torchsig.utils.file_handlers.npy¶
torchsig.utils.file_handlers.npy¶
File-handler that exposes a directory of standard NumPy *.npy files as a
TorchSig Signal dataset.
A TorchSig dataset is described by three co-located artefacts:
* One or more ``*.npy`` files - each file stores a 1-D NumPy array of
complex samples.
A ``metadata.csv`` file - one row per global waveform index, containing
index,label,modcod,sample_rate.An ``info.json`` file - a tiny JSON document that must contain at least
{\"size\": <int>}and defines the advertised length of the dataset.
The heavy binary payload lives in the *.npy files; the human-readable
description lives in the CSV. This separation keeps loading fast (memory-mapped
NumPy) while allowing easy inspection and editing of labels, modulation codes,
etc.
Classes
Read a directory that contains |