Clipping predictions

[deven-gqc]

Hello people, I’m using tsai to make predictions on a dataset and I can’t figure out the reason behind the output of the model.

If you take a look at the image below, the output of the predicted model keeps getting flattened out. What could be the reason behind this? Is it the loss function? or the architecture?

Here is the code and the shapes of the inputs

batch_tfms = [TSStandardize(by_sample=True)]
learn = TSRegressor(X, y, batch_tfms=batch_tfms, splits=split, 
                    arch=InceptionTimePlus, metrics=mae, bs=params['bs'], train_metrics=True)
lr = learn.lr_find()
learn.fine_tune(100, lr[0])

[oguiza]

Hi @deven-gqc,
Sorry for my late reply.
There's no default clipping applied in tsai.
I've tried reproducing your issue with dummy data but haven't been able to.
Here's a gist I've used.