-
Notifications
You must be signed in to change notification settings - Fork 82
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
MDN implementation #20
Conversation
commit 735b2c3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 23:01:54 2020 +0900 Change acoustic model from RMDN to Conv1dResnet commit 171d06d Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 22:42:29 2020 +0900 Change num_gaussians of timelag and duration model from 30 to 1000 commit d034811 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 12:03:32 2020 +0900 Minor fix. commit 48f3bf7 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 11:54:47 2020 +0900 Minor fix commit 9e07c70 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 11:47:58 2020 +0900 Change batch_size from 2 to 1. commit 6014169 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 11:35:29 2020 +0900 Change num_gaussians from default(30) to 10 commit 62dda2f Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 11:34:45 2020 +0900 Use torch.sum instead of torch.prod to avoid loss goes to nan. commit bef2d3f Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 01:59:41 2020 +0900 Change duration model from RMDN to LSTM commit edd2397 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 01:34:14 2020 +0900 Change acoustic model to RMDN commit 647dca3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 01:33:53 2020 +0900 Correct comments. commit 0a87efe Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 01:10:43 2020 +0900 Minor fix commit ac88d1d Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 00:59:41 2020 +0900 Add mdn support. commit b08327c Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 00:59:29 2020 +0900 Delete debug print. commit 33d810b Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 00:58:44 2020 +0900 Discard values of torch.return_types.max commit 07f5d02 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 00:06:27 2020 +0900 Add 03-mdn-world recipe. commit bc60f9c Author: Tarou Shirani <taroushirani@gmail.com> Date: Sat Sep 26 00:01:27 2020 +0900 Adapt to the data structure of NNSVS. commit 07f122a Merge: 69b5f71 7d9e184 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Sep 25 21:32:28 2020 +0900 Merge branch 'master' into mdn_dev commit 69b5f71 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Sep 25 21:30:02 2020 +0900 Refactoring. commit 2ad7bbf Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Sep 25 20:37:00 2020 +0900 Add MDN implementation.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for your great contribution! I added a few comments on numerical stability and efficiency. Besides, I have one suggestion to make it work better: use a small number of Gaussians! 500 sounds extremely large to me. In my experience, 10 would be good for a general choice.
I looked into the Hono's paper again and they used 1 mixture component. This is reasonable to me for duration modeling since the output is 1 dimension and the task is not very complicated. We might want to increase the number of Gaussians for acoustic modeling though.
For reference, https://static.googleusercontent.com/media/research.google.com/ja//pubs/archive/42020.pdf, Heiga uses 1-to-16 mixtures for DNN-based TTS task. According to the paper, 2 mixture was good enough (8 mixture was the best).
So, could you try adjusting the number of Gaussians? 1, 2, 4, 8, 16 would be good to go. You might feel 1 is not very effective, but sometimes it works comparable to 10 mixtures and training becomes very stable. https://arxiv.org/abs/1807.07281
nnsvs/mdn.py
Outdated
# p(y|x,w) = exp(log p(y|x,w)) | ||
loss = torch.exp(g.log_prob(target)) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks like this is numerically unstable. Could you consider using logsumexp
(https://pytorch.org/docs/stable/generated/torch.logsumexp.html)? You can refer my implementation https://github.com/r9y9/wavenet_vocoder/blob/63d3fe0a85a69aa25d80700974fe282c96fb6b65/wavenet_vocoder/mixture.py#L202-L218 as an example.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you for your comment. I changed mdn_loss according to your advice.
nnsvs/mdn.py
Outdated
mode = torch.zeros(batch_size, max_T, out_dim) | ||
for i in range(batch_size): | ||
for j in range(max_T): | ||
for k in range(out_dim): | ||
mode[i, j, k] = mu[i, j, max_component[i, j], k] | ||
return mode |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks like this is not efficient... Can we do better at this?
https://github.com/r9y9/wavenet_vocoder/blob/63d3fe0a85a69aa25d80700974fe282c96fb6b65/wavenet_vocoder/mixture.py#L221-L270 I implemented sampling from Gaussian mixtures a while ago. It might be useful as a reference.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you for your comment. Do you mean I should do this without for loop, or I should do this in more probabilistic manner? I borrow to_one_hot from your wavenet_vocoder/wavenet_vocoer/mixture.py and re-implement this without for loop. And I add mdn_sample to sample predicted variables in probabilistic manner.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I meant it would be great removing the for loop
for efficiency. Taking a mean is okay.
Thank you for your comment. I changed the code according to your advice and add some test code (and sorry for forgetting squash merge). I'll try smaller number of Gaussians later. |
commit 8e9de33 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:27:04 2020 +0900 Minor fix commit e86a197 Merge: 6c9d4e9 605a095 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sun Oct 18 21:07:30 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 6c9d4e9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Oct 13 18:18:43 2020 +0900 Change to use most probable mu as conditional mode. commit 2b9649c Merge: fa901d8 9aeaef1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:55:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit fa901d8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:27 2020 +0900 Minor fix commit 4869f47 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:05 2020 +0900 Remove debug print commit 83cf6de Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:42:34 2020 +0900 Delete debug print. commit 6a61a6b Merge: 5b02542 31b7bf8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:50 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 5b02542 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:03 2020 +0900 Fix typo commit 31b7bf8 Merge: 8fc1e4c 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:05:54 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 8fc1e4c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:04:29 2020 +0900 Change num_gaussians commit 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:02:54 2020 +0900 Change RMDN to MDN commit 2b5e03d Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:00:53 2020 +0900 Fix typo commit 68aaba5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 22:01:58 2020 +0900 Fix scaling of sigma commit 3925221 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 17:21:22 2020 +0900 Fix to handle predictions without dyanmic_features commit 2cfc81f Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:50:20 2020 +0900 Change to call multi_stream_mlpg from predict_* commit fbd40d3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:49:56 2020 +0900 Remove mdn_sample_mode and add mdn_get_most_probable_sigma_and_mu. commit 4560322 Merge: e4a66e5 4305844 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 17:58:33 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit e4a66e5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 16:50:40 2020 +0900 Minor fix. commit e489c82 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:48:21 2020 +0900 Change num_gaussians from 4 to 2 commit c3aa7e8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:35:36 2020 +0900 Minor fix. commit 502cae1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:27:48 2020 +0900 Correct comments. commit 8e52754 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:15:00 2020 +0900 Fix arithmetical bug of mdn_loss commit e4c299a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 14:54:17 2020 +0900 Change the caluculation order. commit 291e808 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:49:24 2020 +0900 Change MDN to RMDN commit 4b49d9b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:45:58 2020 +0900 Add ReLU to RMDN commit 8699803 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:39:26 2020 +0900 Change sigma to log_sigma commit 55feef5 Merge: 68248ac 3a55678 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 02:41:03 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 68248ac Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 01:56:55 2020 +0900 Add note. commit 1c4276b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:53:17 2020 +0900 Change batch_size, nepochs, and models. commit 587e4e4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:51:33 2020 +0900 Minor fix commit 733b35f Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:49:37 2020 +0900 Change to make use of mask. commit 44b4158 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:37:57 2020 +0900 Remove Conv1dResnetMDN commit d273dba Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 23:22:14 2020 +0900 Change MDN to RMDN commit 949cefc Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 22:31:58 2020 +0900 Change RMDN model. commit 09aec48 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:54:32 2020 +0900 Minor fix commit 131d978 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:45:10 2020 +0900 Add Conv1dResnetMDN commit dea9c93 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:53:57 2020 +0900 Remove nn.Softmax for numerical stability. commit 5e18115 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:50:58 2020 +0900 Change epoch to 300 commit c8cdc70 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:48:24 2020 +0900 Change to use mdn_sample instead of mdn_sample_mode commit 1aae4fd Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:01:58 2020 +0900 Change to use nn.ReLU() istead of nn.Tanh() commit cc5ea5b Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 01:03:10 2020 +0900 Fix comment and change not to call forward directly.
I fixed the bug of nnsvs/bin/synthesis.py which fails to synthesis when label_path and out_wav_path are specified instead of utt_list etc. And I don't mind waiting. I'm in no hurry. |
I have some questions about the implementation of mixture density network. (1) I changed pi of MDNLayer into the not softmax-ed version for numerical stability and currently it has no mathematical meaning in itself(unless it is softmax-ed by hand). Is it better to change pi into logpi(logged "pi") by adding F.log_softmax to MDNLayer.forward() ? (2) I found that mdn_loss sometimes goes to nan when num_gaussians is large. In my knowledge, MDN won't work if one of its variance goes to zero and its Gaussian component shrink to a point. Your implementation[1] has the lower limit of log_scale(logged variance) and I guess this is to avoid the collapse of MDN. Is this correct? (3) In your implementation you use logged version of scale. Is this for numerical stability? |
(1) Yes, I think we should apply F.log_softmax to have mixture weights to be consistent with the mathematical formulation. (3) Yes. Otherwise, the scale would quickly go to zero, especially when the number of Gaussians is large. (2): The log scale clipping is for stabilizing the training. I recommend using both 1) log-scale parameterization and 2) clipping to improve numerical stability. |
commit 514087a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 01:43:37 2020 +0900 Change to use log_pi and log_sigma instead of pi and sigma. commit 825a75b Merge: 8e9de33 a94ebc1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:35:02 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 8e9de33 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:27:04 2020 +0900 Minor fix commit e86a197 Merge: 6c9d4e9 605a095 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sun Oct 18 21:07:30 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 6c9d4e9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Oct 13 18:18:43 2020 +0900 Change to use most probable mu as conditional mode. commit 2b9649c Merge: fa901d8 9aeaef1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:55:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit fa901d8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:27 2020 +0900 Minor fix commit 4869f47 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:05 2020 +0900 Remove debug print commit 83cf6de Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:42:34 2020 +0900 Delete debug print. commit 6a61a6b Merge: 5b02542 31b7bf8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:50 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 5b02542 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:03 2020 +0900 Fix typo commit 31b7bf8 Merge: 8fc1e4c 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:05:54 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 8fc1e4c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:04:29 2020 +0900 Change num_gaussians commit 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:02:54 2020 +0900 Change RMDN to MDN commit 2b5e03d Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:00:53 2020 +0900 Fix typo commit 68aaba5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 22:01:58 2020 +0900 Fix scaling of sigma commit 3925221 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 17:21:22 2020 +0900 Fix to handle predictions without dyanmic_features commit 2cfc81f Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:50:20 2020 +0900 Change to call multi_stream_mlpg from predict_* commit fbd40d3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:49:56 2020 +0900 Remove mdn_sample_mode and add mdn_get_most_probable_sigma_and_mu. commit 4560322 Merge: e4a66e5 4305844 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 17:58:33 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit e4a66e5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 16:50:40 2020 +0900 Minor fix. commit e489c82 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:48:21 2020 +0900 Change num_gaussians from 4 to 2 commit c3aa7e8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:35:36 2020 +0900 Minor fix. commit 502cae1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:27:48 2020 +0900 Correct comments. commit 8e52754 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:15:00 2020 +0900 Fix arithmetical bug of mdn_loss commit e4c299a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 14:54:17 2020 +0900 Change the caluculation order. commit 291e808 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:49:24 2020 +0900 Change MDN to RMDN commit 4b49d9b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:45:58 2020 +0900 Add ReLU to RMDN commit 8699803 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:39:26 2020 +0900 Change sigma to log_sigma commit 55feef5 Merge: 68248ac 3a55678 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 02:41:03 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 68248ac Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 01:56:55 2020 +0900 Add note. commit 1c4276b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:53:17 2020 +0900 Change batch_size, nepochs, and models. commit 587e4e4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:51:33 2020 +0900 Minor fix commit 733b35f Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:49:37 2020 +0900 Change to make use of mask. commit 44b4158 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:37:57 2020 +0900 Remove Conv1dResnetMDN commit d273dba Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 23:22:14 2020 +0900 Change MDN to RMDN commit 949cefc Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 22:31:58 2020 +0900 Change RMDN model. commit 09aec48 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:54:32 2020 +0900 Minor fix commit 131d978 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:45:10 2020 +0900 Add Conv1dResnetMDN commit dea9c93 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:53:57 2020 +0900 Remove nn.Softmax for numerical stability. commit 5e18115 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:50:58 2020 +0900 Change epoch to 300 commit c8cdc70 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:48:24 2020 +0900 Change to use mdn_sample instead of mdn_sample_mode commit 1aae4fd Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:01:58 2020 +0900 Change to use nn.ReLU() istead of nn.Tanh() commit cc5ea5b Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 01:03:10 2020 +0900 Fix comment and change not to call forward directly.
Thank you for your rapid comment. I changed to use log_pi and log_sigma instead of pi and sigma as you recommended. And I have another question. I'm afraid that torch.distributions.Normal.log_prob(x) may return -inf when the possibility of x is zero. Is there any need to clamp the result of log_prob with the same way? |
torch.distributions.Normal.log_prob(x) may return -inf even if the possibility is very small but not zero, due to the precision of torch.tensor(pytorch/pytorch#12269) |
I think it's a common issue. Adding a clamp should fix the numerical stability. |
commit 86513fb Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:11:21 2020 +0900 Fix comment commit 56e96c2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:05:21 2020 +0900 Remove comment commit 3d2c046 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:04:43 2020 +0900 Remove lower limit of log_prob commit bc1eef0 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 23:55:30 2020 +0900 Fix variables commit 2d9c35c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 22:45:03 2020 +0900 Add comments commit 37156eb Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 21:10:16 2020 +0900 Add lower limit to torch.distributions.Normal.log_prob commit 17c9ac8 Merge: 514087a 1251bb2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 02:31:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 514087a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 01:43:37 2020 +0900 Change to use log_pi and log_sigma instead of pi and sigma. commit 825a75b Merge: 8e9de33 a94ebc1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:35:02 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 8e9de33 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:27:04 2020 +0900 Minor fix commit e86a197 Merge: 6c9d4e9 605a095 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sun Oct 18 21:07:30 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 6c9d4e9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Oct 13 18:18:43 2020 +0900 Change to use most probable mu as conditional mode. commit 2b9649c Merge: fa901d8 9aeaef1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:55:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit fa901d8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:27 2020 +0900 Minor fix commit 4869f47 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:05 2020 +0900 Remove debug print commit 83cf6de Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:42:34 2020 +0900 Delete debug print. commit 6a61a6b Merge: 5b02542 31b7bf8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:50 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 5b02542 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:03 2020 +0900 Fix typo commit 31b7bf8 Merge: 8fc1e4c 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:05:54 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 8fc1e4c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:04:29 2020 +0900 Change num_gaussians commit 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:02:54 2020 +0900 Change RMDN to MDN commit 2b5e03d Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:00:53 2020 +0900 Fix typo commit 68aaba5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 22:01:58 2020 +0900 Fix scaling of sigma commit 3925221 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 17:21:22 2020 +0900 Fix to handle predictions without dyanmic_features commit 2cfc81f Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:50:20 2020 +0900 Change to call multi_stream_mlpg from predict_* commit fbd40d3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:49:56 2020 +0900 Remove mdn_sample_mode and add mdn_get_most_probable_sigma_and_mu. commit 4560322 Merge: e4a66e5 4305844 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 17:58:33 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit e4a66e5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 16:50:40 2020 +0900 Minor fix. commit e489c82 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:48:21 2020 +0900 Change num_gaussians from 4 to 2 commit c3aa7e8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:35:36 2020 +0900 Minor fix. commit 502cae1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:27:48 2020 +0900 Correct comments. commit 8e52754 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:15:00 2020 +0900 Fix arithmetical bug of mdn_loss commit e4c299a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 14:54:17 2020 +0900 Change the caluculation order. commit 291e808 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:49:24 2020 +0900 Change MDN to RMDN commit 4b49d9b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:45:58 2020 +0900 Add ReLU to RMDN commit 8699803 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:39:26 2020 +0900 Change sigma to log_sigma commit 55feef5 Merge: 68248ac 3a55678 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 02:41:03 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 68248ac Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 01:56:55 2020 +0900 Add note. commit 1c4276b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:53:17 2020 +0900 Change batch_size, nepochs, and models. commit 587e4e4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:51:33 2020 +0900 Minor fix commit 733b35f Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:49:37 2020 +0900 Change to make use of mask. commit 44b4158 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:37:57 2020 +0900 Remove Conv1dResnetMDN commit d273dba Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 23:22:14 2020 +0900 Change MDN to RMDN commit 949cefc Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 22:31:58 2020 +0900 Change RMDN model. commit 09aec48 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:54:32 2020 +0900 Minor fix commit 131d978 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:45:10 2020 +0900 Add Conv1dResnetMDN commit dea9c93 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:53:57 2020 +0900 Remove nn.Softmax for numerical stability. commit 5e18115 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:50:58 2020 +0900 Change epoch to 300 commit c8cdc70 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:48:24 2020 +0900 Change to use mdn_sample instead of mdn_sample_mode commit 1aae4fd Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:01:58 2020 +0900 Change to use nn.ReLU() istead of nn.Tanh() commit cc5ea5b Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 01:03:10 2020 +0900 Fix comment and change not to call forward directly.
Thank you for your comment. I tried to clamp the result of torch.distributions.Normal.log_prob(x) but the numerical instability worsened. I have no idea why this occurred and my codes seems to be numerically unstable yet. |
commit f029a95 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:51:53 2020 +0900 Minor fix commit 0c8eab8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:50:18 2020 +0900 minor fix commit 3c053bf Merge: 86513fb b2c426f Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:29:27 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 86513fb Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:11:21 2020 +0900 Fix comment commit 56e96c2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:05:21 2020 +0900 Remove comment commit 3d2c046 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:04:43 2020 +0900 Remove lower limit of log_prob commit bc1eef0 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 23:55:30 2020 +0900 Fix variables commit 2d9c35c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 22:45:03 2020 +0900 Add comments commit 37156eb Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 21:10:16 2020 +0900 Add lower limit to torch.distributions.Normal.log_prob commit 17c9ac8 Merge: 514087a 1251bb2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 02:31:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 514087a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 01:43:37 2020 +0900 Change to use log_pi and log_sigma instead of pi and sigma. commit 825a75b Merge: 8e9de33 a94ebc1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:35:02 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 8e9de33 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:27:04 2020 +0900 Minor fix commit e86a197 Merge: 6c9d4e9 605a095 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sun Oct 18 21:07:30 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 6c9d4e9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Oct 13 18:18:43 2020 +0900 Change to use most probable mu as conditional mode. commit 2b9649c Merge: fa901d8 9aeaef1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:55:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit fa901d8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:27 2020 +0900 Minor fix commit 4869f47 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:05 2020 +0900 Remove debug print commit 83cf6de Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:42:34 2020 +0900 Delete debug print. commit 6a61a6b Merge: 5b02542 31b7bf8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:50 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 5b02542 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:03 2020 +0900 Fix typo commit 31b7bf8 Merge: 8fc1e4c 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:05:54 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 8fc1e4c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:04:29 2020 +0900 Change num_gaussians commit 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:02:54 2020 +0900 Change RMDN to MDN commit 2b5e03d Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:00:53 2020 +0900 Fix typo commit 68aaba5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 22:01:58 2020 +0900 Fix scaling of sigma commit 3925221 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 17:21:22 2020 +0900 Fix to handle predictions without dyanmic_features commit 2cfc81f Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:50:20 2020 +0900 Change to call multi_stream_mlpg from predict_* commit fbd40d3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:49:56 2020 +0900 Remove mdn_sample_mode and add mdn_get_most_probable_sigma_and_mu. commit 4560322 Merge: e4a66e5 4305844 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 17:58:33 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit e4a66e5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 16:50:40 2020 +0900 Minor fix. commit e489c82 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:48:21 2020 +0900 Change num_gaussians from 4 to 2 commit c3aa7e8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:35:36 2020 +0900 Minor fix. commit 502cae1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:27:48 2020 +0900 Correct comments. commit 8e52754 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:15:00 2020 +0900 Fix arithmetical bug of mdn_loss commit e4c299a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 14:54:17 2020 +0900 Change the caluculation order. commit 291e808 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:49:24 2020 +0900 Change MDN to RMDN commit 4b49d9b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:45:58 2020 +0900 Add ReLU to RMDN commit 8699803 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:39:26 2020 +0900 Change sigma to log_sigma commit 55feef5 Merge: 68248ac 3a55678 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 02:41:03 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 68248ac Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 01:56:55 2020 +0900 Add note. commit 1c4276b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:53:17 2020 +0900 Change batch_size, nepochs, and models. commit 587e4e4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:51:33 2020 +0900 Minor fix commit 733b35f Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:49:37 2020 +0900 Change to make use of mask. commit 44b4158 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:37:57 2020 +0900 Remove Conv1dResnetMDN commit d273dba Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 23:22:14 2020 +0900 Change MDN to RMDN commit 949cefc Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 22:31:58 2020 +0900 Change RMDN model. commit 09aec48 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:54:32 2020 +0900 Minor fix commit 131d978 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:45:10 2020 +0900 Add Conv1dResnetMDN commit dea9c93 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:53:57 2020 +0900 Remove nn.Softmax for numerical stability. commit 5e18115 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:50:58 2020 +0900 Change epoch to 300 commit c8cdc70 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:48:24 2020 +0900 Change to use mdn_sample instead of mdn_sample_mode commit 1aae4fd Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:01:58 2020 +0900 Change to use nn.ReLU() istead of nn.Tanh() commit cc5ea5b Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 01:03:10 2020 +0900 Fix comment and change not to call forward directly.
Umm. I will also look into the issue this weekend! |
I added torch.autograd.set_detect_anomaly(True) to nnsvs/bin/train.py and I found that the back propagation of torch.logsumpexp(=LogsumexpBackward) returned nan. In my knowledge, weight coefficient pi must be between 0 and 1 due to the nature of the mixture density network, and this can apply to torch.distributions.Normal. -inf < log_pi < 0 So the summation of them must be between -inf and 0. If x can be written as x = (x_1, x_2,...,x_n) the partial derivative of logsumexp(x) (d logsumexp(x) /dx_i) is 1/(sum exp(x)) * exp(x_i). This may be solved by setting lower limits to all logged values. (I thought F.log_softmax would be implemented with consideration of such a situation like this, but merely clamping log_pi solved my problem...) |
commit 63a6f87 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 19:15:21 2020 +0900 Add clamping log_pi and and torch.distributions.Normal.log_prob. commit 09a1f11 Merge: f029a95 52168d2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:56:44 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit f029a95 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:51:53 2020 +0900 Minor fix commit 0c8eab8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:50:18 2020 +0900 minor fix commit 3c053bf Merge: 86513fb b2c426f Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:29:27 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 86513fb Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:11:21 2020 +0900 Fix comment commit 56e96c2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:05:21 2020 +0900 Remove comment commit 3d2c046 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:04:43 2020 +0900 Remove lower limit of log_prob commit bc1eef0 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 23:55:30 2020 +0900 Fix variables commit 2d9c35c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 22:45:03 2020 +0900 Add comments commit 37156eb Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 21:10:16 2020 +0900 Add lower limit to torch.distributions.Normal.log_prob commit 17c9ac8 Merge: 514087a 1251bb2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 02:31:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 514087a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 01:43:37 2020 +0900 Change to use log_pi and log_sigma instead of pi and sigma. commit 825a75b Merge: 8e9de33 a94ebc1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:35:02 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 8e9de33 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:27:04 2020 +0900 Minor fix commit e86a197 Merge: 6c9d4e9 605a095 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sun Oct 18 21:07:30 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 6c9d4e9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Oct 13 18:18:43 2020 +0900 Change to use most probable mu as conditional mode. commit 2b9649c Merge: fa901d8 9aeaef1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:55:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit fa901d8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:27 2020 +0900 Minor fix commit 4869f47 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:05 2020 +0900 Remove debug print commit 83cf6de Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:42:34 2020 +0900 Delete debug print. commit 6a61a6b Merge: 5b02542 31b7bf8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:50 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 5b02542 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:03 2020 +0900 Fix typo commit 31b7bf8 Merge: 8fc1e4c 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:05:54 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 8fc1e4c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:04:29 2020 +0900 Change num_gaussians commit 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:02:54 2020 +0900 Change RMDN to MDN commit 2b5e03d Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:00:53 2020 +0900 Fix typo commit 68aaba5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 22:01:58 2020 +0900 Fix scaling of sigma commit 3925221 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 17:21:22 2020 +0900 Fix to handle predictions without dyanmic_features commit 2cfc81f Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:50:20 2020 +0900 Change to call multi_stream_mlpg from predict_* commit fbd40d3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:49:56 2020 +0900 Remove mdn_sample_mode and add mdn_get_most_probable_sigma_and_mu. commit 4560322 Merge: e4a66e5 4305844 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 17:58:33 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit e4a66e5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 16:50:40 2020 +0900 Minor fix. commit e489c82 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:48:21 2020 +0900 Change num_gaussians from 4 to 2 commit c3aa7e8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:35:36 2020 +0900 Minor fix. commit 502cae1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:27:48 2020 +0900 Correct comments. commit 8e52754 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:15:00 2020 +0900 Fix arithmetical bug of mdn_loss commit e4c299a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 14:54:17 2020 +0900 Change the caluculation order. commit 291e808 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:49:24 2020 +0900 Change MDN to RMDN commit 4b49d9b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:45:58 2020 +0900 Add ReLU to RMDN commit 8699803 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:39:26 2020 +0900 Change sigma to log_sigma commit 55feef5 Merge: 68248ac 3a55678 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 02:41:03 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 68248ac Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 01:56:55 2020 +0900 Add note. commit 1c4276b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:53:17 2020 +0900 Change batch_size, nepochs, and models. commit 587e4e4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:51:33 2020 +0900 Minor fix commit 733b35f Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:49:37 2020 +0900 Change to make use of mask. commit 44b4158 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:37:57 2020 +0900 Remove Conv1dResnetMDN commit d273dba Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 23:22:14 2020 +0900 Change MDN to RMDN commit 949cefc Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 22:31:58 2020 +0900 Change RMDN model. commit 09aec48 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:54:32 2020 +0900 Minor fix commit 131d978 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:45:10 2020 +0900 Add Conv1dResnetMDN commit dea9c93 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:53:57 2020 +0900 Remove nn.Softmax for numerical stability. commit 5e18115 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:50:58 2020 +0900 Change epoch to 300 commit c8cdc70 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:48:24 2020 +0900 Change to use mdn_sample instead of mdn_sample_mode commit 1aae4fd Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:01:58 2020 +0900 Change to use nn.ReLU() istead of nn.Tanh() commit cc5ea5b Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 01:03:10 2020 +0900 Fix comment and change not to call forward directly.
Hi, I am going to look into this very soon, hopefully tonight or tomorrow. I will first fix #14 and refactor recipes to reduce dup across recipes. Then, I will review the PR and suggest some changes to make it into good shape. Thanks for your patience! |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
First of all, thank you for your great effort! Overall it looks very good. I tried the PR locally and it worked also well. Please see the individual comments to make the code better.
For numerical instability, I cannot reproduce the issues locally. I tried to uncomment some of the clamp operations, but it just worked okay. If you can provide me detailed steps to reproduce, I can look into it more carefully.
For the commit history, it would be very helpful if you cleaned it up (maybe using git rebase?), so that I can review it easily. That said, I understand using git is a bit difficult (even for me!). I am okay if you happen to mess up the commit history. In that case, I can run git rebase when I merge the PR if necessary.
Many thanks!! I am looking forward to getting the MDN feature into the master.
@@ -0,0 +1,241 @@ | |||
#!/bin/bash |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I deleted the duplication in the latest commit.
egs/nit-song070/03-mdn-world/run.sh
Outdated
model.netG._target_=nnsvs.model.MDN \ | ||
model.netG.hidden_dim=1024 \ | ||
model.netG.num_layers=4 \ | ||
+model.netG.num_gaussians=$num_gaussians \ |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
With #26, these configs can be implemented by an external config. See https://github.com/r9y9/nnsvs/tree/caefa1081bbd8d62ab9eb945033bd3ea02814d18/egs/nit-song070/00-svs-world/conf/train/model for example.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I deleted the duplication in the latest commit.
egs/nit-song070/03-mdn-world/run.sh
Outdated
model.netG._target_=nnsvs.model.MDN \ | ||
model.netG.hidden_dim=1024 \ | ||
model.netG.num_layers=4 \ | ||
+model.netG.num_gaussians=$num_gaussians \ | ||
~model.netG.bidirectional=True \ |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same as above. Also, tabs and whitespaces are mixed. Please use whitespace for consistency.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I deleted the duplication in the latest commit.
egs/nit-song070/03-mdn-world/run.sh
Outdated
model.netG._target_=nnsvs.model.MDN \ | ||
model.netG.hidden_dim=1024 \ | ||
model.netG.num_layers=4 \ | ||
+model.netG.num_gaussians=$num_gaussians \ |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same as above.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I deleted the duplication in the latest commit.
@@ -0,0 +1,215 @@ | |||
# coding: utf-8 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This file is a dup. Please use a symlink instead, or extend the existing 00-svs-world recipe with external config files.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I changed to use symlink in the latest commit.
tests/test_mdn.py
Outdated
sys.path.append(os.path.join(os.path.dirname(__file__), '..', 'nnsvs')) | ||
import mdn |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think the first line is unnecessary. Please consider replacing the second line with from nnsvs import mdn
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I changed import directive as you recommended in the latest commit.
nnsvs/bin/generate.py
Outdated
out = multi_stream_mlpg( | ||
out, scaler.var_, windows, model_config.stream_sizes, | ||
model_config.has_dynamic_features) | ||
if model.prediction_type == "probabilistic": |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Please consider changing this like:
if not model.is_deterministic()
or if model.prediction_type() == ENUM_VALUE
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I changed to use predictin_type() of BaseModel in the latest commit.
nnsvs/gen.py
Outdated
# De-normalization and rounding | ||
lag = np.round(timelag_out_scaler.inverse_transform(y.data.numpy())) | ||
# Run model | ||
if timelag_model.prediction_type == "probabilistic": |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same as above.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I changed to use predictin_type() of BaseModel in the latest commit.
nnsvs/gen.py
Outdated
|
||
# Apply denormalization | ||
pred_durations = duration_out_scaler.inverse_transform(pred_durations) | ||
if duration_model.prediction_type == "probabilistic": |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same as above.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I changed to use predictin_type() of BaseModel in the latest commit.
nnsvs/gen.py
Outdated
|
||
# Apply denormalization | ||
pred_acoustic = acoustic_out_scaler.inverse_transform(pred_acoustic) | ||
if acoustic_model.prediction_type == "probabilistic": |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same as above.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I changed to use predictin_type() of BaseModel in the latest commit.
commit b53bea4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:12:26 2020 +0900 Delete debug print commit 54b80a9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:06:43 2020 +0900 Change to call prediction_type correctly. commit 4736dbe Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:56:38 2020 +0900 Minor fix commit a953daf Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:32:11 2020 +0900 Delete wrong README.md commit 68cad87 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:30:50 2020 +0900 Fix wrong symlink commit 8c42416 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:26:50 2020 +0900 Split clamping values commit e5e5f43 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:21:46 2020 +0900 Follow google's style of docstring commit c98b9cb Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:21:31 2020 +0900 Add user defined conf of hydra commit 181412d Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:20:44 2020 +0900 Revet deletion of unnecessary values commit 7e6a175 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:19:49 2020 +0900 Use ENUM type instead of str to specify the prediction type commit e491de8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:19:00 2020 +0900 Use data_prep.py of 00-svs-world. commit a743943 Merge: 5dfedd1 25c9f01 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 21:41:34 2020 +0900 Merge branch 'master' into mdn_dev commit 25c9f01 Merge: 4d90dbb ffa9fa6 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Mon Nov 2 16:33:43 2020 +0900 Merge pull request nnsvs#28 from r9y9/fix-pitch-diff-features hed: Fix JP pitch diff features commit ffa9fa6 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Mon Nov 2 16:14:28 2020 +0900 hed: Fix JP pitch diff features ref r9y9/nnmnkwii#109 commit 4d90dbb Merge: c3dd805 ce87053 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Mon Nov 2 00:48:58 2020 +0900 Merge pull request nnsvs#27 from r9y9/basemodel Add BaseModel interface commit ce87053 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Mon Nov 2 00:01:15 2020 +0900 Add BaseModel interface and add inference method in case we want to run different operations for training and inferente time (e.g., shallow AR) commit 5dfedd1 Merge: 63a6f87 c3dd805 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 00:00:32 2020 +0900 Merge branch 'master' into mdn_dev commit c3dd805 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sun Nov 1 01:25:54 2020 +0900 Add python docstring style in README commit caefa10 Merge: 91a1788 867e588 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 21:55:19 2020 +0900 Merge pull request nnsvs#26 from r9y9/hydra-config-dir Recipes: allow users to have custom model configs in a recipe level commit 867e588 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 21:45:05 2020 +0900 Fix README commit 065c5ac Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 21:43:31 2020 +0900 recipes: add an example on how to use custom model configs in a recipe commit a00734a Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 21:35:28 2020 +0900 recipes: allow users to control models by specifying thier configs without manually editing nnsvs/bin/conf/train/model/*.yaml finally fixes nnsvs#14 Thanks the hydra authors! commit 92d3eaa Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 21:32:26 2020 +0900 train: rename config names commit db85a6e Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 21:22:54 2020 +0900 hydra: add --config-dir for train_*.sh commit c402230 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 21:14:46 2020 +0900 rm old dirs commit 91a1788 Merge: 9143538 50ec407 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 21:11:39 2020 +0900 Merge pull request nnsvs#25 from r9y9/recipes-remove-dup Recipe refactors commit 50ec407 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 20:59:47 2020 +0900 Recipe refactors - made common scripts and removed the most of duplicated scripts - put the common hed file to egs/_common/hed - renamed utils to local following kaldi commit 9143538 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 19:32:08 2020 +0900 Update gitignore commit 63f44d3 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 19:30:36 2020 +0900 Fix traling spaces commit 28e2f9f Merge: 7d9e184 6aeedf7 Author: Ryuichi Yamamoto <zryuichi@gmail.com> Date: Sat Oct 31 19:16:15 2020 +0900 Merge pull request nnsvs#23 from taroushirani/detect_anomaly_pr use_detect_anomaly feature commit 6aeedf7 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 20:18:24 2020 +0900 Squashed commit of the following: commit 7bcd463 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 20:14:35 2020 +0900 Add use_detect_anomaly feature commit 63a6f87 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 19:15:21 2020 +0900 Add clamping log_pi and and torch.distributions.Normal.log_prob. commit 09a1f11 Merge: f029a95 52168d2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:56:44 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit f029a95 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:51:53 2020 +0900 Minor fix commit 0c8eab8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:50:18 2020 +0900 minor fix commit 3c053bf Merge: 86513fb b2c426f Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:29:27 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 86513fb Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:11:21 2020 +0900 Fix comment commit 56e96c2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:05:21 2020 +0900 Remove comment commit 3d2c046 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:04:43 2020 +0900 Remove lower limit of log_prob commit bc1eef0 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 23:55:30 2020 +0900 Fix variables commit 2d9c35c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 22:45:03 2020 +0900 Add comments commit 37156eb Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 21:10:16 2020 +0900 Add lower limit to torch.distributions.Normal.log_prob commit 17c9ac8 Merge: 514087a 1251bb2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 02:31:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 514087a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 01:43:37 2020 +0900 Change to use log_pi and log_sigma instead of pi and sigma. commit 825a75b Merge: 8e9de33 a94ebc1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:35:02 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 8e9de33 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:27:04 2020 +0900 Minor fix commit e86a197 Merge: 6c9d4e9 605a095 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sun Oct 18 21:07:30 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 6c9d4e9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Oct 13 18:18:43 2020 +0900 Change to use most probable mu as conditional mode. commit 2b9649c Merge: fa901d8 9aeaef1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:55:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit fa901d8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:27 2020 +0900 Minor fix commit 4869f47 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:05 2020 +0900 Remove debug print commit 83cf6de Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:42:34 2020 +0900 Delete debug print. commit 6a61a6b Merge: 5b02542 31b7bf8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:50 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 5b02542 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:03 2020 +0900 Fix typo commit 31b7bf8 Merge: 8fc1e4c 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:05:54 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 8fc1e4c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:04:29 2020 +0900 Change num_gaussians commit 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:02:54 2020 +0900 Change RMDN to MDN commit 2b5e03d Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:00:53 2020 +0900 Fix typo commit 68aaba5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 22:01:58 2020 +0900 Fix scaling of sigma commit 3925221 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 17:21:22 2020 +0900 Fix to handle predictions without dyanmic_features commit 2cfc81f Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:50:20 2020 +0900 Change to call multi_stream_mlpg from predict_* commit fbd40d3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:49:56 2020 +0900 Remove mdn_sample_mode and add mdn_get_most_probable_sigma_and_mu. commit 4560322 Merge: e4a66e5 4305844 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 17:58:33 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit e4a66e5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 16:50:40 2020 +0900 Minor fix. commit e489c82 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:48:21 2020 +0900 Change num_gaussians from 4 to 2 commit c3aa7e8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:35:36 2020 +0900 Minor fix. commit 502cae1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:27:48 2020 +0900 Correct comments. commit 8e52754 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:15:00 2020 +0900 Fix arithmetical bug of mdn_loss commit e4c299a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 14:54:17 2020 +0900 Change the caluculation order. commit 291e808 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:49:24 2020 +0900 Change MDN to RMDN commit 4b49d9b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:45:58 2020 +0900 Add ReLU to RMDN commit 8699803 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:39:26 2020 +0900 Change sigma to log_sigma commit 55feef5 Merge: 68248ac 3a55678 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 02:41:03 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 68248ac Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 01:56:55 2020 +0900 Add note. commit 1c4276b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:53:17 2020 +0900 Change batch_size, nepochs, and models. commit 587e4e4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:51:33 2020 +0900 Minor fix commit 733b35f Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:49:37 2020 +0900 Change to make use of mask. commit 44b4158 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:37:57 2020 +0900 Remove Conv1dResnetMDN commit d273dba Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 23:22:14 2020 +0900 Change MDN to RMDN commit 949cefc Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 22:31:58 2020 +0900 Change RMDN model. commit 09aec48 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:54:32 2020 +0900 Minor fix commit 131d978 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:45:10 2020 +0900 Add Conv1dResnetMDN commit dea9c93 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:53:57 2020 +0900 Remove nn.Softmax for numerical stability. commit 5e18115 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:50:58 2020 +0900 Change epoch to 300 commit c8cdc70 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:48:24 2020 +0900 Change to use mdn_sample instead of mdn_sample_mode commit 1aae4fd Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:01:58 2020 +0900 Change to use nn.ReLU() istead of nn.Tanh() commit cc5ea5b Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 01:03:10 2020 +0900 Fix comment and change not to call forward directly.
commit ca446a9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:53:27 2020 +0900 Fix import directive commit 14370a5 Merge: b53bea4 11088a1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:51:35 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit b53bea4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:12:26 2020 +0900 Delete debug print commit 54b80a9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:06:43 2020 +0900 Change to call prediction_type correctly. commit 4736dbe Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:56:38 2020 +0900 Minor fix commit a953daf Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:32:11 2020 +0900 Delete wrong README.md commit 68cad87 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:30:50 2020 +0900 Fix wrong symlink commit 8c42416 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:26:50 2020 +0900 Split clamping values commit e5e5f43 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:21:46 2020 +0900 Follow google's style of docstring commit c98b9cb Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:21:31 2020 +0900 Add user defined conf of hydra commit 181412d Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:20:44 2020 +0900 Revet deletion of unnecessary values commit 7e6a175 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:19:49 2020 +0900 Use ENUM type instead of str to specify the prediction type commit e491de8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:19:00 2020 +0900 Use data_prep.py of 00-svs-world. commit a743943 Merge: 5dfedd1 25c9f01 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 21:41:34 2020 +0900 Merge branch 'master' into mdn_dev commit 5dfedd1 Merge: 63a6f87 c3dd805 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 00:00:32 2020 +0900 Merge branch 'master' into mdn_dev commit 63a6f87 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 19:15:21 2020 +0900 Add clamping log_pi and and torch.distributions.Normal.log_prob. commit 09a1f11 Merge: f029a95 52168d2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:56:44 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit f029a95 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:51:53 2020 +0900 Minor fix commit 0c8eab8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:50:18 2020 +0900 minor fix commit 3c053bf Merge: 86513fb b2c426f Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:29:27 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 86513fb Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:11:21 2020 +0900 Fix comment commit 56e96c2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:05:21 2020 +0900 Remove comment commit 3d2c046 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:04:43 2020 +0900 Remove lower limit of log_prob commit bc1eef0 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 23:55:30 2020 +0900 Fix variables commit 2d9c35c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 22:45:03 2020 +0900 Add comments commit 37156eb Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 21:10:16 2020 +0900 Add lower limit to torch.distributions.Normal.log_prob commit 17c9ac8 Merge: 514087a 1251bb2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 02:31:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 514087a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 01:43:37 2020 +0900 Change to use log_pi and log_sigma instead of pi and sigma. commit 825a75b Merge: 8e9de33 a94ebc1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:35:02 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 8e9de33 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:27:04 2020 +0900 Minor fix commit e86a197 Merge: 6c9d4e9 605a095 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sun Oct 18 21:07:30 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 6c9d4e9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Oct 13 18:18:43 2020 +0900 Change to use most probable mu as conditional mode. commit 2b9649c Merge: fa901d8 9aeaef1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:55:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit fa901d8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:27 2020 +0900 Minor fix commit 4869f47 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:05 2020 +0900 Remove debug print commit 83cf6de Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:42:34 2020 +0900 Delete debug print. commit 6a61a6b Merge: 5b02542 31b7bf8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:50 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 5b02542 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:03 2020 +0900 Fix typo commit 31b7bf8 Merge: 8fc1e4c 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:05:54 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 8fc1e4c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:04:29 2020 +0900 Change num_gaussians commit 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:02:54 2020 +0900 Change RMDN to MDN commit 2b5e03d Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:00:53 2020 +0900 Fix typo commit 68aaba5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 22:01:58 2020 +0900 Fix scaling of sigma commit 3925221 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 17:21:22 2020 +0900 Fix to handle predictions without dyanmic_features commit 2cfc81f Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:50:20 2020 +0900 Change to call multi_stream_mlpg from predict_* commit fbd40d3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:49:56 2020 +0900 Remove mdn_sample_mode and add mdn_get_most_probable_sigma_and_mu. commit 4560322 Merge: e4a66e5 4305844 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 17:58:33 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit e4a66e5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 16:50:40 2020 +0900 Minor fix. commit e489c82 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:48:21 2020 +0900 Change num_gaussians from 4 to 2 commit c3aa7e8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:35:36 2020 +0900 Minor fix. commit 502cae1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:27:48 2020 +0900 Correct comments. commit 8e52754 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:15:00 2020 +0900 Fix arithmetical bug of mdn_loss commit e4c299a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 14:54:17 2020 +0900 Change the caluculation order. commit 291e808 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:49:24 2020 +0900 Change MDN to RMDN commit 4b49d9b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:45:58 2020 +0900 Add ReLU to RMDN commit 8699803 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:39:26 2020 +0900 Change sigma to log_sigma commit 55feef5 Merge: 68248ac 3a55678 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 02:41:03 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 68248ac Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 01:56:55 2020 +0900 Add note. commit 1c4276b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:53:17 2020 +0900 Change batch_size, nepochs, and models. commit 587e4e4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:51:33 2020 +0900 Minor fix commit 733b35f Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:49:37 2020 +0900 Change to make use of mask. commit 44b4158 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:37:57 2020 +0900 Remove Conv1dResnetMDN commit d273dba Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 23:22:14 2020 +0900 Change MDN to RMDN commit 949cefc Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 22:31:58 2020 +0900 Change RMDN model. commit 09aec48 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:54:32 2020 +0900 Minor fix commit 131d978 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:45:10 2020 +0900 Add Conv1dResnetMDN commit dea9c93 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:53:57 2020 +0900 Remove nn.Softmax for numerical stability. commit 5e18115 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:50:58 2020 +0900 Change epoch to 300 commit c8cdc70 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:48:24 2020 +0900 Change to use mdn_sample instead of mdn_sample_mode commit 1aae4fd Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:01:58 2020 +0900 Change to use nn.ReLU() istead of nn.Tanh() commit cc5ea5b Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 01:03:10 2020 +0900 Fix comment and change not to call forward directly.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The code is getting in a good shape! I left a few minor comments. Once the minor issues are addressed, I am going to merge the PR.
nnsvs/model.py
Outdated
@@ -74,14 +70,52 @@ def __init__(self, in_dim, hidden_dim, out_dim, num_layers=1, bidirectional=True | |||
super(LSTMRNN, self).__init__() | |||
self.hidden_dim = hidden_dim | |||
self.num_layers = num_layers | |||
self.relu = nn.ReLU() |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks like this is not used anymore?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I deleted unnecessary ReLU in the latest commit.
@@ -25,7 +24,6 @@ def __init__(self, dim, dilation=1): | |||
WNConv1d(dim, dim, kernel_size=1), | |||
) | |||
self.shortcut = WNConv1d(dim, dim, kernel_size=1) | |||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
minor: could you keep the new line here?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I added new line there in the latest commit.
@@ -46,7 +44,6 @@ def __init__(self, in_dim, hidden_dim, out_dim, num_layers=4, dropout=0.0): | |||
] | |||
|
|||
self.model = nn.Sequential(*model) | |||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same as above
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I added new line there in the latest commit.
@@ -60,7 +57,6 @@ def __init__(self, in_dim, hidden_dim, out_dim, num_layers=2, dropout=0.0): | |||
self.last_linear = nn.Linear(hidden_dim, out_dim) | |||
self.relu = nn.ReLU() | |||
self.dropout = nn.Dropout(dropout) | |||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same as above
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I added new line there in the latest commit.
self.num_direction = 2 if bidirectional else 1 | ||
self.lstm = nn.LSTM(in_dim, hidden_dim, num_layers, | ||
bidirectional=bidirectional, batch_first=True, dropout=dropout) | ||
self.hidden2out = nn.Linear(self.num_direction*self.hidden_dim, out_dim) | ||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same as above
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I added new line there in the latest commit.
commit 6fe86b4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 01:39:26 2020 +0900 Add spaces and delete unnnecessary ReLU commit cdebfb6 Merge: ca446a9 b707a5f Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 01:00:17 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit ca446a9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:53:27 2020 +0900 Fix import directive commit 14370a5 Merge: b53bea4 11088a1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:51:35 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit b53bea4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:12:26 2020 +0900 Delete debug print commit 54b80a9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Nov 3 00:06:43 2020 +0900 Change to call prediction_type correctly. commit 4736dbe Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:56:38 2020 +0900 Minor fix commit a953daf Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:32:11 2020 +0900 Delete wrong README.md commit 68cad87 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:30:50 2020 +0900 Fix wrong symlink commit 8c42416 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:26:50 2020 +0900 Split clamping values commit e5e5f43 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:21:46 2020 +0900 Follow google's style of docstring commit c98b9cb Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:21:31 2020 +0900 Add user defined conf of hydra commit 181412d Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:20:44 2020 +0900 Revet deletion of unnecessary values commit 7e6a175 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:19:49 2020 +0900 Use ENUM type instead of str to specify the prediction type commit e491de8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 23:19:00 2020 +0900 Use data_prep.py of 00-svs-world. commit a743943 Merge: 5dfedd1 25c9f01 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 21:41:34 2020 +0900 Merge branch 'master' into mdn_dev commit 5dfedd1 Merge: 63a6f87 c3dd805 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Nov 2 00:00:32 2020 +0900 Merge branch 'master' into mdn_dev commit 63a6f87 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 19:15:21 2020 +0900 Add clamping log_pi and and torch.distributions.Normal.log_prob. commit 09a1f11 Merge: f029a95 52168d2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:56:44 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit f029a95 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:51:53 2020 +0900 Minor fix commit 0c8eab8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:50:18 2020 +0900 minor fix commit 3c053bf Merge: 86513fb b2c426f Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:29:27 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 86513fb Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:11:21 2020 +0900 Fix comment commit 56e96c2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:05:21 2020 +0900 Remove comment commit 3d2c046 Author: Tarou Shirani <taroushirani@gmail.com> Date: Fri Oct 30 00:04:43 2020 +0900 Remove lower limit of log_prob commit bc1eef0 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 23:55:30 2020 +0900 Fix variables commit 2d9c35c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 22:45:03 2020 +0900 Add comments commit 37156eb Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 21:10:16 2020 +0900 Add lower limit to torch.distributions.Normal.log_prob commit 17c9ac8 Merge: 514087a 1251bb2 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 02:31:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 514087a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 29 01:43:37 2020 +0900 Change to use log_pi and log_sigma instead of pi and sigma. commit 825a75b Merge: 8e9de33 a94ebc1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:35:02 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 8e9de33 Author: Tarou Shirani <taroushirani@gmail.com> Date: Mon Oct 19 19:27:04 2020 +0900 Minor fix commit e86a197 Merge: 6c9d4e9 605a095 Author: Tarou Shirani <taroushirani@gmail.com> Date: Sun Oct 18 21:07:30 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 6c9d4e9 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Oct 13 18:18:43 2020 +0900 Change to use most probable mu as conditional mode. commit 2b9649c Merge: fa901d8 9aeaef1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:55:34 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit fa901d8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:27 2020 +0900 Minor fix commit 4869f47 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:53:05 2020 +0900 Remove debug print commit 83cf6de Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:42:34 2020 +0900 Delete debug print. commit 6a61a6b Merge: 5b02542 31b7bf8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:50 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 5b02542 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:25:03 2020 +0900 Fix typo commit 31b7bf8 Merge: 8fc1e4c 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:05:54 2020 +0900 Merge branch 'mdn_dev' of https://github.com/taroushirani/nnsvs into mdn_dev commit 8fc1e4c Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:04:29 2020 +0900 Change num_gaussians commit 8455087 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:02:54 2020 +0900 Change RMDN to MDN commit 2b5e03d Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 8 01:00:53 2020 +0900 Fix typo commit 68aaba5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 22:01:58 2020 +0900 Fix scaling of sigma commit 3925221 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 17:21:22 2020 +0900 Fix to handle predictions without dyanmic_features commit 2cfc81f Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:50:20 2020 +0900 Change to call multi_stream_mlpg from predict_* commit fbd40d3 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Oct 7 16:49:56 2020 +0900 Remove mdn_sample_mode and add mdn_get_most_probable_sigma_and_mu. commit 4560322 Merge: e4a66e5 4305844 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 17:58:33 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit e4a66e5 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 16:50:40 2020 +0900 Minor fix. commit e489c82 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:48:21 2020 +0900 Change num_gaussians from 4 to 2 commit c3aa7e8 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:35:36 2020 +0900 Minor fix. commit 502cae1 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:27:48 2020 +0900 Correct comments. commit 8e52754 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 15:15:00 2020 +0900 Fix arithmetical bug of mdn_loss commit e4c299a Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 14:54:17 2020 +0900 Change the caluculation order. commit 291e808 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:49:24 2020 +0900 Change MDN to RMDN commit 4b49d9b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:45:58 2020 +0900 Add ReLU to RMDN commit 8699803 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 09:39:26 2020 +0900 Change sigma to log_sigma commit 55feef5 Merge: 68248ac 3a55678 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 02:41:03 2020 +0900 Merge branch 'mdn_pr' into mdn_dev commit 68248ac Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 01:56:55 2020 +0900 Add note. commit 1c4276b Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:53:17 2020 +0900 Change batch_size, nepochs, and models. commit 587e4e4 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:51:33 2020 +0900 Minor fix commit 733b35f Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:49:37 2020 +0900 Change to make use of mask. commit 44b4158 Author: Tarou Shirani <taroushirani@gmail.com> Date: Thu Oct 1 00:37:57 2020 +0900 Remove Conv1dResnetMDN commit d273dba Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 23:22:14 2020 +0900 Change MDN to RMDN commit 949cefc Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 22:31:58 2020 +0900 Change RMDN model. commit 09aec48 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:54:32 2020 +0900 Minor fix commit 131d978 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 19:45:10 2020 +0900 Add Conv1dResnetMDN commit dea9c93 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:53:57 2020 +0900 Remove nn.Softmax for numerical stability. commit 5e18115 Author: Tarou Shirani <taroushirani@gmail.com> Date: Wed Sep 30 17:50:58 2020 +0900 Change epoch to 300 commit c8cdc70 Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:48:24 2020 +0900 Change to use mdn_sample instead of mdn_sample_mode commit 1aae4fd Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 20:01:58 2020 +0900 Change to use nn.ReLU() istead of nn.Tanh() commit cc5ea5b Author: Tarou Shirani <taroushirani@gmail.com> Date: Tue Sep 29 01:03:10 2020 +0900 Fix comment and change not to call forward directly.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Let's merge and play with it.
Many thanks for your great contribution!! |
Thank you for your code review. I probably fixed all problems you pointed out above. I met numerical instability when I tried Conv1dResnetMDN(namely Conv1dResnet + MDN). I'll submit PR after MDN is merged. As far as I know, git rebase re-assign new hash to each commit and make the log straight, but it will destroy the repository which refers to the old hashes. To the amateur eye, I feel "squash merge" is safer than rebase. |
Hello, I made sample implementation of Mixture Density Networks for better timelag prediction following Sinsy[1].
Contrary to my expectations, there is no statistically significance among the current timelag model and MDN-based models.
(As for duration model, the results of MDN-based models are poorer than that of conventional one(see statistical analysis below)).
I checked my code repeatedly but I could not find my mistakes. Could anyone please advise me to improve the quality?
In this experiment I use NIT-SONG070 data-set, and the number of Gaussian components of MDN and RMDN model are
set to 500, 100 respectively.
Statistical analysis
timelag: ShapiroResult(statistic=0.943882942199707, pvalue=9.166962328110984e-12)
duration: ShapiroResult(statistic=0.6752718687057495, pvalue=1.4900654531575684e-37)
Errors seems not to be distributed normally, so we use a non-parametric test.
timelag:KruskalResult(statistic=0.12859137306783802, pvalue=0.9377276850847137)
duration:KruskalResult(statistic=18.355098064661238, pvalue=0.00010333349007178473)
There seems to be statistically significance among duration errors of these models.
The conventional model is statistically more accurate than other 2 models.