This is a package for extracting facial expressions from SLEAP analyses.
Things that it does:
- iterate (repeatedly) over each mouse and each week (each mouse and each experiment)
- get project files (experimental) structure
- initialize an iterator over project structure
- get daq data from CSV file
- read CSV files
- save each column from CSV file
- Note: CSV columns are of differing lengths
- get “beh_metadata” from json metadata
- read JSON file
- grab the values for key “beh_metadata”
- get the values of sub key “trialArray”
- get the values of subkey “ITIArray”
- get video metadata from *.mp4 file (with ffmpeg.probe)
- read in the *.mp4 metadata
- select the correct video stream
- get the average frames per second
- get SLEAP data from *.h5 file
- open h5 file
- get transposed values of key “tracks” (tracking_locations)
- fill missing locations (linear regress. fit)
- get transposed values of key “edge_inds”
- get values of key “edge_names”
- get transposed values of “instance_scores”
- get transposed values of “point_scores”
- get values of “track_occupancy”
- get transposed values of “tracking_scores”
- get decoded values of “node_names” (make sure theres no encoding issues)
- deconstruct SLEAP points into x and y points (across all frames)
- iterate over each node
- breakup the 4D array “tracks” into 1D array for x and y values respectively
- Note: [frame, node, x/y, color] for greyscale the color dimension is 1D (i.e. essentially the 4D array is 3D because the color dimension is constant)
- iterate over each frame
- assign mouse, week, frame #, and timestamp (using average frames per second)
- Split data into individual trials by trial type using the Speaker and LED data from the CSV daq data
- initialize trial iterators for the consistently documented points from the daq CSV
- iterate over each trial in “trialArray”
- get the index of 10sec before and 13sec after trial start
- for each feature, grab the start and end indices
- store data from each trial in a pd.dataframe
- concatenate all pd.dataframes together for each video
- concatenate the pd.dataframes from each video together for each mouse (base expr split)
- Prepare the data
- (opt.) mean center across all points for a single trial
- mean center across all trials for a single experiment
- mean center across all experiments for a single mouse
- mean center across all mice
- (opt.) z-score mean centered data
- Analyze the data
- Perform 2D and 3D PCAs on all data (raw, centered, by trial, by week, by mouse, etc…)
- apply gaussian kernel to PCA outputs
- Save the data
- write everything to HDF5 file(s)
Table of Contents
pip install sleapyfacessleapyfaces is distributed under the terms of the MIT license.