Dev #19
Dev #19
Conversation
Tensorflow v2
|
Let's wait for the regressor refactoring on #22 before me make a new release! |
|
The pull request build failure above failed with |
|
The dev branch build on my fork passed - |
…base classifier to base/estimators/_classifier.py. Tests pass.
…ase regressor to base/estimators/_regressor.py. Tests pass.
…e (e.g. a network without its final layer) into deeplearning/[architecture name]. Moved base network to base/estimators/_network.py. Tests pass.
…them to the network, classifier, regressor structure already used for CNN and MLP.
…hey take on Travis.
…hey take on Travis.
…t data moved to adjust_parameters(X) and extra adjustment added to handle the short series that might be used for forecasting. Modified slice_data() to handle y that can be regression values or int class labels and where y can be 1D array (regression) or 2D array (classification). Modified pre_processing() to handle y that can be 1D array or 2D array.
* fixed forecasting test for time series regressors * Update test_regressors.py Updated test to reflect that the model lists are dicts now * Removing MCDCNNRegressor_quick from model lists (and so tests) to confirm other tests complete successfully * MCDCNNRegressor seems to need extra epochs * including mcdcnn back into the model lists * Update test_regressors.py Co-authored-by: James Large <44509982+James-Large@users.noreply.github.com>
Sktime upgrade
…sle for one particular test case just because its not getting enough data
|
My lord, the tests passed. Validation tests have been marked as slow and not to run on travis, as too often there's one or two envs where it hits the 10 min time out while the rest are fine. After release, there'll be some issues put up and we'll do a roadmap post for 0.3.0 in the future, one of them will be revamping the tests and streamlining them. Part of me dreads merging this and then one of the tests fails due to bad initialisation again on master Either way, think this is a long over due merge. Good to submit reviews? |
|
Thanks for persevering @james-large. Good to merge 👍 |
|
Sorry for the delay! |
|
Fantastic to see this merge go through 🥳 |
|
@Withington I don't think we've set up the CI so that tags automatically trigger uploads to pypi. Implementing that would be really helpful. The sktime base repo could serve as an example. Otherwise we can do it by downloading the build artefacts and manually uploading them pypi. |
Long time no PR, getting this into master before working on new things. Proposing that this updates sktime-dl to 0.2.0
See #14 for most of the changes.
Major:
Also included are some updates to the setup.py and travis tests.