Fix Chronos notebook

[abdulfatir]

Hi!

Thank you for trying Chronos out! It looks like the predictions from Chronos are misaligned when you compute the metrics. This PR fixes this issue. Furthermore, there were multiple inefficiencies in the inference loop for Chronos:

• Pandas DataFrame was being filtered in a loop.
• Predictions were being generated one time series at a time. In this notebook I have used a batch size of 12, which should fit on a typical consumer GPU, even for the large model.

Here are the final numbers after these fixes:

	metric	MLP	NBEATS	Chronos-Tiny	Chronos-Large
0	mae	566.162569	551.727285	582.312115	563.345977
1	smape	0.064215	0.062675	0.065570	0.063084
2	Time	23.294084	49.143765	19.080808	138.417686

In case you're wondering what was the bug in the previous version: the index in forecast_df was incorrect after sorting. You can fix the previous version by simply changing:

forecast_df = forecast_df.drop(columns=['SortKey'])

to

forecast_df = forecast_df.drop(columns=['SortKey']).reset_index(drop=True)

Based on these results, I kindly ask you to modify your code and blog post. Furthermore, I would also request you to post an update on your social media profiles. Please let me know if you need any help from our side running Chronos! Thank you again!