pilot7747 / crowd-embeddings

Run

To reproduce the LSTM experiment, run

python3 -m benchmarks --name lstm_crowd_embeddings \
         --dataset=imdb \
         --train_path=data/imdb/train_crowd_alpha047.csv  --val_path=data/imdb/val_clean_alpha047.csv --test_path=data/imdb/test.csv \
         --log_dir=$HOME/crowd-embeddings/logs --checkpoint_root=$HOME/crowd-embeddings/checkpoints   \
         --backbone=lstm --approach=crowd_embedding   \
         --num_workers=8 \
         --max_epochs=25 --batch_size=32 \
         --inference_policy=top_k \
         --reproduction_iters=10 \
         --lr=0.0009886 --dropout=0.1 \
         --tune_iters=0 \
         --gpus=1 \
         2>&1 | tee train.log

Data

We have noticed that this service does not support Git LFS storage, so some data files are not available. To access these files, you can use this link which leads to a Google Drive folder created from an anonymous Google account.