[BUG] plot_model() is not working consistently when fh has gaps

[tugra-alp]

I used compare_model() to select top 3 models, and combine them with blend_model() function.

fh=np.arange(25,73) : 1 day gap and 2 days forecast horizon in each split (Hourly data).

custom_split = SlidingWindowSplitter(fh=np.arange(25,73) , window_length=1200, step_length=720)
exp = TimeSeriesExperiment()
exp.setup(data=data1, fold_strategy = custom_split, preprocess = False, session_id = 42 )
.
.
.
exp.plot_model(blended_model, plot='forecast')

After created blended_model object to see model performance, used plot_model and got this error.

Note: Even i create single model to see model performance, same error raise up.
Note: Sometimes this code snippet works fine, but when i re-run it, it comes with same error.

[ngupta23]

You seem to be passing a lot of custom things to the splitter. I can not tell the issue just by looking at the pictures. Can you share a minimally reproducible example (code + data) so I can debug? Thanks!

[ngupta23]

Can you try passing fh = np.arange(1, 73) and check if it works? I don't think intermittent fh is supported yet. Just a guess by looking at the image and the difference in the shapes in the error message. Once you confirm that it works by making fh continuous, I can see if a fix can be released shortly.

[tugra-alp]

When i try fh = np.arange(1, 73) , it worked ! However, as i mentioned above if i set fh with gap it doesn't work although it was working before.

[tugra-alp]

As a Temporary solution,
This problem can solve if we set data_kwargs when plotting.
Ex:
fh=np.arange(25,73) : 1 day gap and 2 days forecast horizon in each split (Hourly data).

exp.plot_model(estimator=model, data_kwargs = {'fh' : 72})

[ngupta23]

Fixed, it will be available in the next release.