Refactoring solver classes

About this version

• Refactored Solver class
• Refactored TimeSeriesForecaster class
• Changed default config values
• Added new default config values
• Added new results

Adding new modules

About this Version

• Added Informer
• Added MAE and MSE metrics
• Updated README
• Added StepScheduler
• Added new result

Bug fixing

About this version

• Fixed a minor bug which occurs when using gpu
  Since I found bias added is not moved to gpu, I added lines like bias.to(device).

Bug fixing

About this version

• Fixed predict bug
  Fixed a bug that bias is used wrongly.

Fixing ADD_LAST_STEP_VAL

About this version

• Fixed ADD_LAST_STEP_VAL
  Before, ADD_LAST_STEP_VAL can be set to True only when input/output features are the same. We fixed ADD_LAST_STEP_VAL can be set to True even if input/output features are different.

• Updated README.md

Bug fixing

About this version

Changed to initialize X_scaler and y_scaler separately
Expanded ADD_LAST_STEP_VAL to output each feature
Fixed Seq2Seq to work on multiple output features

New Benchmark Result

• Added schedulers
• Added deterministic results
• Fixed Seq2Seq
• Updated NBeats result
• Added Seq2Seq result

Bug fixing (+α)

About this PR

Fixed StandardScaler dimension bug
Modified Seq2Seq (added a new layer to regressor and bias terms)
Added TrendBlock to NBeats

• Fixed minor bugs

First benchmark result

About this version

Added the first benchmark result on benchmark directory (added and tested modules around it)

Bug fixing

About this version

Fixed a minor bug of predict method