Editing TensorFlow tutorial #1369

iriark01 · 2020-09-30T10:09:50Z

Ready for engineering review.

Please look at the source for commented-out questions.

You can also see a preview (on the VPN): https://os.mbed.com/docs/mbed-os/TensforFlow_edit/mbed-os-pelion/machine-learning-with-tensorflow-and-mbed-os.html

docs/reference/TensorFlow.md

spartacoos · 2020-09-30T11:11:43Z

docs/reference/TensorFlow.md

+* Git.

-**Getting started**
+* xxd.<!--unless is exists by default on all Linux computers?-->


Well, you are right, it does come with Ubuntu although it could be the case some minimalist distros don't include it for some reason. But since the tutorial focuses on Ubuntu we could leave this out.

Suggested change

* xxd.

spartacoos · 2020-09-30T12:42:42Z

docs/reference/TensorFlow.md

+    <span class="notes">**Note**: On a Mac, you might have to use gmake instead of make.</span>

-[Watch here](http://www.youtube.com/watch?v=x5MhydijWmc)
+* Python 2.7. We recommend [using pyenv to manage Python versions](https://pypi.org/project/pyenv/).<!--The install page for Mbed CLI now requires 3.7. Why are we asking for 2.7?-->


We tried with Python 3.7 but had issues from the TensorFlow side. See point 5 here.

And that doesn't cause problems with Mbed CLI?

Python 3.x please! What are the issues on the TF-side (apart from outdated tutorial)?

Yes, just tested again with a Python 3.7 environment and it worked fine.

docs/reference/TensorFlow.md

spartacoos · 2020-09-30T15:55:50Z

docs/reference/TensorFlow.md


-The code samples audio from the microphone on the K66F. The audio is run through a Fast Fourier transform to create a spectrogram. The spectrogram is then fed into a pre-trained machine learning model. The model uses a [convolutional neural network](https://developers.google.com/machine-learning/practica/image-classification/convolutional-neural-networks) to identify whether the sample represents either the command “yes” or “no”, silence, or an unknown input. We will explore how this works in more detail later in the guide.
+    <!--The install page for Mbed CLI now requires 3.7. Why are we asking for 2.7?-->
+    <!--need to explain what `config root .` and `deploy` do and why we need them - neither one is part of a standard workflow where you use Mbed CLI to import an application, so this is a special case-->


From what I understand, mbed config root . sets the current mbed-os folder as the root (meaning mbed command actually knows what Mbed OS project we want to run the following commands on) which then allows us to run mbed deploy which can pull the libraries needed (in our case in tensorflow/tensorflow/lite/micro/tools/make/gen/mbed_cortex-m4/prj/micro_speech/mbed/mbed-os/requirements.txt) and installs them. I was not aware that this is not a standard workflow, this is the workflow presented in the original guide hosted on developer.arm.com. If you know of a better way of doing this I'd be glad to explore that workflow and update the guide.

@donatieng can you confirm/deny?

See above, I think this is a shortcoming of the Makefile here

are we good here?

spartacoos · 2020-09-30T15:58:04Z

docs/reference/TensorFlow.md

-The code samples audio from the microphone on the K66F. The audio is run through a Fast Fourier transform to create a spectrogram. The spectrogram is then fed into a pre-trained machine learning model. The model uses a [convolutional neural network](https://developers.google.com/machine-learning/practica/image-classification/convolutional-neural-networks) to identify whether the sample represents either the command “yes” or “no”, silence, or an unknown input. We will explore how this works in more detail later in the guide.
+    <!--The install page for Mbed CLI now requires 3.7. Why are we asking for 2.7?-->
+    <!--need to explain what `config root .` and `deploy` do and why we need them - neither one is part of a standard workflow where you use Mbed CLI to import an application, so this is a special case-->
+    <!--and why are we compiling here? We compile again two steps down, with the flash parameter-->


You are right. This is the format presented in the original guide referenced above. However, we could take that out. One reason could be that we compile to ensure everything is working and only then we compile + flash the device.

The first compile step now has to come out, since it compiled with an old version of Mbed OS.

spartacoos · 2020-09-30T16:17:02Z

docs/reference/TensorFlow.md

+    <!--and why are we compiling here? We compile again two steps down, with the flash parameter-->

-Here are descriptions of some interesting source files:
+    For some compilers<!--we only support two, and you specifically asked for GNU, so in what case will this happen?-->, you may get a compilation error in `mbed_rtc_time.cpp`. Go to  `mbed-os/platform/mbed_rtc_time.h`  and comment out line 32 and line 37:


That was coming from the TensorFlow instructions at some point while we were developing the demo. However, I can't seem to find this warning anymore on their repo other than in the K66F instructions which I pushed myself but I just tried this workflow again and it works without commenting anything out so they must have fixed this issue. We can omit this step.

spartacoos · 2020-09-30T16:19:07Z

docs/reference/TensorFlow.md


-   [command_responder.cc](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/command_responder.cc) is called every time a potential command has been identified.
+1. Flash the application to the board:
+    <!--I compiled two steps ago - that probably needs to be removed-->


Yes, we could either remove the previous compile step or this one to be less redundant. I'd say it might be better to keep this one as it flashes automatically to the board.

spartacoos · 2020-09-30T16:22:41Z

docs/reference/TensorFlow.md

+    ```
+    mbed compile -m K66F -t GCC_ARM –flash
+    ```
+1. Deploy the example to your K66F<!--but we flshed already. What are we doing here?-->


Again, following the format in the original guide on developer.arm.com. They had two ways of doing things. One where we compile and flash on the same command mbed compile -m K66F -t GCC_ARM –flash and one where we just compile then copy the binary over to the board to flash it, as shown below with cp ./BUILD/K66F/GCC_ARM/mbed.bin /Volumes/K66F/ .

spartacoos · 2020-09-30T16:23:26Z

docs/reference/TensorFlow.md

-   [main.cc](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/main.cc). This file is the entry point for the Mbed program, which runs the machine learning model using TensorFlow Lite for Microcontrollers.
+    Copy the binary file that we built earlier to the USB storage.

+    Note: if you have skipped the previous steps<!--which ones?-->, download the [binary file]() to proceed.


all the steps up to compiling steps

spartacoos · 2020-09-30T16:25:51Z

docs/reference/TensorFlow.md

+    cp ./BUILD/K66F/GCC_ARM/mbed.bin /Volumes/K66F/
+    ```

+    Depending on your operating system <!--aren't we making everyone use Linux?-->, the exact copy command and paths may vary.


Good catch. I should have removed this from the original guide.

iriark01 · 2020-10-01T08:45:55Z

On hold until it's updated to Mbed OS 6

spartacoos · 2020-10-01T09:51:01Z

docs/reference/TensorFlow.md

+The model was trained on the one-second samples of audio we saw above. In the training data, the word “yes” or “no” is spoken at the start of the sample, and the entire word is contained within that one-second. However, when the application is running, there is no guarantee that a user will begin speaking at the very beginning of our one-second sample. If the user starts saying “yes” at the end of the sample instead of the beginning, the model might not be able to understand the word. This is because the model uses the position of the features within the sample to help predict which word was spoken.

-# Deploy the example to your K66F
+To solve this problem, our code runs inference as often as it can<!--here you lost me, probably because I don't deal with ML - is this in the training again? where is it getting new samples?-->, depending on the speed of the device, and averages all of the results within a rolling 1000ms window. The code in the file [recognize_commands.cc](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.cc) performs this action. When the average for a given category in a set of predictions goes above the threshold<!--lost me here too - are we talking about training or recognising actual users?-->, as defined in [recognize_commands.h](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.h), we can assume a valid result.


This part doesn't refer to training data but rather samples when the model is running on the device. Because we don't know at what point in the "window" a person will say the keyword this means we need to run inference many times so that we effectively take many snapshots of many windows then we analyse those "windows" and then take the average result as that's what is most likely being said. For a deeper explanation you can search "Windowing" on this book on TinyML.

Suggested change

To solve this problem, our code runs inference as often as it can, depending on the speed of the device, and averages all of the results within a rolling 1000ms window. The code in the file [recognize_commands.cc](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.cc) performs this action. When the average for a given category in a set of predictions goes above the threshold, as defined in [recognize_commands.h](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.h), we can assume a valid result.

To solve this problem, our code runs inference as often as it can, depending on the speed of the device, and averages all of the results within a rolling 1000ms window. The code in the file [recognize_commands.cc](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.cc) performs this action. When the average for a given category in a set of predictions goes above the threshold, as defined in [recognize_commands.h](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.h), we can assume a valid result.

spartacoos · 2020-10-01T10:04:16Z

docs/reference/TensorFlow.md

+    1. Save the file.

-   Modify the code and deploy to the device.
+1. Copy the `tiny_conv_micro_features_model_data.cc` <!--where is this file? it's not the one I was just editing-->file into the `tensorflow/tensorflow/lite/micro/tools/make/gen/mbed_cortex-m4/prj/micro_speech/mbed/tensorflow/lite/micro/examples/micro_speech/micro_features` directory.


Apologies. This file was originally called tiny_conv_micro_features_model_data.cc when the project was first made but this is now just micro_features/model.cc .

Suggested change

1. Copy the `tiny_conv_micro_features_model_data.cc` file into the `tensorflow/tensorflow/lite/micro/tools/make/gen/mbed_cortex-m4/prj/micro_speech/mbed/tensorflow/lite/micro/examples/micro_speech/micro_features` directory.

1. Copy the `model.cc` file into the `tensorflow/tensorflow/lite/micro/tools/make/gen/mbed_cortex-m4/prj/micro_speech/mbed/tensorflow/lite/micro/examples/micro_speech/micro_features` directory.

spartacoos · 2020-10-01T10:12:36Z

docs/reference/TensorFlow.md

+    ```

-To save time, we will skip this step and instead download the [tiny_conv.tflite](https://developer.arm.com/-/media/Files/downloads/Machine%20learning%20how-to%20guides/tiny_conv.tflite?revision=495eb362-4325-49b8-b3ba-3141df0c9b95&la=en&hash=0F37BA2C5DE95A1561979CDD18973171767A47C3).
+    The code uses this array to map the output of the model to the correct value. Because we specified our wanted_words<!--is that underscore correct?--> as “up, down” in the training script, we should update this array to reflect these words in the same order.


Yes, the underscore is correct but it should actually be all caps "WANTED_WORDS" as on the python training script.

Suggested change

The code uses this array to map the output of the model to the correct value. Because we specified our wanted_words as “up, down” in the training script, we should update this array to reflect these words in the same order.

The code uses this array to map the output of the model to the correct value. Because we specified our WANTED_WORDS as “up, down” in the training script, we should update this array to reflect these words in the same order.

spartacoos · 2020-10-01T10:13:51Z

docs/reference/TensorFlow.md

+    ```

-and open the file:
+1. Copy the binary to the USB storage of the device<!--why aren't we using flash?-->, using the same method that you used earlier.


I guess we can leave this out and show it as before. I'm not sure what the logic behind showing two ways of doing it was when the original guide was made.

Suggested change

1. Copy the binary to the USB storage of the device, using the same method that you used earlier.

spartacoos

Let me know if I missed any of your questions or if you need me to look at some of the questions in more detail such as why we are flashing in two different ways. I can discuss with Alessandro Grande from Arm who made the original guide with Pete Warden from Google.

spartacoos

I think this covers your comments plus the porting to Mbed OS 6.3. Feel free to ping me on Slack if anything else needs changing.

iriark01

Need to approve some changes I've made, and I would like @donatieng to weigh in on some things.

iriark01 · 2020-10-16T07:31:55Z

docs/reference/TensorFlow.md

-The code samples audio from the microphone on the K66F. The audio is run through a Fast Fourier transform to create a spectrogram. The spectrogram is then fed into a pre-trained machine learning model. The model uses a [convolutional neural network](https://developers.google.com/machine-learning/practica/image-classification/convolutional-neural-networks) to identify whether the sample represents either the command “yes” or “no”, silence, or an unknown input. We will explore how this works in more detail later in the guide.
+    <!--The install page for Mbed CLI now requires 3.7. Why are we asking for 2.7?-->
+    <!--need to explain what `config root .` and `deploy` do and why we need them - neither one is part of a standard workflow where you use Mbed CLI to import an application, so this is a special case-->
+    <!--and why are we compiling here? We compile again two steps down, with the flash parameter-->


The first compile step now has to come out, since it compiled with an old version of Mbed OS.

iriark01 · 2020-10-16T07:32:41Z

docs/reference/TensorFlow.md

+* Git.

-**Getting started**
+* xxd.<!--unless is exists by default on all Linux computers?-->


Suggested change

* xxd.

docs/reference/TensorFlow.md

iriark01 · 2020-10-16T07:34:57Z

docs/reference/TensorFlow.md

+The model was trained on the one-second samples of audio we saw above. In the training data, the word “yes” or “no” is spoken at the start of the sample, and the entire word is contained within that one-second. However, when the application is running, there is no guarantee that a user will begin speaking at the very beginning of our one-second sample. If the user starts saying “yes” at the end of the sample instead of the beginning, the model might not be able to understand the word. This is because the model uses the position of the features within the sample to help predict which word was spoken.

-# Deploy the example to your K66F
+To solve this problem, our code runs inference as often as it can<!--here you lost me, probably because I don't deal with ML - is this in the training again? where is it getting new samples?-->, depending on the speed of the device, and averages all of the results within a rolling 1000ms window. The code in the file [recognize_commands.cc](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.cc) performs this action. When the average for a given category in a set of predictions goes above the threshold<!--lost me here too - are we talking about training or recognising actual users?-->, as defined in [recognize_commands.h](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.h), we can assume a valid result.


Suggested change

To solve this problem, our code runs inference as often as it can, depending on the speed of the device, and averages all of the results within a rolling 1000ms window. The code in the file [recognize_commands.cc](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.cc) performs this action. When the average for a given category in a set of predictions goes above the threshold, as defined in [recognize_commands.h](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.h), we can assume a valid result.

To solve this problem, our code runs inference as often as it can, depending on the speed of the device, and averages all of the results within a rolling 1000ms window. The code in the file [recognize_commands.cc](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.cc) performs this action. When the average for a given category in a set of predictions goes above the threshold, as defined in [recognize_commands.h](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/micro_speech/recognize_commands.h), we can assume a valid result.

iriark01 · 2020-10-16T07:36:15Z

docs/reference/TensorFlow.md

+    1. Save the file.

-   Modify the code and deploy to the device.
+1. Copy the `tiny_conv_micro_features_model_data.cc` <!--where is this file? it's not the one I was just editing-->file into the `tensorflow/tensorflow/lite/micro/tools/make/gen/mbed_cortex-m4/prj/micro_speech/mbed/tensorflow/lite/micro/examples/micro_speech/micro_features` directory.


Suggested change

1. Copy the `tiny_conv_micro_features_model_data.cc` file into the `tensorflow/tensorflow/lite/micro/tools/make/gen/mbed_cortex-m4/prj/micro_speech/mbed/tensorflow/lite/micro/examples/micro_speech/micro_features` directory.

1. Copy the `model.cc` file into the `tensorflow/tensorflow/lite/micro/tools/make/gen/mbed_cortex-m4/prj/micro_speech/mbed/tensorflow/lite/micro/examples/micro_speech/micro_features` directory.

iriark01 · 2020-10-16T07:36:36Z

docs/reference/TensorFlow.md

+    ```

-To save time, we will skip this step and instead download the [tiny_conv.tflite](https://developer.arm.com/-/media/Files/downloads/Machine%20learning%20how-to%20guides/tiny_conv.tflite?revision=495eb362-4325-49b8-b3ba-3141df0c9b95&la=en&hash=0F37BA2C5DE95A1561979CDD18973171767A47C3).
+    The code uses this array to map the output of the model to the correct value. Because we specified our wanted_words<!--is that underscore correct?--> as “up, down” in the training script, we should update this array to reflect these words in the same order.


Suggested change

The code uses this array to map the output of the model to the correct value. Because we specified our wanted_words as “up, down” in the training script, we should update this array to reflect these words in the same order.

The code uses this array to map the output of the model to the correct value. Because we specified our WANTED_WORDS as “up, down” in the training script, we should update this array to reflect these words in the same order.

docs/reference/TensorFlow.md

iriark01 · 2020-10-16T07:37:06Z

docs/reference/TensorFlow.md

+    ```

-and open the file:
+1. Copy the binary to the USB storage of the device<!--why aren't we using flash?-->, using the same method that you used earlier.


Suggested change

1. Copy the binary to the USB storage of the device, using the same method that you used earlier.

iriark01 · 2020-10-16T07:37:47Z

docs/reference/TensorFlow.md

+    <span class="notes">**Note**: On a Mac, you might have to use gmake instead of make.</span>

-[Watch here](http://www.youtube.com/watch?v=x5MhydijWmc)
+* Python 2.7. We recommend [using pyenv to manage Python versions](https://pypi.org/project/pyenv/).<!--The install page for Mbed CLI now requires 3.7. Why are we asking for 2.7?-->


And that doesn't cause problems with Mbed CLI?

iriark01 · 2020-10-16T07:38:28Z

docs/reference/TensorFlow.md


-The code samples audio from the microphone on the K66F. The audio is run through a Fast Fourier transform to create a spectrogram. The spectrogram is then fed into a pre-trained machine learning model. The model uses a [convolutional neural network](https://developers.google.com/machine-learning/practica/image-classification/convolutional-neural-networks) to identify whether the sample represents either the command “yes” or “no”, silence, or an unknown input. We will explore how this works in more detail later in the guide.
+    <!--The install page for Mbed CLI now requires 3.7. Why are we asking for 2.7?-->
+    <!--need to explain what `config root .` and `deploy` do and why we need them - neither one is part of a standard workflow where you use Mbed CLI to import an application, so this is a special case-->


@donatieng can you confirm/deny?

donatieng

Thanks @COTASPAR - some feedback below.

docs/reference/TensorFlow.md

donatieng · 2020-10-27T12:59:58Z

docs/reference/TensorFlow.md

+    <span class="notes">**Note**: On a Mac, you might have to use gmake instead of make.</span>

-[Watch here](http://www.youtube.com/watch?v=x5MhydijWmc)
+* Python 2.7. We recommend [using pyenv to manage Python versions](https://pypi.org/project/pyenv/).<!--The install page for Mbed CLI now requires 3.7. Why are we asking for 2.7?-->


Python 3.x please! What are the issues on the TF-side (apart from outdated tutorial)?

donatieng · 2020-10-27T13:02:40Z

docs/reference/TensorFlow.md

+    cd tensorflow/tensorflow/lite/micro/mbed/
+    ```
+    
+1. Change the DebugLog function to use UnbufferedSerial instead since the Serial API was deprecated in Mbed OS 6:


Can we upstream this? Or link an open PR in the doc?

Many of the other TFLu examples still use older versions of Mbed OS and thus if we upstream it that could break those examples. Ideally, all of the examples would be ported to the latest version of mbed. However, I think for now this should be okay while perhaps we try to engage with the TFLu team to port the other examples.

As soon as this is published we will start to work with the TF team to discuss next steps.

donatieng · 2020-10-27T13:06:13Z

docs/reference/TensorFlow.md


-# Download and build the sample application
+    ```
+    mbed config root .


Probably a shortcoming of the TFLiteU Makefile above, ideally this wouldn't be needed so candidate for upstreaming too.

docs/reference/TensorFlow.md

Adding explanation for terminal output in reference to this [comment](#1369 (comment))

Fixing a few spelling mistakes and explanation of the benefits of local ML inference. Co-authored-by: Donatien Garnier <donatien.garnier@arm.com>

We are putting the version in the prerequisites.

In reference to [this](#1369 (comment))

In reference to [this suggested change](#1369 (comment))

Instead we just use the --flash command again. In reference to [this](#1369 (comment)).

spartacoos

Implemented the changes from the comments.

donatieng

Did a bit of clean-up but it looks good, thanks a lot for the work @COTASPAR !

iriark01 added 6 commits September 30, 2020 07:03

Generating preview

43f87d9

Moving file

574fb6d

Edit part 1: getting the example application running

6a940f6

Edit part 2: how it works

c3b7506

First draft ready

934e9e4

Moving file

5b4e027

spartacoos reviewed Sep 30, 2020

View reviewed changes

docs/reference/TensorFlow.md Outdated Show resolved Hide resolved

spartacoos reviewed Sep 30, 2020

View reviewed changes

docs/reference/TensorFlow.md Outdated Show resolved Hide resolved

spartacoos reviewed Sep 30, 2020

View reviewed changes

docs/reference/TensorFlow.md Outdated Show resolved Hide resolved

spartacoos reviewed Sep 30, 2020

View reviewed changes

iriark01 added code dependency wait to merge labels Oct 1, 2020

iriark01 self-assigned this Oct 1, 2020

spartacoos reviewed Oct 1, 2020

View reviewed changes

Updated instructions to work with Mbed OS 6.3.0

b828a0f

spartacoos suggested changes Oct 13, 2020

View reviewed changes

spartacoos closed this Oct 13, 2020

iriark01 reopened this Oct 13, 2020

iriark01 commented Oct 16, 2020

View reviewed changes

donatieng suggested changes Oct 27, 2020

View reviewed changes

spartacoos and others added 17 commits November 11, 2020 15:14

Update TensorFlow.md

436315f

Adding explanation for terminal output in reference to this [comment](#1369 (comment))

Update docs/reference/TensorFlow.md

6334f27

Fixing a few spelling mistakes and explanation of the benefits of local ML inference. Co-authored-by: Donatien Garnier <donatien.garnier@arm.com>

Updating prerequisites

ae740e5

Improving wording of the introduction to the demo

d34adc1

Clarification as to why mbed config root is used

9c496e4

Missing word

34bf7bb

Removing Make version error

a1e98e8

We are putting the version in the prerequisites.

Spelling mistake and wrong RespondToCommand info

37ffa6a

command responder code doesn't need to change k66f

e19c560

Explaining the model size constant

3a4df20

adding the --flash and --sterm flags

adda305

In reference to [this](#1369 (comment))

Correct constant from Python script

97313a8

In reference to [this](#1369 (comment))

Fixing file name changed in TensorFlow repo

19144d4

In reference to [this](#1369 (comment))

Removing comments

aa73e65

In reference to [this suggested change](#1369 (comment))

Removed instruction to move binary to the board

101e3cb

Instead we just use the --flash command again. In reference to [this](#1369 (comment)).

Fixing images link to make them local ./image/x

c8e3aec

Adding images folder

7bd9779

spartacoos approved these changes Nov 13, 2020

View reviewed changes

Donatien Garnier added 2 commits November 13, 2020 18:03

Fixed image links

2ca98fc

Removed inline comments

6e3b76e

donatieng approved these changes Nov 13, 2020

View reviewed changes

donatieng merged commit ccf5a54 into TensorFlow Nov 15, 2020

	1. Copy the `tiny_conv_micro_features_model_data.cc` <!--where is this file? it's not the one I was just editing-->file into the `tensorflow/tensorflow/lite/micro/tools/make/gen/mbed_cortex-m4/prj/micro_speech/mbed/tensorflow/lite/micro/examples/micro_speech/micro_features` directory.
	1. Copy the `model.cc` file into the `tensorflow/tensorflow/lite/micro/tools/make/gen/mbed_cortex-m4/prj/micro_speech/mbed/tensorflow/lite/micro/examples/micro_speech/micro_features` directory.

	The code uses this array to map the output of the model to the correct value. Because we specified our wanted_words<!--is that underscore correct?--> as “up, down” in the training script, we should update this array to reflect these words in the same order.
	The code uses this array to map the output of the model to the correct value. Because we specified our WANTED_WORDS as “up, down” in the training script, we should update this array to reflect these words in the same order.

Editing TensorFlow tutorial #1369

Editing TensorFlow tutorial #1369

Uh oh!

Conversation

iriark01 commented Sep 30, 2020

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

spartacoos Sep 30, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

iriark01 commented Oct 1, 2020

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

spartacoos left a comment

Choose a reason for hiding this comment

Uh oh!

spartacoos left a comment

Choose a reason for hiding this comment

Uh oh!

iriark01 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

spartacoos Sep 30, 2020 •

edited

Loading