v1.7.6

[marcopeix]

New Features

• [FEAT]: Support providing DataLoader arguments to optimize GPU usage @jasminerienecker ([FEAT] support providing DataLoader arguments to optimize GPU usage #1186)
• [FEAT]: Set activation function in GRN of TFT @marcopeix (FEAT - Set activation function in GRN of TFT #1175)
• [FEAT]: Conformal Predictions in NeuralForecast @JQGoh ([FEAT] Conformal Predictions in NeuralForecast #1171)

Bug Fixes

• [FIX]: Ability load models saved using versions before 1.7 @tylernisonoff (Fix: Ability load models saved using versions before 1.7 #1207)
• [FIX]: Conformal prediction issues @elephaint ([FIX] Conformal prediction issues #1179)
• [FIX]: Feature importance when using only hist_exog in TFT fails @elephaint ([FIX] Feature importance when using only hist_exog in TFT fails #1174)
• [FIX]: Remove unused output layer NBEATSx @elephaint ([FIX] Remove unused output layer NBEATSx #1168)
• [FIX]: Fix Tweedie loss @elephaint ([FIX] Fix Tweedie loss #1164)
• [FIX]: MLPMultivariate incorrect static_exog parsing @elephaint ([FIX] MLPMultivariate incorrect static_exog parsing #1170)
• [FIX]: Deprecate activation functions for GRU @marcopeix (feat: Deprecate activation functions for GRU #1198)

Documentation

• [DOC]: Tutorial on cross-validation @marcopeix (DOC: Tutorial on cross-validation #1176)
• [DOC]: Build docs on release only @elephaint ([FIX] Build docs on release only #1183)

---

This discussion was created from the release v1.7.6.

[VivCh14]

After upgrading to this the neuralforecast lib does not work. I had a working model-training script that showed the tqdm progressbar when I trained the NHITS and BiTCN models but now after the upgrade, neither numba seems to be used nor the progress bar is shown and the model does not train. Has anyone had similar troubles ?