Explore modules ¶

All functions listed in this page can be used to do a variety of operations. We have separated the functions into multiple modules.

Standard I/O routines ¶

`hdf5read` (fname[, key])	Read input raw DAS file.
`load_model` (fname, model)	Load saved model's parameter dictionary to initialized model.
`save_data` (data, dir_dst, fname)	Save raw data region as JPG image.
`load_image` (data[, rgb, to_numpy, squeeze])	Load input data as tensor image.
`load_bulk` (dname, size[, rgb, to_numpy, labeled])	Load multiple images from directory.

DAS data loader ¶

`set_from_file` (filename[, img_size, stride, ...])	Create set of images out of single DAS data file.
`mean_shift` (data[, loops])	Shit all values from input 2D array to match mean value of 0.5.
`set_creation` (datapath[, img_size, stride, ...])	Create PyTorch tensor training set of single-channel data images.
`prepare_loader` (datapath, img_size, stride, ...)	Create custom data loader with unlabeled, single-channel, data images directly extracted from raw data files.

Data Distribution ¶

`gauss_single` (x, a, b, c)
`gauss_double` (x, a, b, c, d, e, f)
`distfit` (ax, data, ngauss[, amp, mu, sig, ...])	Fit gaussian to 2D patch distribution.

Evaluation ¶

`decode_plot` (model, datapath[, img_size, ...])	Show original and decoded images for 5 random images.
`embedding_plot` (model, datapath[, img_size, ...])	Display latent representation in 2D space.
`epoch_recon` (models, datapath[, img_size, ...])	Display image reconstruction accuracy across epochs.
`label_2d_latent` (model, data_loader[, embeddings])	Scatter plot of 2D latent space with label-based color schemme.
`success_rate` (model, target, img_size, ...[, ...])	Number of reconstructed images for which 70% (or the percentage defined by the success_threshold argument) of the pixels are 90% similar.

Figures ¶

`plot_prob_map` (data, prob[, xmin, xmax, ...])	Comparative plots between raw data and corresponding probability map.
`plot_map_stage` (data, NeuralNet, loss_hist, ...)	Plot stage of probability map.

Mapping ¶

`minute_prob` (data_path, model_file, depth[, ...])	Get surface wave probabilities for every consecutive square regions.
`minute_prob_test` (hdf5_file, model_file, depth)	Get surface wave probabilities for every consecutive square regions.
`extract_prob_map` (data, model[, img_size, ...])	Calculate probability map using fed raw data and training model.

Cross-correlation ¶

`avg_fft` (data[, fs])
`xcorr_freq` (data, xcorr[, lag_range, threshold])

Phase-weighted stack ¶

`phase_weight_stack` (fpath, idx[, weight, path])
`ts_weighting` (fname, data, map_path)

Single ¶

suplearn_simple (model, criterion, optimizer, ...)

Simple, non-optimized, supervised training with validation step performed at regular intervals during batch iteration for single node, single processor execution.