[ROADMAP] Sprint3

[lkorczowski]

SPRINT BACKLOG

in chronological order

• [x] extensive pytest (each breaking point in RASR and related functions), see #24
• [x] REFACTORING: remove completely the 2D compatibility > only 3D matrices will be managed. removing related parameters (window_len, window_overlap, etc.) and epoching steps. See #5
• [x] passing the parameters from RASR to _fit_eeg_distribution() and checking valid keys. See https://github.com/bertrandlalo/timeflux_rasr/issues/15#issuecomment-585231931
• [ ] more documentation (insure accurate explanation of parameters, add instruction to how epoch the data before feeding to RASR)
• [x] blending (with parameter for option), see https://github.com/bertrandlalo/timeflux_rasr/issues/15#issuecomment-584583649
• [ ] detailed report for python rasr behaviour for different parameters (with comparison with matlab using default parameters) (e.g. screenshot, description, analysis of the outliers, etc.)
• [ ] Visually validate on a new dataset

PRODUCT BACKLOG (out of scope)

• IIR filter for mixing matrix and reconstruction matrix estimation #10
• solve the pymanopt converge of non_linear_eigh(), see #14

[mesca]

It is very important that the library can be used by non-specialist users. That means:

• Clearly document the input parameters
• Automatically infer secondary parameter values when possible
• Provide sensible defaults, with the number of required parameters kept to the minimum

[mesca]

Tests must be extensive, and must address every possible breaking point in the code. It is usually a bad practice to do them at the end. Have they been done on the fly, some issues raised in #17 would have been detected earlier.

[mesca]

I would add:

• [ ] Visually validate on a new dataset

We will acquire the data along with a synchronized video feed, and annotate classic artifacts ourselves.

[lkorczowski]

It is very important that the library can be used by non-specialist users. That means:

* Clearly document the input parameters

* Automatically infer secondary parameter values when possible

* Provide sensible defaults, with the number of required parameters kept to the minimum

• all inputs params are documented. The only case is 3 parameters that are passed by default to _fit_eeg_distribution and should not be changed by the user.
• only srate is required, all the others parameters are based on default. Based on my observations in the report, I can tune better the parameters that matlab.

[lkorczowski]

Tests must be extensive, and must address every possible breaking point in the code. It is usually a bad practice to do them at the end. Have they been done on the fly, some issues raised in #17 would have been detected earlier.

I understand, this procedure is new for me but I'm learning. I'll work on that until your expectations are met.

[lkorczowski]

In #21, @bertrandlalo mentioned that passing parameters from RASR to _fit_eeg_distribution() may use **kwargs. However, I couldn't find the best pratice to avoid copypasta of docstrings and avoid the initialization of the parameters in RASR().

I'm adding this problem in the backlog. I don't think it is critical right now. I simply can't figure out myself what is the best practice here.

[lkorczowski]

[ ] Visually validate on a new dataset

Added to backlog

[lkorczowski]

following https://github.com/bertrandlalo/timeflux_rasr/issues/5#issuecomment-584352525 , I added the code refactoring to remove 2D matrix compatibility.

[lkorczowski]

following https://github.com/bertrandlalo/timeflux_rasr/issues/5#issuecomment-584352525 (removal of 2D compatibility) , I propose to develop a specific node for blending outside of RASR(). Rational discussed in https://github.com/bertrandlalo/timeflux_rasr/issues/4#issuecomment-584582494

[lkorczowski]

Following https://github.com/bertrandlalo/timeflux_rasr/pull/21#issuecomment-585227534 , I am adding the task to pass the parameters from RASR to _fit_eeg_distribution using **kwargs and:

• document only useful parameters for the user in RASR()
• only take valid parameters inside _fit_eeg_distribution() as **kwargs could pass any parameters from RASR that aren't defined.

[lkorczowski]

Sprint3 starts now.

[lkorczowski]

FYI 1bf6683 added a feature for testing and passing **kwargs arguments properly while still checking for invalids key parameters. feature done

[lkorczowski]

@mesca could you send me the new dataset ? Thanks

[lkorczowski]

@mesca in news on the #27 review ?

[lkorczowski]

27 and #30 fixed most of the points

34 tested on new dataset