jzhoubu / lstmexplainer

Update

• 2019/01/30

  • Meeting: Pivot direction to SHAP

• 2019/01/31

  • Direction: Produce a LSTM model from Kaggle dataset for Experiment
  • Experiment:
    • Requirement: {keras: 2.2.2, tensorflow: 1.10.0, shap: 0.28.3}
    • Data set: Stock time series data
    • Shape
      • X_train: (2709, 60, 4) --> (batch, time_stamps, features)
      • y_train: (2709, )
      • LSTM first layer input: (60, 4)-->(time_stamps, features)

      model.add(LSTM(units=50, return_sequences=True, input_shape=(X_train.shape[1],4)))

      • SHAP values output: (1, 5, 60, 4)
      • shap_values[0][1] shape: (60, 4) --> (time_stamps, features)

• 2019/01/31

  • Experiment:
    • On server LSTM

• 2019/02/03

  • Meeting at everning 22:00