Lorenzo Stella

@MariPlaza I am not familiar with SHAP or LIME, maybe you could explain what you have tried with those, or what blocks you from using them in the context of...

> My question to you is if the TFT implementation on gluon-ts has (or will have) the interpretable insights as well as the PyTorch implementation has @Gandor26 may help answering...

@MInYangP-OU [this test module](https://github.com/awslabs/gluonts/blob/4c8eefddf014e52f6889419f8783c64e0671ea1b/test/torch/model/test_multivariate_estimators.py#L26) imports the estimator exactly as you say, and works fine. Are you sure you’re on the latest dev branch?

@NeoWang9999 was the data sampled at 3-second frequency, and some time windows happen to be missing? If that's the case, you probably want to just `resample` your DataFrame to have...

For example using https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.asfreq.html

@sunnyccshine I suspect this is not a bug. Does the following work, or raise a similar issue? ```python import urllib urllib.request.urlopen("https://www.google.com").read() ```

Should predictors not allow being called without `n`? (That is, should `n` not have a default value?)

This line seems too complicated: ```python median_forecast = np.array([np.median(sample) for sample in forecast]) ``` If `forecast` is an array-like object, should one not be able to do something like ```python...

Another note: `predict_many` may be the preferred way of implementing some of the predictors (e.g. NN-based ones that prefer to process multiple entries in one batch). What is the idea...

> I wanted to use the same code for both examples. The axis-parameter doesn't work where you only have a single value per time-stamp, I think. Sure. One problem I...