abuzarmahmood / blech_clust Goto Github PK

Can be turned into a confluence page or something else public

yaml allows commenting, which can provide better interactivity with users

Can use this to simply tee print statements to a log file per animal
Maybe can even write to specific sections of a log file

https://stackoverflow.com/a/616672

This will help with testing and checking for coverage as well as understanding of flow in scripts
Will also allow for data validation.

This will allow checking for cutoff in blech_clust

Will include:

1. blech_clust.py : Username

this will include further processing lie #31 , but NOT analyses.
Analyses can be included in blech_clust (e.g. spectrogram, coherence etc), but separately

Set parameters for thresholding a channel

If number of waveforms detected greater than threshold, allow options:

1. Spit out sheet marking which electrodes are significant and with how many total, hits, and misses
2. Only output channels (plots and data) with significant waveforms

Would be great to have a combined one that works for both spike sorting and EMG...but separate ones would be a start

pip install -r pip_requirements.txt # Install pip requirements (not covered by conda)
should be changed to
pip install -r pip_requirements_base.txt # Install pip requirements (not covered by conda)

This can be merged with #20 to check both logs, and outputs.

Can have a submodule where outputs for every step are listed. Module is called at top of every file...if previous step has never been logged, or cannot detect outputs of previous step, then it just prevents execution of the script

Would be good to have this for the main spike-sorting pipeline

update algorithm to not do all-to-all comparisons

1. One file per channel
2. One file per signal type
3. Dataset with EMG

This could be a txt file written to the plots folder for that channel

This will remove confusion between using index and actual number
And will automatically store it as metadata

Currently the downsampling is set to 100, which works for 40 minute long recordings but not long. The downsampling should be updated to reflect the recording length to create realistic plots.
Example images of usual recording session length + long length plots below:

This way, given the same .info and sorted .csv, and the same version of blech_clust, one will get the same output

1. Perform ICA and ask user to select noise channels
2. Remove noise channels and reconstruct LFP from remaining ICA channels
3. Save appropriate figures (plots of pre ICA, ICA, and post-cleaning channels) and metadata (which channels were removed and ICA params)

Add it as a separate quality control measure
A log can be added to the QA folder showing which channels were chopped off
Can be done after blech_common_average_reference
This might be something so important that blech_process confirms that the user is aware of this before continuing

That is, when indexing emg_data.npy, we are assuming the order of emg_data follows the CAR groups in the electrode layout file.
While this is likely true, this should be made explicit somewhere (maybe save emg_data.npy as an xarray)

Along with other options...just add templates as a single option.
Suggestion:
Have a templates directory (in blech_clust dir) where people can add their templates

https://docs.github.com/en/repositories/releasing-projects-on-github/managing-releases-in-a-repository

if it's upper case, it throws an error

Includes:

1. Unit similarity check
2. Unit drift check (ANOVA for spike count over time) #16

Ignore ones below...not sure if they can be counted as quality assurance
3. Population dynamicity check (relative to other recorded populations)
4. Number + Fraction of taste responsive + discriminative + palatability encoding neurons

Currently, this is conditioned on knowledge of whether/when the headstage fell off, and generated spike trains
But once headstage fall-off is calculated, this can be done generally without reference to spike-trians

For example, instead of asking the user to provide the index of the tastant delivery input, pull that directly from the data.
ALSO - if an info file already exists, find it and skip running the code entirely.

All features used in clustering
This may be useful in determining if slightly varying waveforms are due to drift, noise, or multiple neurons
E.g. if amplitude is changing over time, very likely to be same neuron; as opposed to jumping between 2 or more amplitudes over time

refer to #48

Since gaping has a temporal component, the classifier can leverage this information.

Maybe an hmm on the spectrum of filtered EMG (not spectrum of envelope)

Or provide link to downloading one to help with testing and learning to use

This is to remove dead/noisy channels or at the least process them separately.
Number of electrodes can be determined with KMeans and AIC or BIC.