Refactor testing module

[timonmerk]

Many thanks for your first contribution @SamedVossberg!

Currently branch #323 was merged into main. Due to many changes in that branch, some points should be still adressed for the lsl stream generation and testing.

In general I would want to refactor the nm_generator.py module s.t. it's split into three different modules: 1. MNE-LSL Player, 2. MNE-LSL Stream and 3. Offline generator.

Tests:

• Test MNE-LSL Stream generation after player was initiated. Check:
  1. if stream is available
  2. if stream sfreq, channel names and types match the ones specified
  3. if the pulled data is equivallent to the one retrieved from the example data

Adapt then the example notebook. Additional, but for now optional, add plain Open-BCI ipynb example.

I will implement some of those changes in a new branch and notify you about remaining points.

Other points:

• Currently the pynput is located in generator module. I would pull this functionality out from the nm_*.py files. I think lateron it would make sense to create an nm_application file where the keyboard "interrput" is handled.
• In LSLStream, the example_stream get's currently always set to "example_stream", if the stream_name parameter is None, then check which streams are available and intialize the first one.
• The plot_7_lsl_example.ipynb can in fact be a .py file, sphinx-gallery can render those into ipynb. Markdown cells are rendered out from comments.

[toni-neurosc]

Totally agree on the nm_generator split, the classes inside are very different among them. I would also want to make a diagram of the current Stream class hierarchy because there is some complexity creeping up that we need to keep in check.

For example, the hierarchy NMStream -> _GenericStream -> Stream is really weird right now. _GenericStream docstring says This class can be inhereted for different types of offline streams but really it's handling everything since it contains logic to process LSL streams and also inherits form NMStream which also contains online stream logic.

I also think that raw_data_generator should be just a method in OfflineStream, which does not exist currently but I think it should.

Additionally, _GenericStream lacks a constructor since all the initialization is already handled in NMStream, which I think it proves NMStream holds too much responsability for a supposedly abstract base class.

Finally, I think we should distinguish between LSL streams (which are receiving over-the-network, both the actual LSLstream and the mock stream) and the NMStream, which is really more of a data handler and processor... it's doing a variety of things, except maybe stream data. Things it does:

• holding a data generator that produces batches out of the data source
• handling settings, checking their validity and passing them around
• generating timestamps
• generating log info about the data processing
• initializing threads for multiprocessing
• holding the actual Data processor which itself holds the pre-normalization, feature calculation and post-normalization logic.
• receiving results from Data processor and writing them out to files

So, this is A LOT. I think we should figure out a way to define clear boundaries between these tasks, and either redefine our inheritance hierarchy or separate into components that we can combine at runtime depending on settings. Or perhaps both.

[SamedVossberg]

Yes I think that would make totally sense @toni-neurosc and @timonmerk.

As I am still trying to get a complete overview it's quite hard for me to do the refactoring and finding a perfect solution. Unwinding the streams and hierarchies would make absolute sense though @toni-neurosc and would make the code base easier readable and understandable.

@timonmerk I just pushed updated testing which checks for stream sfreq, channel names and types matching the ones specified. I also changed the plot_7_lsl_example.ipynb file to a .py in the format of the other examples.

I will continue with the Tests:

• [ ] Check if stream is available (this happens implicitly already but I will write a test specifically for this)
• [ ] comparing pulled data with example data

And I will address the point:

• [ ] Stream name handling (currently "example_stream")

[timonmerk]

Can we close this @SamedVossberg?

[SamedVossberg]

Yes think this is done and can be closed after the changes in #326