tsai

tsai copied to clipboard

Hi @oguiza, I get an error when I try to use the XceptionTImePlus for multi-step forecasting. I tried a simple code as well and it still gives me the same error: learn = TSForecaster(X_train, y_train, splits=splits, tfms=tfms, bs=512, arch=XceptionTimePlus, arch_config={}, metrics=rmse)# , cbs=FastAIPruningCallback(trial))

Error found:

Hi @MOREDataset, Sorry for my late reply. I believe this issue is fixed now. I've run the following code and don't get the issue you mention:

X, y, splits = get_UCR_data('LSST', split_data=False)
tfms = [None, TSClassification()]
learn = TSForecaster(X, y, splits=splits, tfms=tfms, bs=512, arch=XceptionTimePlus, arch_config={})
learn.fit_one_cycle(1)

getting the same error for the MiniRochetPlus architecture in TSForecaster.

Hi @MOREDataset, Sorry for my late reply. I believe this issue is fixed now. I've run the following code and don't get the issue you mention:

X, y, splits = get_UCR_data('LSST', split_data=False)
tfms = [None, TSClassification()]
learn = TSForecaster(X, y, splits=splits, tfms=tfms, bs=512, arch=XceptionTimePlus, arch_config={})
learn.fit_one_cycle(1)

这是来自QQ邮箱的假期自动回复邮件。您的邮件已收到，谢谢！

Hey there @oguiza, thank you for your efforts but I just tried the model again for time series forecasting and got the same error.

Hi @MOREDataset , @parshinsh, @mPowc , I've found custom_head was actually missing in 3 models: MiniRocketPlus, XceptiontimePlus, and FCNPlus. The issue is fixed now (#597).