oeawai_challenge

Instrument Classfication Challenge for the OEAW AI Summer School: https://www.kaggle.com/c/oeawai/

Our score in private board was 0.77231 (F1). It was achieved by an geometric average of the outputs of two models:

SpectralResNet18 (private score alone: 0.75906) learns the MFC trained for 9 epochs with Adam Optimizer, lr 0.0001, batch size 8 on GeForce GTX1080

MSResNet (private score alone: 0.71199) learns the raw audio trained for 10 epochs with Adam Optimizer, lr 0.0001, batch size 8 on GeForce GTX1080

Name		Name	Last commit message	Last commit date
Latest commit History 58 Commits
.DS_Store		.DS_Store
.gitignore		.gitignore
FFTMulitScale1DResNet.py		FFTMulitScale1DResNet.py
FeatureFNN.py		FeatureFNN.py
Gif and sounds for challenge introduction.ipynb		Gif and sounds for challenge introduction.ipynb
LSTM.py		LSTM.py
MelRawCombined.py		MelRawCombined.py
MulitScale1DResNet.py		MulitScale1DResNet.py
OurSolution.ipynb		OurSolution.ipynb
README.md		README.md
Resnet.py		Resnet.py
SVM solution.ipynb		SVM solution.ipynb
SpectralResNet.py		SpectralResNet.py
SpectrogramCNN.py		SpectrogramCNN.py
cnn-example.ipynb		cnn-example.ipynb
evaluation_utils.py		evaluation_utils.py
load dataset example.ipynb		load dataset example.ipynb
main.py		main.py
math_utils.py		math_utils.py
testDataset.py		testDataset.py
trainDataset.py		trainDataset.py
train_utils.py		train_utils.py
validation_split.py		validation_split.py

Provide feedback