Sagar Mishra

It seems like there's no perfect editor to render `.rst`. GitHub and [online editors](https://livesphinx.herokuapp.com/) don't render code blocks. PyCharm seems to be rendering them fine though. References also work correctly...

Following is a small code-snippet to showcase how "states are preserved" / "contract is maintained": ```python import sktime.classification.deep_learning as dl from sktime.base import load, load_dl import numpy as np from...

Thanks for the review! My only problem with a separate `load_dl()` was that we'd have to modify tests to call `load_dl()` instead of `load()` for relevant cases, not to mention...

If we add new estimators that require a different logic than already provided in base classes(having custom objects for example), then the author would have to expand the class method...

Perhaps we don't understand each other. The additional code is due to the fact that non-DL and DL estimators are being stored in different ways. For the sake of discussion,...

> question: is there a way to avoid the creation of the temp folder? This may upset virus scanners etc. Wouldn't this be a problem with `keras`'s folder creation too?...

It was based on the implementation of #3336, not branched from it - sorry I didn't explicitly mention that. I had removed the tests here since they are being worked...

I could be wrong but the reason why #3336 passes and this doesn't might be because all (non-DL) estimators comply with expected test specifications. In contrast, DL estimators probably don't...

I see we've been discussing two different tests here. Let's break the conversation down: There are two 'problematic' test functions that we were initially skipping - `test_persistence_via_pickle()` & `test_fit_idempotent()`. You...

So here's what I propose - we do not change `test_persistence_via_pickle`, just update the doc to explicitly mention that we are checking for in-memory serialization. We add a `test_file_persistence_via_pickle` where...