posdc

• Bayes decoding of animal's position in an linear environment
• Depending on the customized package neuralib, Documentation
• Core function place_bayes

---

How to set up in a local machine?

1. Clone or download the project locally

2. Create conda environment

conda create -n posdc python~=3.10.0 -y
conda activate posdc

3. Install required packages

• First cd to the directory with pyproject.toml

pip install -e .

---

How to run in terminal?

See help

posdoc -h

Bayes decoding of animal's position in an linear environment

options:
  -h, --help            show this help message and exit
  -F FILE, --file FILE  hdf data file path
  --deconv              use deconv activity, otherwise, df_f

Decoding Option:
  --length VALUE        trial length in cm
  --spatial-bin VALUE   spatial bin size in cm
  --temporal-bin VALUE  temporal bin size in second, CURRENTLY NOT USE, DIRECTLY USE THE SAMPLING RATE OF THE DF/F
  --run                 whether select only the running epoch
  --no-position-ds      not interpolate position to neural activity shape, then interpolate neural activity to position
  --invalid-cache       invalid all the cache, and recompute

Train/Test Option:
  --train NAME           train the decoder in which behavioral session
  --cv CROSS_VALIDATION  nfold for model cross validation
  --no-shuffle           whether without shuffle the data for non-repeated cv
  --repeats N_REPEATS    run as repeat kfold cv, make number of results to `n_cv * n_repeats`
  --train-fraction TRAIN_FRACTION 
                         fraction of data for train set if `random_split` in cv, the rest will be utilized in test set.(require for --train=random-split)

Random Option:
  --random-neuron NUMBER number of random neurons
  --seed VALUE           seed for random number generator

Rastermap sorting Options:
  --rastermap            sort activity heatmap using rastermap
  --rastermap-bin VALUE  bin size for number of total neurons

Plotting Options:
  --ignore-foreach       whether to ignore foreach cv plot, and only plot the summary result
  -O PATH, --output PATH output figures to a directory

Examples

Example of 5-fold cross validation on light session using 500 neurons, and test on the rest

posdoc --file [HDF_FILE] --run --train light-cv --cv 5 --random-neuron 500

Example of 10 repeated 5-fold cross validation on light session using 500 neurons, and test on the rest

posdoc --file [HDF_FILE] --run --train light-cv --cv 5 --random-neuron 500 --repeats 10

Remarks

• --invalid-cache: To invalid local cache when changing Decoding Option
• --rastermap: show sorted heatmap by rastermap
• --ignore-foreach: ignore foreach cv plots, and only plot summary
• --output: specify output dir for saving the figures
• See other TRAIN_OPTIONS if needed: check PositionDecodeOptions.train_test_split()

---

Output Files

Figures

• decode_foreach_cv*.pdf: Foreach cross validation results
• decode_summary.pdf: Summary of cross validation in light versus dark

Parquet file

• Example output from 5-fold non-repeated cross-validation

n_cv (i64)	session (str)	n_trials (i64)	decode_err (f64)	trial_indices (list[i64])	binned_err (list[f64])
0	light	10	3.635864	[5, 7, … 47]	[22.904429, 18.559892, … 17.09…]
0	dark	81	8.658199	[51, 52, … 126]	[30.071842, 23.799756, … 30.65…]
1	light	10	5.738808	[0, 1, … 47]	[22.249407, 12.599134, … 17.27…]
1	dark	81	9.701644	[51, 52, … 126]	[27.65329, 17.79454, … 27.6105…]
2	light	9	4.084523	[3, 9, … 47]	[16.561472, 19.38851, … 19.172…]
2	dark	81	9.151024	[51, 52, … 126]	[27.966511, 20.448634, … 28.03…]
3	light	9	3.85323	[4, 6, … 47]	[38.950859, 13.562917, … 27.73…]
3	dark	81	8.98436	[51, 52, … 126]	[28.859011, 19.658932, … 27.56…]
4	light	9	4.877178	[2, 24, … 47]	[26.855452, 5.773164, … 16.729…]
4	dark	81	9.153177	[51, 52, … 126]	[27.265134, 19.030994, … 28.35…]

• n_cv: Number of the cross validation
• session: Which session of the test dataset
• n_trials: Number of trials of the test dataset
• decode_err: Mean/median decode error for the session
• trial_indices: trial indices for test set
• binned_err: Binned decoding error, length equal to number of bins

Cache

• [FILE]_position_cache.npz: Position information
• [FILE]_ratemap_cache.npy: RateMap (position binned activity)

---

TODO/BUGFIX

• If get anRuntimeError: pr approach the infinite, usually happened in --deconv, try to specify less neuron using --random-neuron
• Test for the --seed usage
• Statistic for cv summary if needed
• Other behavioral session support

Annotations

Numpy Array

Numpy array are annotated with following expression Array[dtype, shape]. For example, Array[int, [3, 3]] is a 3x3 int array or (3, 3) in short.