joshvarty / audiotagging

Working on: https://www.kaggle.com/c/freesound-audio-tagging-2019

AudioTagging

Working on: https://www.kaggle.com/c/freesound-audio-tagging-2019

Running

1. Run 00_Preprocess.ipynb (Takes ~2.5 hours 😢)

   • Converts audio files into images and saves them
   • Turns string labels into binary indicators
   • Perform label smoothing on the noisy dataset
   • Merge the train_curated.csv and train_noisy.csv into train_merged.csv

2. Run 01_BasicModel.ipynb

   • Generates a single (balanced) validation fold based on the curated training set
   • Defines a few simple image transforms
   • Creates an ImageDataBunch with batch image normalization .normalize()
   • Creates a vgg16_bn learner that uses mixup data augmentation

3. Run src/trainAll.py

   • Performs a full training cycle with my best known hyperparameters and network
   • Creates a /kfolds folder with validation set predictions
   • Creates a /model_predictions folder with test set predictions
   • Creates a /model_source folder with the exact source used to generate a given score

Optional

• 00_EDA.ipynb is Exploratory Data Analysis

  • Visualize class balance
  • Visualize audio length
  • Find incorrect audio file 77b925c2.wav
  • Example on how to create a validation set that is
    • Only taken from curated dataset
    • Is balanced according to labels (using MultilabelStratifiedKFold)

• 01_ExploringActivations.ipynb

  • Looking at the activations of the network to make sure nothing seems problematic

• 03_ImageStats.ipynb

  • My attempt to compute image statistics for normalization (similar to imagenet_stats or mnist_stats)
  • Unfortunately using the statistics from this doesn't improve performance
    • I have probably made a mistake and am misunderstanding how the statistics should be calculated.