Brave (#20)

* commit changes

* commit changes

* commit changes

* commit changes

* commit changes

* Update workflows.py

* Updates

* Updates

* Updates

* Changes

* Pinning requirements

* Pinning dev requirements

* Updated datasource

* Added data cleaning and preprocessing logic

* Code linting

* Code linting

* Added black to requirements

* Updated gitignore file

* Updated preprocessing function

* Added training script

* Created flyte workflow

* workflowoutput

Co-authored-by: Ali Abbas Jaffri <[email protected]>

.gitignore

@@ -143,3 +143,6 @@ dmypy.json
 
 # Pyre type checker
 .pyre/
+
+*/gtzan/*
+projects/bravemusic/bravemusic/gtzan/

projects/bravemusic/.dockerignore

@@ -0,0 +1 @@
+!.flyte

projects/bravemusic/.flyte/remote-config.yaml

@@ -0,0 +1,17 @@
+admin:
+  # For GRPC endpoints you might want to use dns:///flyte.myexample.com
+  endpoint: dns:///playground.hosted.unionai.cloud
+  authType: Pkce
+  # Change insecure flag to ensure that you use the right setting for your environment
+  insecure: false
+storage:
+  type: stow
+  stow:
+    kind: s3
+    config:
+      auth_type: iam
+      region: us-east-2
+logger:
+  # Logger settings to control logger output. Useful to debug logger:
+  show-source: true
+  level: 1

projects/bravemusic/.flyte/remote.config

@@ -0,0 +1,6 @@
+[sdk]
+# This option specifies the python packages where-in to search for workflows and tasks workflow packages. These workflows and tasks are then serialized during the `make serialize` commands
+workflow_packages=bravemusic
+
+[auth]
+raw_output_data_prefix=s3://open-compute-playground

projects/bravemusic/.flyte/sandbox-config.yaml

@@ -0,0 +1,19 @@
+admin:
+  # For GRPC endpoints you might want to use dns:///flyte.myexample.com
+  endpoint: dns:///localhost:30081
+  authType: Pkce
+  insecure: true
+logger:
+  show-source: true
+  level: 0
+storage:
+  connection:
+    access-key: minio
+    auth-type: accesskey
+    disable-ssl: true
+    endpoint: http://localhost:30084
+    region: us-east-1
+    secret-key: miniostorage
+  type: minio
+  container: "my-s3-bucket"
+  enable-multicontainer: true

projects/bravemusic/.flyte/sandbox.config

@@ -0,0 +1,6 @@
+[sdk]
+# This option specifies the python packages where-in to search for workflows and tasks workflow packages. These workflows and tasks are then serialized during the `make serialize` commands
+workflow_packages=bravemusic
+
+[auth]
+raw_output_data_prefix=s3://my-s3-bucket/flytelab

projects/bravemusic/.gitpod.yml

@@ -0,0 +1,7 @@
+tasks:
+  - init: |
+      python -m venv ~/venvs/brave
+      source ~/venvs/brave/bin/activate
+      pip install -r requirements.txt -r requirements-dev.txt
+    command: python3 projects/bravemusic/bravemusic/workflows.py
+

projects/bravemusic/Dockerfile

@@ -0,0 +1,39 @@
+FROM python:3.9-slim-buster
+
+WORKDIR /root
+ENV VENV /opt/venv
+ENV LANG C.UTF-8
+ENV LC_ALL C.UTF-8
+ENV PYTHONPATH /root
+
+# e.g. flyte.config or sandbox.config
+ARG config
+
+RUN apt-get update && \
+    apt-get install -y \
+    libsm6 \
+    libxext6 \
+    libxrender-dev \
+    ffmpeg \
+    build-essential
+
+# Install the AWS cli separately to prevent issues with boto being written over
+RUN pip3 install awscli
+
+ENV VENV /opt/venv
+
+# Virtual environment
+RUN python3 -m venv ${VENV}
+ENV PATH="${VENV}/bin:$PATH"
+
+# Install Python dependencies
+COPY requirements.txt /root
+RUN pip install -r /root/requirements.txt
+
+COPY bravemusic /root/bravemusic
+COPY $config /root/flyte.config
+
+# This image is supplied by the build script and will be used to determine the version
+# when registering tasks, workflows, and launch plans
+ARG image
+ENV FLYTE_INTERNAL_IMAGE $image

projects/bravemusic/README.md

@@ -0,0 +1,28 @@
+# Design Doc: Brave-Hyenas-2
+## MLOps Community: Engineering labs
+
+| Team name           |      brave-hyenas-2                      |
+|---------------------|:----------------------------------------:|
+|Project name         |      brave-hyenas-2                      | 
+| Project description |     Hackathon - brave-hyenas-2 team      |
+|Using GPUs? (Yes/No) |  No                                      | 
+
+
+
+### Problem Statement
+What problem are you solving?  
+It’s usually hard to identify correctly what kind of music genre is playing thus our team embraced in tackling to classify music genre using deep learning.  
+
+
+### ......
+
+
+
+
+
+![new](https://user-images.githubusercontent.com/85021780/161294904-a4158856-0558-424f-9f07-85aef8f4b423.jpg)  
+
+
+
+
+### Solution (working progress) ....

projects/bravemusic/bravemusic/datasource.py

@@ -0,0 +1,33 @@
+import os
+import tarfile
+import git
+
+GIT_URL = "https://huggingface.co/datasets/marsyas/gtzan"
+GTZAN_PATH = "./gtzan"
+GTZAN_ZIP_FILE_PATH = "./gtzan/data"
+GTZAN_ZIP_FILE_NAME = "genres.tar.gz"
+
+
+class Progress(git.remote.RemoteProgress):
+    def update(self, op_code, cur_count, max_count=None, message=""):
+        print(self._cur_line)
+
+
+def download_gtzan_repo():
+    if not os.path.isdir(GTZAN_PATH) or not any(os.scandir(GTZAN_PATH)):
+        git.Repo.clone_from(url=GIT_URL, to_path=GTZAN_PATH, progress=Progress())
+        extract_gtzan_repo_tarball()
+    else:
+        print("dataset already exists")
+
+
+def extract_gtzan_repo_tarball():
+    # open file
+    file = tarfile.open(f"{GTZAN_ZIP_FILE_PATH}/{GTZAN_ZIP_FILE_NAME}")
+    # extracting file
+    file.extractall(GTZAN_ZIP_FILE_PATH)
+    file.close()
+
+
+if __name__ == "__main__":
+    download_gtzan_repo()

projects/bravemusic/bravemusic/preprocess.py

@@ -0,0 +1,95 @@
+import os
+import math
+import librosa
+from datasource import download_gtzan_repo, GTZAN_ZIP_FILE_PATH
+
+SAMPLE_RATE = 22050
+TRACK_DURATION = 30  # measured in seconds
+SAMPLES_PER_TRACK = SAMPLE_RATE * TRACK_DURATION
+BAD_FORMATS = ["jazz.00054.wav"]
+
+
+def clean_dataset():
+    for (dir_path, dir_names, filenames) in os.walk(f"{GTZAN_ZIP_FILE_PATH}/genres/"):
+        print(dir_path)
+        [
+            os.remove(f"{dir_path}{filename}")
+            for filename in filenames
+            if not filename.endswith(".wav")
+        ]
+        [
+            os.renames(
+                old=f"{dir_path}/{filename}",
+                new=f"{dir_path}/{filename}".replace("._", ""),
+            )
+            for filename in filenames
+            if f"{dir_path}/{filename}".startswith("._")
+        ]
+        [
+            os.remove(f"{dir_path}/{filename}")
+            for filename in filenames
+            if filename.startswith("._")
+        ]
+
+
+def preprocess(
+    dataset_path: str,
+    num_mfcc: int = 13,
+    n_fft: int = 2048,
+    hop_length: int = 512,
+    num_segments: int = 10,
+) -> dict:
+    data = {"mapping": [], "labels": [], "mfcc": []}
+
+    samples_per_segment = int(SAMPLES_PER_TRACK / num_segments)
+    num_mfcc_vectors_per_segment = math.ceil(samples_per_segment / hop_length)
+
+    # loop through all genre sub-folder
+    for i, (dir_path, dir_names, filenames) in enumerate(
+        os.walk(f"{GTZAN_ZIP_FILE_PATH}/genres/")
+    ):
+        # ensure we're processing a genre sub-folder level
+        if dir_path is not dataset_path:
+            # save genre label (i.e., sub-folder name) in the mapping
+            semantic_label = dir_path.split("/")[-1]
+            print(semantic_label)
+            data["mapping"].append(semantic_label)
+            print("Processing: {}".format(semantic_label))
+
+            # process all audio files in genre sub-dir
+            for f in filenames:
+                if f not in BAD_FORMATS:
+                    # load audio file
+                    file_path = os.path.join(dir_path, f)
+                    signal, sample_rate = librosa.load(path=file_path, sr=SAMPLE_RATE)
+
+                    # process all segments of audio file
+                    for d in range(num_segments):
+
+                        # calculate start and finish sample for current segment
+                        start = samples_per_segment * d
+                        finish = start + samples_per_segment
+
+                        # extract mfcc
+                        mfcc = librosa.feature.mfcc(
+                            y=signal[start:finish],
+                            sr=sample_rate,
+                            n_mfcc=num_mfcc,
+                            n_fft=n_fft,
+                            hop_length=hop_length,
+                        )
+                        mfcc = mfcc.T
+
+                        # store only mfcc feature with expected number of vectors
+                        if len(mfcc) == num_mfcc_vectors_per_segment:
+                            data["mfcc"].append(mfcc.tolist())
+                            data["labels"].append(i - 1)
+                            # print("{}, segment:{}".format(file_path, d + 1))
+    return data
+
+
+if __name__ == "__main__":
+    download_gtzan_repo()
+    clean_dataset()
+    data = preprocess(dataset_path=GTZAN_ZIP_FILE_PATH)
+    print(data)

projects/bravemusic/bravemusic/train.py

@@ -0,0 +1,76 @@
+import json
+import typing
+import warnings
+import numpy as np
+from tensorflow import keras
+from dataclasses import dataclass
+from preprocess import preprocess
+from datasource import GTZAN_ZIP_FILE_PATH
+from dataclasses_json import dataclass_json
+from flytekit.types.directory import FlyteDirectory
+from sklearn.model_selection import train_test_split
+
+
+warnings.filterwarnings("ignore")
+MODELSAVE = [typing.TypeVar("str")]
+model_file = typing.NamedTuple("Model", model=FlyteDirectory[MODELSAVE])
+
+
+@dataclass_json
+@dataclass
+class Hyperparameters(object):
+    batch_size: int = 32
+    metrics: str = "accuracy"
+    loss = ("sparse_categorical_crossentropy",)
+    epochs: int = 30
+    learning_rate: float = 0.0001
+
+
+def train(
+    data: dict,
+    hp: Hyperparameters
+) -> model_file:
+    # with open("data.json", "r") as fp:
+    #     data = json.load(fp)
+
+    # convert lists to numpy arrays
+    X = np.array(data["mfcc"])
+    y = np.array(data["labels"])
+
+    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)
+
+    model = keras.Sequential(
+        [
+            keras.layers.Flatten(input_shape=(X.shape[1], X.shape[2])),
+            keras.layers.Dense(512, activation="relu"),
+            keras.layers.Dense(256, activation="relu"),
+            keras.layers.Dense(64, activation="relu"),
+            keras.layers.Dense(10, activation="softmax"),
+        ]
+    )
+    optimiser = keras.optimizers.Adam(learning_rate=hp.learning_rate)
+    model.compile(
+        optimizer=optimiser,
+        loss=hp.loss,
+        metrics=[hp.metrics],
+    )
+    # train model
+    model.fit(
+        X_train,
+        y_train,
+        validation_data=(X_test, y_test),
+        batch_size=hp.batch_size,
+        epochs=hp.epochs,
+    )
+
+    Dir = "model"
+    model.save(Dir)
+    return model
+
+
+if __name__ == '__main__':
+    data = preprocess(dataset_path=GTZAN_ZIP_FILE_PATH)
+    model = train(
+        data=data,
+        hp=Hyperparameters(epochs=1)
+    )