[MXNET-1210 ] Gluon Audio - Example #13325

gaurav-gireesh · 2018-11-20T01:57:09Z

Description

Contribute an Example for performing a task with Audio data that demonstrates the following abilities:

to be able to load audio (only .wav files supported currently) files and make an AudioDataset (NDArrays),
apply some popular audio transforms on the audio data( example scaling, MEL, MFCC etc.),
load the Dataset using Gluon's DataLoader, train a neural network ( Ex: MLP) with this transformed audio dataset,
perform a simple audio data related task such as sounds classification - 1 audio clip with 1 label( Multiclass sound classification problem).
have an end to end example for a sample Audio multi class classification task (Urban Sounds Classification)

Note : This example uses AudioFolderDataset and applies transforms to extract features from audio file and does a classification task. Current design of the AudioFolderDataset is below:
https://cwiki.apache.org/confluence/display/MXNET/Gluon+-+Audio

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
Changes are complete (i.e. I finished coding on this PR)
All changes have test coverage:
Steps in the example can be followed to test the example
Code is well-documented
To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change
Example contributed

Comments

This feature allows to take first step in audio data related tasks and needs to be extended for more generic as well as advanced use cases
- Design here- Dev list discussion here

Testing

Tested on p2.8xlarge instance ( Deep Learning AMI (Ubuntu) Version 18.0 )

stu1130 · 2018-11-20T02:12:34Z

@mxnet-label-bot add [pr-awaiting-review]
Thanks for your contribution @gaurav-gireesh

gaurav-gireesh · 2018-11-20T17:13:24Z

@sandeep-krishnamurthy @roywei Example created.
Could you help review this PR?
Thank you.

roywei

generally looks good, few comments. Really looking forward to seeing this design can be extended on more complex models.

roywei · 2018-11-20T18:20:41Z

example/gluon/urban_sounds/datasets.py

+try:
+ import librosa
+except ImportError as e:
+ warnings.warn("gluon/contrib/data/audio/datasets.py : librosa dependency could not be resolved or \


change warning message, not in contrib now

roywei · 2018-11-20T18:50:34Z

example/gluon/urban_sounds/README.md

@@ -0,0 +1,22 @@
+# Urban Sounds classification in MXNet


Link your design doc

roywei · 2018-11-20T19:27:18Z

example/gluon/urban_sounds/datasets.py

+ train_csv: str, default None
+ train_csv should be populated by the training csv filename
+ file_format: str, default '.wav'
+ The format of the audio files(.wav, .mp3)


mp3 is not supported now, right?

roywei · 2018-11-20T19:42:04Z

example/gluon/urban_sounds/datasets.py

+ label = self.items[idx][1]
+
+ if librosa is not None:
+ X1, _ = librosa.load(filename, res_type='kaiser_fast')


what is 'kaiser_fast' for 'res_type'

roywei · 2018-11-20T19:44:09Z

example/gluon/urban_sounds/datasets.py

+ return len(self.items)
+
+
+ def transform_first(self, fn, lazy=True):


why default for lazy is True, but False is passed in super?
Explain the reason and recommendation of choosing default values in doc string or comment

roywei · 2018-11-20T19:49:47Z

example/gluon/urban_sounds/transforms.py

+try:
+ import librosa
+except ImportError as e:
+ warnings.warn("gluon/contrib/data/audio/transforms.py : librosa dependency could not be resolved or \


change message

stu1130

Great work! @gaurav-gireesh

stu1130 · 2018-11-20T19:24:39Z

example/gluon/urban_sounds/datasets.py

+ self._train_csv = train_csv
+ if file_format.lower() not in self._exts:
+ warnings.warn("format {} not supported currently.".format(file_format))
+ return


I think raise is better here

if file_format.lower() not in self._exts: raise NotImplementedError('format {} not supported currently.".format(file_format)')

stu1130 · 2018-11-20T19:31:11Z

example/gluon/urban_sounds/datasets.py

+ for folder in sorted(os.listdir(root)):
+ path = os.path.join(root, folder)
+ if not os.path.isdir(path):
+ warnings.warn('Ignoring %s, which is not a directory.'%path, stacklevel=3)


nit: unify warning formatting style: either % or .foramt()

stu1130 · 2018-11-20T19:37:54Z

example/gluon/urban_sounds/datasets.py

+ with open("./synset.txt", "w") as synsets_file:
+ for item in self.synsets:
+ synsets_file.write(item+os.linesep)
+ print("Synsets is generated as synset.txt")


nit: one more space between generated and as

stu1130 · 2018-11-20T19:42:00Z

example/gluon/urban_sounds/datasets.py

+ self._label = nd.array(label_tmp)
+ for i, _ in enumerate(data_tmp):
+ if self._format not in data_tmp[i]:
+ self.items.append((data_tmp[i]+self._format, self._label[i]))


can we do the self.items.append in line 114 to reduce the # of for-loop?

Yes. Thanks.

stu1130 · 2018-11-20T19:54:38Z

example/gluon/urban_sounds/train.py

+def evaluate_accuracy(data_iterator, net):
+ """Function to evaluate accuracy of any data iterator passed to it as an argument"""
+ acc = mx.metric.Accuracy()
+ for _, (data, label) in enumerate(data_iterator):


if we don't use the index here why not just use regular for loop?

Yes! Refactored the code.

stu1130 · 2018-11-20T19:54:57Z

example/gluon/urban_sounds/train.py

+
+ for e in range(epochs):
+ cumulative_loss = 0
+ for _, (data, label) in enumerate(audio_train_loader):


Removed the variable.

stu1130 · 2018-11-20T20:19:57Z

example/gluon/urban_sounds/transforms.py

+ y = x.asnumpy()
+ else:
+ warnings.warn("MFCC - allowed datatypes mx.nd.NDArray and numpy.ndarray")
+ return x


the return value is not what user expects, raise exception would be better here IMO

Yes. I agree. However, other frameworks also return the unmodified value in these cases instead of raising error notifying them with a warning. Users can then use some other libraries to extract mfcc. This is the rationale behind this implementation.

stu1130 · 2018-11-20T20:21:13Z

example/gluon/urban_sounds/transforms.py

+
+ def forward(self, x):
+ if isinstance(x, np.ndarray):
+ return nd.array(x/self.scale_factor)


check self.scale_factor is not zero

Thanks. Added that test.

roywei

Could you add in README what are the steps/commands to run training and prediction, what arguments should users pass, what's your training time, accuracy and some sample output?

roywei

looks good now, few more spellings inline, and please address the comment above

roywei · 2018-11-26T23:51:31Z

example/gluon/urban_sounds/datasets.py

+ root/drilling/26.wav
+ root/dog_barking/42.wav
+ OR
+ Files(wav) and a csv file that has filename and associated label


nit: file name

roywei · 2018-11-26T23:51:44Z

example/gluon/urban_sounds/datasets.py

+ Parameters
+ ----------
+ fn : callable
+ A transformer function that takes the first elemtn of a sample


nit: element

roywei · 2018-11-26T23:52:25Z

example/gluon/urban_sounds/transforms.py

+ max_len : int
+ Length to which the array will be padded or trimmed to.
+ fill_value: int or float
+ If there is a need of padding, what value to padd at the end of the input array


roywei · 2018-11-26T23:52:54Z

example/gluon/urban_sounds/transforms.py

+ sampling_rate: int, default 22050
+ sampling rate of the input audio signal
+ num_fft: int, default 2048
+ length of the Fast fourier transform window


nit: Fourier

sandeep-krishnamurthy

Thanks for your contributions.

should this be under folder - example/gluon/audio/urban_sounds ?
Please run lint checks.
requirements.txt will be useful.
prepare_data.py will be useful.

sandeep-krishnamurthy · 2018-11-27T00:05:08Z

example/gluon/urban_sounds/README.md

+To be able to run this example:
+
+1. Download the dataset(train.zip, test.zip) required for this example from the location:
+**https://drive.google.com/drive/folders/0By0bAi7hOBAFUHVXd1JCN3MwTEU**


nit: Why bold? Should be hyperlinks only?

sandeep-krishnamurthy · 2018-11-27T00:06:10Z

example/gluon/urban_sounds/README.md

+**https://drive.google.com/drive/folders/0By0bAi7hOBAFUHVXd1JCN3MwTEU**
+
+
+2. Extract both the zip archives into the **current directory** - after unzipping you would get 2 new folders namely,\


can we provide simple python script to download and prepare data? something like prepare_dataset.py that users can run as 1st step.
Less the manual step the better.

sandeep-krishnamurthy · 2018-11-27T00:06:56Z

example/gluon/urban_sounds/README.md

+
+3. Apache MXNet is installed on the machine. For instructions, go to the link: **https://mxnet.incubator.apache.org/install/**
+
+4. Librosa is installed. To install, use the commands


Should we add a note to say, against which version of Librosa, this example is tested and working fine.

Added the version in the README as well as in requirements.txt.

sandeep-krishnamurthy · 2018-11-27T00:08:16Z

example/gluon/urban_sounds/datasets.py

+try:
+ import librosa
+except ImportError as e:
+ warnings.warn("librosa dependency could not be resolved or \


This will continue execution after printing warning. Is this intended?

Raise an ImportError here

Raised an Import error with a warning.

sandeep-krishnamurthy · 2018-11-27T00:09:42Z

example/gluon/urban_sounds/datasets.py

+ Attributes
+ ----------
+ synsets : list
+ List of class names. `synsets[i]` is the name for the integer label `i`


for ith label

Addressed this.

sandeep-krishnamurthy · 2018-11-27T00:11:56Z

example/gluon/urban_sounds/datasets.py

+ """
+ self.synsets = []
+ self.items = []
+ if self._train_csv is None:


Please add code comments for each of this section or better have separate functions to handled if csv is provided or folder.

if not self._train_csv ?

Refactored code to handle the logic in separate functions.

sandeep-krishnamurthy · 2018-11-27T00:14:20Z

example/gluon/urban_sounds/model.py

+ net.add(gluon.nn.Dense(256, activation="relu")) # 1st layer (256 nodes)
+ net.add(gluon.nn.Dense(256, activation="relu")) # 2nd hidden layer
+ net.add(gluon.nn.Dense(num_labels))
+ net.collect_params().initialize(mx.init.Normal(1.))


xavier is better? or doesn't matter?

Changed to Xavier.

sandeep-krishnamurthy · 2018-11-27T00:15:42Z

example/gluon/urban_sounds/predict.py

+if __name__ == '__main__':
+ try:
+ import argparse
+ parser = argparse.ArgumentParser(description="Urban Sounds clsssification example - MXNet")


MXNet Gluon

Thanks. Corrected this.

sandeep-krishnamurthy · 2018-11-27T00:16:50Z

example/gluon/urban_sounds/predict.py

+ pred_dir = args.pred
+
+ except ImportError:
+ warnings.warn("Argparse module not installed! passing default arguments.")


It would be good, if you can add requirements.txt file in this example folder. Have a step in readme to install pre-requisites using this requirements.txt

you could also provide a setup script to install all dependencies

Good point. I have added a requirements.txt file and a step in README for installing pre-requisites.

sandeep-krishnamurthy · 2018-11-27T00:20:57Z

example/gluon/urban_sounds/transforms.py

+ return x / self.scale_factor
+
+
+class PadTrim(Block):


This looks like a generally useful transforms. Post this PR, can you please see if this can be part of gluon transforms?

Yes. Thanks.

vandanavk · 2018-11-27T17:17:21Z

example/gluon/urban_sounds/datasets.py

+ if skip_header:
+ skip_rows = 1
+ else:
+ skip_rows = 0


maybe

skip_rows = 0 if skip_header: skip_rows = 1

?

vandanavk · 2018-11-27T17:17:53Z

example/gluon/urban_sounds/datasets.py

+try:
+ import librosa
+except ImportError as e:
+ warnings.warn("librosa dependency could not be resolved or \


Raise an ImportError here

vandanavk · 2018-11-27T17:18:42Z

example/gluon/urban_sounds/datasets.py

+ """
+ self.synsets = []
+ self.items = []
+ if self._train_csv is None:


if not self._train_csv ?

vandanavk · 2018-11-27T17:22:40Z

example/gluon/urban_sounds/datasets.py

+ for line in traincsv:
+ skipped_rows = skipped_rows + 1
+ if skipped_rows <= skip_rows:
+ continue


for skipping multiple rows in csv, could you explore https://python-forum.io/Thread-How-to-Loop-CSV-File-Beginning-at-Specific-Row?pid=29676#pid29676 or https://stackoverflow.com/questions/40403971/skip-multiple-rows-in-python ?

Yes. Good point Vandana. Made use of itertools and csv modules to start read at a particular row.

vandanavk · 2018-11-27T17:23:56Z

example/gluon/urban_sounds/datasets.py

+ def __getitem__(self, idx):
+ """Retrieve the item (data, label) stored at idx in items"""
+ filename = self.items[idx][0]
+ label = self.items[idx][1]


filename, label = self.items[idx]

vandanavk · 2018-11-27T17:28:15Z

example/gluon/urban_sounds/predict.py

+ pred_dir = args.pred
+
+ except ImportError:
+ warnings.warn("Argparse module not installed! passing default arguments.")


you could also provide a setup script to install all dependencies

vandanavk · 2018-11-27T17:33:05Z

example/gluon/urban_sounds/train.py

+ if args.train:
+ training_dir = args.train
+ else:
+ training_dir = './Train'


training_dir = './Train' ... ... try: ... ... if args: if args.train: training_dir = args.train

similar for the ones below

vandanavk · 2018-11-27T17:34:11Z

example/gluon/urban_sounds/train.py

+ training_dir = './Train'
+ training_csv = './train.csv'
+ eps = 30
+ batch_sz = 32


move this initialization before try, then it wont be required inside try or except.

Moved the initialization block above try catch blocks.

vandanavk · 2018-11-27T17:36:38Z

example/gluon/urban_sounds/transforms.py

+
+ def forward(self, x):
+ if isinstance(x, np.ndarray):
+ y = x


what is the default value for y?

Since, the argument x could be of numpy.ndarray or mxnet.nd.NDArray types, I do a conversion to numpy types as librosa needs numpy argument. So y is unchanged if it is of type numpy else it is converted into numpy before being passed for MFCC feature extraction.

vandanavk · 2018-11-27T17:38:23Z

example/gluon/urban_sounds/transforms.py

+ specs = librosa.feature.melspectrogram(x, sr=self._sampling_rate,\
+ n_fft=self._num_fft, n_mels=self._num_mels, hop_length=self._hop_length)
+ return nd.array(specs)
+


new line at the end

stu1130

some minor changes
The rest LGTM

stu1130 · 2018-11-28T02:30:54Z

example/gluon/urban_sounds/datasets.py

+ fn : callable
+ A transformer function that takes the first element of a sample
+ as input and returns the transformed element.
+ lazy : bool, default True


nit: default False

Thanks. corrected this.

stu1130 · 2018-11-28T02:33:37Z

example/gluon/urban_sounds/datasets.py

+ The transformed dataset.
+
+ """
+ return super(AudioFolderDataset, self).transform_first(fn, lazy=False)


Suggested change

return super(AudioFolderDataset, self).transform_first(fn, lazy=False)

return super(AudioFolderDataset, self).transform_first(fn, lazy=lazy)

Passing lazy from the arguments now.

stu1130 · 2018-11-28T02:51:18Z

example/gluon/urban_sounds/train.py

+
+
+def train(train_dir=None, train_csv=None, epochs=30, batch_size=32):
+ """The function responsible for running the training the model."""


nit: unify the docstring style

Suggested change

"""The function responsible for running the training the model."""

"""Function responsible for running the training the model."""

Corrected this.

stu1130 · 2018-11-28T02:58:19Z

example/gluon/urban_sounds/train.py

+
+ if e%5 == 0:
+ train_accuracy = evaluate_accuracy(audio_train_loader, net)
+ print("Epoch %s. Loss: %s Train accuracy : %s " % (e, cumulative_loss/num_examples, train_accuracy))


nit: unify the print style

Suggested change

print("Epoch %s. Loss: %s Train accuracy : %s " % (e, cumulative_loss/num_examples, train_accuracy))

print("Epoch {}. Loss: {} Train accuracy : {} ".format(e, cumulative_loss/ num_examples, train_accuracy))

kalyc

Added minor comments inline - can you add details about where has this example been tested?

kalyc · 2018-11-28T20:46:20Z

example/gluon/urban_sounds/README.md

+The details of the dataset and the link to download it are given below:
+
+
+Urban Sounds Dataset:


nit: make header

kalyc · 2018-11-28T20:46:52Z

example/gluon/urban_sounds/README.md

+## Description
+ The dataset contains 8732 wav files which are audio samples(<= 4s)) of street sounds like engine_idling, car_horn, children_playing, dog_barking and so on.
+ The task is to classify these audio samples into one of the 10 labels.
+


nit: please add list of available labels here as well

kalyc · 2018-11-28T20:48:26Z

example/gluon/urban_sounds/README.md

+
+To be able to run this example:
+
+1. `pip install -r ./requirements.txt`


Would mention that we need to go back to the directory: cd ../
would just leave it to pip install requirements.txt

kalyc · 2018-11-28T20:48:54Z

example/gluon/urban_sounds/README.md

+*https://librosa.github.io/librosa/install.html*
+
+2. Download the dataset(train.zip, test.zip) required for this example from the location:
+https://drive.google.com/drive/folders/0By0bAi7hOBAFUHVXd1JCN3MwTEU


Could you move this to a public S3 bucket instead?

You can use https://registry.opendata.aws/ this page to onboard your dataset onto a public S3 bucket - I would highly recommend doing so as the 1. Google drive link is external and we don't use it to store data in production
2. S3 buckets are more reliable and the example would not have any issues related to data availability in the future

kalyc · 2018-11-28T20:49:24Z

example/gluon/urban_sounds/README.md

+
+3. Extract both the zip archives into the **current directory** - after unzipping you would get 2 new folders namely,\
+ **Train** and **Test** and two csv files - **train.csv**, **test.csv**
+


please add comment about how the folder structure should look like for more clarity

kalyc · 2018-11-28T20:58:05Z

example/gluon/urban_sounds/train.py

+if __name__ == '__main__':
+ training_dir = './Train'
+ training_csv = './train.csv'
+ eps = 30


rename to epochs

kalyc · 2018-11-28T20:58:14Z

example/gluon/urban_sounds/train.py

+ training_dir = './Train'
+ training_csv = './train.csv'
+ eps = 30
+ batch_sz = 32


rename to batch_size

kalyc · 2018-11-28T20:59:35Z

example/gluon/urban_sounds/transforms.py

+ x = x.asnumpy()
+ specs = librosa.feature.melspectrogram(x, sr=self._sampling_rate,\
+ n_fft=self._num_fft, n_mels=self._num_mels, hop_length=self._hop_length)
+ return nd.array(specs)


add blank line at the end of every python file

kalyc · 2018-11-28T20:59:59Z

example/gluon/urban_sounds/transforms.py

+ return x
+
+
+class MEL(Block):


please use more verbose name here - what does MEL stand for?

The name MEL refers to a MEL scale frequently used with audio data(pitches and frequencies of audio signal). This isnt an abbreviation. And also, the idea was to have similar names as they appear in other framework so that people can intuitively port.

kalyc · 2018-11-28T21:00:27Z

example/gluon/urban_sounds/transforms.py

+
+# coding: utf-8
+# pylint: disable= arguments-differ
+"Audio transforms."


roywei

Thanks for the contribution! LGTM!

vandanavk

LGTM

kalyc · 2018-11-29T21:20:04Z

example/gluon/urban_sounds/README.md

@@ -0,0 +1,101 @@
+# Urban Sounds classification in MXNet


nit: Classification

kalyc

LGTM

sandeep-krishnamurthy

Thank you for this example, and patiently responding to all feedback comments.
LGTM.

This will be merged after CI pipeline is Green.

gaurav-gireesh · 2018-11-30T19:32:40Z

@marcoabreu Hi Marco! Can this PR be merged, if the required tests are passing, even if the new test steps are running? Thanks!

marcoabreu · 2018-11-30T19:36:49Z

Hey, yeah feel free to move ahead. The new statuses are equivalent to the old big pipeline :) Am Fr., 30. Nov. 2018, 20:33 hat Gaurav Gireesh <[email protected]> geschrieben:

…

@marcoabreu <https://github.com/marcoabreu> Hi Marco! Can this PR be merged, if the required tests are passing, even if the new test steps are running? Thanks! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#13325 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ARxB6_LLj3mI2tWiothGhS0lxIBZIeueks5u0Yf_gaJpZM4YqOjQ> .

gaurav-gireesh · 2018-12-01T18:00:35Z

@mxnet-label-bot remove [pr-awaiting-merge]

…ile (#13478) * updated to v1.5.0 * Bumped minor version from 1.4.0 to 1.5.0 on master * added Anirudh as maintainer for R package ... adding something useful and re-trigger PR check * Updated license file for clojure, onnx-tensorrt, gtest, R-package * Get the correct include path in pip package (#13452) * add find_include_path API * address reviewer comment * change return type from list to string * add unit test * address reviewer comment * address reviewer comment * address reviewer comment * address reviewer comment * fix include path problem in pip package * add comment * fix lint error * address reviewer comment * address reviewer comment * Use ~/.ccache as default ccache directory so is not cache is not erased on reboot (#13431) * Skip flaky test #13446 (#13480) * Rewrite dataloader with process pool, improves responsiveness and reliability (#13447) * fix recordio.py * rewrite dataloader with pool * fix batch as tuple * fix prefetching * fix pylint * picklable function * use pickle * add missing commit * Fix errors in docstrings for subgraph op; use code directive (#13463) * [MXNET-1158] JVM Memory Management Documentation (#13105) * update train_mnist * Add documentation for JVM Memory Management * update doc * address nit picks * address nit picks * Grammar and clarity edits for memory management doc * Edits for scala memory management * Update memory-management.md * Update memory-management.md * Update memory-management.md * capitalization fix * Update row_sparse tutorial (#13414) Update row_sparse tutorial * Add resiliency to onnx export code (#13426) * Added resiliency to onnx export code - With previous infer-shape implementation, if input shape was list instead of tuple or if extra non-existent parameters were provided, the code would still work. The fixes in this commit make sure that behavior is restored to prevent any compatibility issues with existing export code. * Fixed name of net in unittest * Fix pylint * [MXNET-1185] Support large array in several operators (part 1) (#13418) * fix a few operators with large arrays (# of elements) * fix bug in broadcast_div and add tests * address reviewer comment * add unit test * add empty line * retrigger CI * [MXNET-1210 ] Gluon Audio - Example (#13325) * Initialized the example * Addressed PR comments, about existing synset.txt file - no overwrite * RST - docstring issues fixed * added README * Addressed PR comments * Addressed PR comments, checking Divide by 0 * Raising error if format is not supported. * changed a line for ndarray of labels * Trigger CI * Trigger CI * PR comments addressed around skip_header argument * Addressed PR comments around librosa import * PR Comments * Passing lazy=lazy from argument * Added PR comments, labels to README.MD * Trigger CI * Addressing PR Comments in README * Modified README.md * Added example under audio folder * Retrigger CI * Retrigger CI * ONNX export: Instance normalization, Shape (#12920) * ONNX import/export: Make backend_rep common * ONNX export: Instance Normalization * ONNX export: Shape operator * Clarify dependency on OpenCV in CNN Visualization tutorial. (#13495) * clarify ops faq regarding docs strings (#13492) * Add graph_compact operator. (#13436) * add graph_compact. * fix. * add doc. * add tests for graph_compact. * address comments. * update docs. * trigger CI * Deprecate Jenkinsfile (#13474) * update github location for sampled_block.py (#13508) Updated to https://github.com/dmlc/gluon-nlp/blob/master/src/gluonnlp/model/sampled_block.py * #13453 [Clojure] - Add Spec Validations to the Optimizer namespace (#13499) * ONNX export: Logical operators (#12852) * Fix cmake options parsing in dev_menu (#13458) Add GPU+MKLDNN unittests to dev_menu * Revert "Manually track num_max_thread (#12380)" (#13501) This reverts commit 7541021. * Feature/mkldnn static 2 (#13503) * build mkldnn as static lib * update makefile to statically build mkldnn * build static mkldnn * fix static name * fix static name * update static for mac * rename mkldnn dep in ci * remove moving mkldnn dynamic lib * remove commented code * remove mkldnn dnaymic for unitest * force static for mkldnn lib * remove dynamic mkldnn bind * only link windows * add mkldnn.mk * try force linking * remove mkldnn dynanmic check * remove test mkldnn install * fix spacing * fix index * add artifacts * add comment about windows * remove static * update makefile * fix toctree Sphinx errors (#13489) * fix toctree errors * nudging file for CI * Disabled flaky test test_gluon_data.test_recordimage_dataset_with_data_loader_multiworker (#13527) * [MXNET-1234] Fix shape inference problems in Activation backward (#13409) * Provide a failing test for ReLU activation shape inference bug * Fix Activation backward shape inference fixes: #13333 * Add softsign Activation to test_gluon.py * Use activation in GPU if we are using CUDNN and not MKLDNN as it's happening right now * Don't disable MKLDNN

* Initialized the example * Addressed PR comments, about existing synset.txt file - no overwrite * RST - docstring issues fixed * added README * Addressed PR comments * Addressed PR comments, checking Divide by 0 * Raising error if format is not supported. * changed a line for ndarray of labels * Trigger CI * Trigger CI * PR comments addressed around skip_header argument * Addressed PR comments around librosa import * PR Comments * Passing lazy=lazy from argument * Added PR comments, labels to README.MD * Trigger CI * Addressing PR Comments in README * Modified README.md * Added example under audio folder * Retrigger CI * Retrigger CI

…ile (apache#13478) * updated to v1.5.0 * Bumped minor version from 1.4.0 to 1.5.0 on master * added Anirudh as maintainer for R package ... adding something useful and re-trigger PR check * Updated license file for clojure, onnx-tensorrt, gtest, R-package * Get the correct include path in pip package (apache#13452) * add find_include_path API * address reviewer comment * change return type from list to string * add unit test * address reviewer comment * address reviewer comment * address reviewer comment * address reviewer comment * fix include path problem in pip package * add comment * fix lint error * address reviewer comment * address reviewer comment * Use ~/.ccache as default ccache directory so is not cache is not erased on reboot (apache#13431) * Skip flaky test apache#13446 (apache#13480) * Rewrite dataloader with process pool, improves responsiveness and reliability (apache#13447) * fix recordio.py * rewrite dataloader with pool * fix batch as tuple * fix prefetching * fix pylint * picklable function * use pickle * add missing commit * Fix errors in docstrings for subgraph op; use code directive (apache#13463) * [MXNET-1158] JVM Memory Management Documentation (apache#13105) * update train_mnist * Add documentation for JVM Memory Management * update doc * address nit picks * address nit picks * Grammar and clarity edits for memory management doc * Edits for scala memory management * Update memory-management.md * Update memory-management.md * Update memory-management.md * capitalization fix * Update row_sparse tutorial (apache#13414) Update row_sparse tutorial * Add resiliency to onnx export code (apache#13426) * Added resiliency to onnx export code - With previous infer-shape implementation, if input shape was list instead of tuple or if extra non-existent parameters were provided, the code would still work. The fixes in this commit make sure that behavior is restored to prevent any compatibility issues with existing export code. * Fixed name of net in unittest * Fix pylint * [MXNET-1185] Support large array in several operators (part 1) (apache#13418) * fix a few operators with large arrays (# of elements) * fix bug in broadcast_div and add tests * address reviewer comment * add unit test * add empty line * retrigger CI * [MXNET-1210 ] Gluon Audio - Example (apache#13325) * Initialized the example * Addressed PR comments, about existing synset.txt file - no overwrite * RST - docstring issues fixed * added README * Addressed PR comments * Addressed PR comments, checking Divide by 0 * Raising error if format is not supported. * changed a line for ndarray of labels * Trigger CI * Trigger CI * PR comments addressed around skip_header argument * Addressed PR comments around librosa import * PR Comments * Passing lazy=lazy from argument * Added PR comments, labels to README.MD * Trigger CI * Addressing PR Comments in README * Modified README.md * Added example under audio folder * Retrigger CI * Retrigger CI * ONNX export: Instance normalization, Shape (apache#12920) * ONNX import/export: Make backend_rep common * ONNX export: Instance Normalization * ONNX export: Shape operator * Clarify dependency on OpenCV in CNN Visualization tutorial. (apache#13495) * clarify ops faq regarding docs strings (apache#13492) * Add graph_compact operator. (apache#13436) * add graph_compact. * fix. * add doc. * add tests for graph_compact. * address comments. * update docs. * trigger CI * Deprecate Jenkinsfile (apache#13474) * update github location for sampled_block.py (apache#13508) Updated to https://github.com/dmlc/gluon-nlp/blob/master/src/gluonnlp/model/sampled_block.py * apache#13453 [Clojure] - Add Spec Validations to the Optimizer namespace (apache#13499) * ONNX export: Logical operators (apache#12852) * Fix cmake options parsing in dev_menu (apache#13458) Add GPU+MKLDNN unittests to dev_menu * Revert "Manually track num_max_thread (apache#12380)" (apache#13501) This reverts commit 7541021. * Feature/mkldnn static 2 (apache#13503) * build mkldnn as static lib * update makefile to statically build mkldnn * build static mkldnn * fix static name * fix static name * update static for mac * rename mkldnn dep in ci * remove moving mkldnn dynamic lib * remove commented code * remove mkldnn dnaymic for unitest * force static for mkldnn lib * remove dynamic mkldnn bind * only link windows * add mkldnn.mk * try force linking * remove mkldnn dynanmic check * remove test mkldnn install * fix spacing * fix index * add artifacts * add comment about windows * remove static * update makefile * fix toctree Sphinx errors (apache#13489) * fix toctree errors * nudging file for CI * Disabled flaky test test_gluon_data.test_recordimage_dataset_with_data_loader_multiworker (apache#13527) * [MXNET-1234] Fix shape inference problems in Activation backward (apache#13409) * Provide a failing test for ReLU activation shape inference bug * Fix Activation backward shape inference fixes: apache#13333 * Add softsign Activation to test_gluon.py * Use activation in GPU if we are using CUDNN and not MKLDNN as it's happening right now * Don't disable MKLDNN

gaurav-gireesh added 4 commits November 16, 2018 18:53

Initialized the example

7fee29a

Addressed PR comments, about existing synset.txt file - no overwrite

8360a4e

RST - docstring issues fixed

5e00682

added README

3385a7d

gaurav-gireesh requested a review from szha as a code owner November 20, 2018 01:57

marcoabreu added the pr-awaiting-review PR is waiting for code review label Nov 20, 2018

szha mentioned this pull request Nov 20, 2018

[MXNET-1210 ] Gluon Audio #13241

Closed

8 tasks

roywei reviewed Nov 20, 2018

View reviewed changes

stu1130 reviewed Nov 20, 2018

View reviewed changes

gaurav-gireesh added 6 commits November 20, 2018 12:39

Addressed PR comments

6d029ae

Addressed PR comments, checking Divide by 0

1e30f7c

Raising error if format is not supported.

662749b

changed a line for ndarray of labels

acf48c4

Trigger CI

5e37fb8

Trigger CI

4fe850c

roywei reviewed Nov 26, 2018

View reviewed changes

sandeep-krishnamurthy reviewed Nov 27, 2018

View reviewed changes

PR comments addressed around skip_header argument

214d4ba

vandanavk reviewed Nov 27, 2018

View reviewed changes

Addressed PR comments around librosa import

75e1507

stu1130 reviewed Nov 28, 2018

View reviewed changes

gaurav-gireesh added 2 commits November 27, 2018 20:13

PR Comments

cc3714a

Passing lazy=lazy from argument

51101f2

kalyc reviewed Nov 28, 2018

View reviewed changes

Added PR comments, labels to README.MD

c41b9b3

roywei approved these changes Nov 28, 2018

View reviewed changes

vandanavk approved these changes Nov 29, 2018

View reviewed changes

Trigger CI

5eef58f

kalyc reviewed Nov 29, 2018

View reviewed changes

example/gluon/urban_sounds/README.md Outdated

@@ -0,0 +1,101 @@

# Urban Sounds classification in MXNet

Copy link

Contributor

kalyc Nov 29, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: Classification

gaurav-gireesh added 2 commits November 29, 2018 13:28

Addressing PR Comments in README

2465b0c

Modified README.md

4e0d541

kalyc approved these changes Nov 29, 2018

View reviewed changes

Added example under audio folder

74106e0

sandeep-krishnamurthy approved these changes Nov 30, 2018

View reviewed changes

sandeep-krishnamurthy added pr-awaiting-merge Review and CI is complete. Ready to Merge and removed pr-awaiting-review PR is waiting for code review labels Nov 30, 2018

gaurav-gireesh added 2 commits November 30, 2018 14:22

Retrigger CI

5eb923e

Retrigger CI

5461bc7

sandeep-krishnamurthy merged commit baeada4 into apache:master Dec 1, 2018

marcoabreu removed the pr-awaiting-merge Review and CI is complete. Ready to Merge label Dec 1, 2018

		return len(self.items)


		def transform_first(self, fn, lazy=True):

		https://drive.google.com/drive/folders/0By0bAi7hOBAFUHVXd1JCN3MwTEU


		2. Extract both the zip archives into the current directory - after unzipping you would get 2 new folders namely,\


		3. Apache MXNet is installed on the machine. For instructions, go to the link: https://mxnet.incubator.apache.org/install/

		4. Librosa is installed. To install, use the commands

	return super(AudioFolderDataset, self).transform_first(fn, lazy=False)
	return super(AudioFolderDataset, self).transform_first(fn, lazy=lazy)



		def train(train_dir=None, train_csv=None, epochs=30, batch_size=32):
		"""The function responsible for running the training the model."""

	"""The function responsible for running the training the model."""
	"""Function responsible for running the training the model."""

	print("Epoch %s. Loss: %s Train accuracy : %s " % (e, cumulative_loss/num_examples, train_accuracy))
	print("Epoch {}. Loss: {} Train accuracy : {} ".format(e, cumulative_loss/ num_examples, train_accuracy))

		The details of the dataset and the link to download it are given below:


		Urban Sounds Dataset:


		To be able to run this example:

		1. `pip install -r ./requirements.txt`


		3. Extract both the zip archives into the current directory - after unzipping you would get 2 new folders namely,\
		Train and Test and two csv files - train.csv, test.csv

[MXNET-1210 ] Gluon Audio - Example #13325

[MXNET-1210 ] Gluon Audio - Example #13325

Conversation

gaurav-gireesh commented Nov 20, 2018 • edited Loading

Description

Checklist

Essentials

Comments

Testing

stu1130 commented Nov 20, 2018

gaurav-gireesh commented Nov 20, 2018

roywei left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

stu1130 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

roywei left a comment • edited Loading

Choose a reason for hiding this comment

roywei left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sandeep-krishnamurthy left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

stu1130 left a comment

gaurav-gireesh commented Nov 20, 2018 •

edited

Loading

roywei left a comment •

edited

Loading

roywei left a comment •

edited

Loading