PYTHONPATH="."  luigi  \
--module train CoOccurrenceTraining  \
--project config.base_rnn \
--local-scheduler  \
--data-frames-preparation-extra-params '{
  "sample_days": 500, 
  "test_days": 30,
  "window_trip": 5,
  "column_stratification": "user_id",
  "filter_last_step": true,
  "balance_sample_step": 200000,
  "filter_trip_size": 0 }' \
--test-size 0 \
--early-stopping-min-delta 0.0001 \
--learning-rate 0.001 \
--metrics='["loss"]' \
--batch-size 128 \
--loss-function ce \
--epochs 100


PYTHONPATH="." luigi --module evaluation EvaluationTask \
--model-task-class "train.CoOccurrenceTraining" \
--model-task-id CoOccurrenceTraining____mars_gym_model_b____563c1ab497 \
--file "/media/workspace/booking_challenge/output/booking/dataset/test_500_30_5.csv"  \
--local-scheduler \
--model-eval "coocorrence"

mars-gym run supervised --project config.conf1_rnn \
  --recommender-module-class model.RNNAttModel \
  --recommender-extra-params '{
    "n_factors": 100, 
    "hidden_size": 300, 
    "n_layers": 1, 
    "dropout": 0.3, 
    "window_trip": 10,
    "from_index_mapping": false,
    "path_item_embedding": false, 
    "freeze_embedding": false}' \
  --data-frames-preparation-extra-params '{
    "test_split": 0.1, 
    "window_trip": 10,
    "user_features_file": "all_user_features_10.csv",
    "column_stratification": "user_id",
    "filter_last_step": true,
    "balance_sample_step": 200000,
    "filter_trip_size": 0 }' \
  --test-size 0 \
  --val-size 0.1 \
  --early-stopping-min-delta 0.0001 \
  --early-stopping-patience 5 \
  --learning-rate 0.001 \
  --metrics='["loss", "top_k_acc", "top_k_acc2"]' \
  --batch-size 128 \
  --loss-function ce \
  --loss-function-class loss.FocalLoss \
  --loss-function-params '{
    "alpha":1,
    "gamma":3,
    "c": 0.8,
    "epsilon": 0.1
    }' \
  --epochs 100 \
  --generator-workers 8

mars-gym run supervised --project config.conf1_rnn \
  --recommender-module-class model.NARMModel \
  --recommender-extra-params '{
    "n_factors": 50, 
    "hidden_size": 300, 
    "n_layers": 1, 
    "dropout": 0.2, 
    "n_user_features": 10,
    "n_dense_features": 5,
    "from_index_mapping": false,
    "path_item_embedding": false, 
    "freeze_embedding": false}' \
  --data-frames-preparation-extra-params '{
    "test_split": 0.1, 
    "window_trip": 10,
    "user_features_file": "all_user_features_10.csv",
    "column_stratification": "user_id",
    "filter_last_step": true,
    "balance_sample_step": 200000,
    "filter_trip_size": 0 }' \
  --test-size 0 \
  --val-size 0.1 \
  --early-stopping-min-delta 0.0001 \
  --early-stopping-patience 5 \
  --learning-rate 0.001 \
  --metrics='["loss", "top_k_acc", "top_k_acc2"]' \
  --batch-size 64 \
  --optimizer "radam" \
  --optimizer-params '{
    "weight_decay": 0.01
    }' \
  --loss-function ce \
  --loss-function-class loss.FocalLoss \
  --loss-function-params '{
    "alpha":1,
    "gamma":3,
    "c": 0.5,
    "epsilon": 0.1
    }' \
  --epochs 100 \
  --monitor-metric val_top_k_acc \
  --monitor-mode max \
  --generator-workers 4

mars-gym run supervised --project config.conf1_rnn \
  --recommender-module-class model.Caser \
  --recommender-extra-params '{
      "n_factors": 100, 
      "p_L": 10, 
      "p_nh": 16,
      "p_nv": 4,  
      "dropout": 0.2, 
      "hist_size": 10, 
      "from_index_mapping": false,
      "path_item_embedding": false, 
      "freeze_embedding": false}' \
  --data-frames-preparation-extra-params '{
    "test_split": 0.1, 
    "window_trip": 10,
    "column_stratification": "user_id",
    "filter_last_step": true,
    "balance_sample_step": 200000,
    "filter_trip_size": 0 }' \
  --test-size 0 \
  --val-size 0.1 \
  --early-stopping-min-delta 0.0001 \
  --learning-rate 0.001 \
  --metrics='["loss", "top_k_acc"]' \
  --loss-function ce \
  --loss-function-class loss.FocalLoss \
  --batch-size 128 \
  --epochs 100

mars-gym run supervised --project config.conf1_rnn \
  --recommender-module-class model.MLTransformerModel \
  --recommender-extra-params '{
      "n_factors": 100, 
      "n_hid": 50,
      "n_head": 2,
      "n_layers": 1,
      "num_filters": 100,
      "dropout": 0.2, 
      "hist_size": 5, 
      "from_index_mapping": false,
      "path_item_embedding": false, 
      "freeze_embedding": false}' \
  --data-frames-preparation-extra-params '{
    "sample_days": 500, 
    "test_days": 30,
    "window_trip": 5,
    "column_stratification": "user_id",
    "filter_last_step": true,
    "balance_sample_step": 500000,
    "filter_trip_size": 0 }' \
  --test-size 0 \
  --early-stopping-min-delta 0.0001 \
  --learning-rate 0.001 \
  --metrics='["loss"]' \
  --loss-function ce \
  --batch-size 128 \
  --epochs 100