search

deep-dance / core

GNU General Public License v3.0
2 stars 1 forks source link

Run 3 - experiments #52

Open erak opened 3 years ago

erak commented 3 years ago 
dvc exp run --queue -S train.epochs=10 -S train.batch_size=32 -S train.look_back=70 -S train.lstm_layer=32 -S train.mdn_layer=5 -S train.dancers=all -S train.validation_split=0.3 -S generate.look_back=70 -S generate.dancers=all -S generate.validation_split=0.3

dvc exp run --queue -S train.epochs=10 -S train.batch_size=32 -S train.look_back=70 -S train.lstm_layer=64 -S train.mdn_layer=5 -S train.dancers=all -S train.validation_split=0.3 -S generate.look_back=70 -S generate.dancers=all -S generate.validation_split=0.3

dvc exp run --queue -S train.epochs=10 -S train.batch_size=32 -S train.look_back=70 -S train.lstm_layer=128 -S train.mdn_layer=5 -S train.dancers=all -S train.validation_split=0.3 -S generate.look_back=70 -S generate.dancers=all -S generate.validation_split=0.3
erak commented 3 years ago

I've checked the output an did a little runtime analysis for the experiments mention above (for all dancers, only impro, validation_split: 0.3 and test_size: 0.05):