-
Notifications
You must be signed in to change notification settings - Fork 15
Data Format and Configuration Parameters
YASS takes as input 2 files: the raw data file and a 2D geometry file. The raw input (generally 2-byte integer format, i.e. 'int16' in python) has column-structure:
Channel_1_Time_1, Channel_2_Time_1,... Channel_C_Time_t, Channel_C+1_Time_t...
The geometry file is an ASCII txt file containing the X and Y coordinates of each electrode (in the same order as the raw binary file) with no comma separation:
-7.350000000000000000e+02 4.500000000000000000e+02
-6.750000000000000000e+02 4.500000000000000000e+02
-6.150000000000000000e+02 4.500000000000000000e+02
...
The config.yaml file contains several sections and it can be edited either using a text editor (recommended) or using the YASS GUI (limited changes only).
data:
`root_folder: [insert folder location of install]`
`# recordings filename (must be a binary file), details about the recordings`
`# are specified in the recordings section`
`recordings: [insert binary filename]`
`# channel geometry filename , supports txt (one x, y pair per line,`
`# separated by spaces) or a npy file with shape (n_channels, 2),`
`# where every row contains a x, y pair. see yass.geometry.parse for details`
`geometry: [insert name of geometry text file]`
recordings:
`# precision of the recording – must be a valid numpy dtype`
`dtype: int16 [only int16 is supported currently]`
`# recording rate (in Hz)`
`sampling_rate: [sampling rate] `
`# number of channels`
`n_channels: [number of channels]`
# channels spatial radius to consider them neighbors, see
# yass.geometry.find_channel_neighbors for details
spatial_radius: [distance between channels + 10]
# temporal length of waveforms in ms. It must capture
# the full shape of waveforms but longer means slower
spike_size_ms: 3 [3ms is default]
`# chunks to run clustering on (in seconds)`
`# leave blank to run clustering on entire dataset [not recommended]`
`clustering_chunk: [0, 300] # default clustering is run on first 5mins of data`
`# chunks to run final deconv on (in seconds)`
`# leave blank to run it on full`
`final_deconv_chunk: # default leave blank`
resources:
`# CPU multi-processing flag: 1 = use multiple cores`
`multi_processing: 1`
`# Number of CPU cores to use; recommended to set to # of physical cores available on CPU`
`n_processors: 16`
`# Length of processing chunks; if memory issues arise, decrease value`
`n_sec_chunk: 10`
`# number of GPUs to use [multi-gpu options being currently implemented]`
`n_gpu_processors: 1`
`# n_sec_chunk for gpu detection; if memory issues arise, decrease value`
`n_sec_chunk_gpu_detect: 0.5`
`# n_sec_chunk for gpu deconvolution; if memory issues arise, decrease value`
`n_sec_chunk_gpu_deconv: 5`
Neural Network training:
detect:
`# model name, it can be an absolute path to a model `
`# (e.g. /path/to/my/model.ckpt) or a name of one of models included in yass `
`# (e.g. detect.pt). Check /src/yass/assets/nn_models for the list of available`
`# models.`
`filename: detect.pt `
`n_filters: [16, 8, 8] `
denoise:
`# model name, same rule as detect filename`
`filename: denoise.pt`
`n_filters: [16, 8, 4]`
`filter_sizes : [5, 11, 21]`
training:
`# input spike train file name. It must be a numpy file of an arroay of n x 2.`
`# n is the number of spikes, the first column is the spike times `
`# (center of spikes), the second column is the unit id.`
`# if you want to run yass without nn to make spike train, leave it empty`
`input_spike_train_filname:`
`# temporal length of spike to go into nn`
`# leave it empty if not training`
`spike_size_ms:`