| |
| |
| """ |
| Created on Wed Jan 29 17:36:42 2025 |
| |
| |
| About: |
| ======= |
| initialize dataloaders and pytorch datasets |
| |
| """ |
| import pickle |
| import torch |
| from torch.utils.data import DataLoader |
| import numpy as np |
|
|
| from dloaders.CountsDset import CountsDset |
| from dloaders.CountsDset import jax_collator as collator |
| from dloaders.init_dataloader import init_dataloader |
| from dloaders.init_time_array import init_time_array |
|
|
| |
| def init_counts_dset( args, |
| task, |
| training_argparse=None, |
| include_dataloader=True ): |
| """ |
| initialize the dataloaders |
| """ |
| |
| |
| |
| if task in ['train', |
| 'resume_train']: |
| only_test = False |
| t_per_sample = args.pred_config['times_from'] == 't_per_sample' |
| t_array_for_all_samples = init_time_array(args) |
| subs_only = (args.pred_config['indel_model_type'] is None) |
| |
| if args.pred_config['subst_model_type'].lower() == 'hky85': |
| emission_alphabet_size = 4 |
| else: |
| emission_alphabet_size = 20 |
| |
| |
| |
| |
| |
| elif task in ['eval']: |
| only_test = True |
| t_per_sample = training_argparse.pred_config['times_from'] == 't_per_sample' |
| t_array_for_all_samples = init_time_array(training_argparse) |
| subs_only = (training_argparse.pred_config['indel_model_type'] is None) |
| |
| if training_argparse.pred_config['subst_model_type'].lower() == 'hky85': |
| emission_alphabet_size = 4 |
| else: |
| emission_alphabet_size = 20 |
| |
| |
| |
| |
| |
| |
| assert type(args.test_dset_splits) == list |
|
|
| print('Test dset:') |
| for s in args.test_dset_splits: |
| print(s) |
| print() |
|
|
| test_dset = CountsDset( data_dir = args.data_dir, |
| split_prefixes = args.test_dset_splits, |
| emission_alphabet_size = emission_alphabet_size, |
| t_per_sample = t_per_sample, |
| subs_only = subs_only, |
| toss_alignments_longer_than = args.toss_alignments_longer_than) |
| out = {'test_dset': test_dset, |
| 't_array_for_all_samples': t_array_for_all_samples} |
|
|
|
|
| |
| if not only_test: |
| assert type(args.train_dset_splits) == list |
|
|
| print('Training dset:') |
| for s in args.train_dset_splits: |
| print(s) |
| print() |
| |
| training_dset = CountsDset( data_dir = args.data_dir, |
| split_prefixes = args.train_dset_splits, |
| emission_alphabet_size = emission_alphabet_size, |
| t_per_sample = t_per_sample, |
| subs_only = subs_only, |
| toss_alignments_longer_than = args.toss_alignments_longer_than) |
| out['training_dset'] = training_dset |
| |
| |
| |
| |
| |
| if include_dataloader: |
| test_dl = init_dataloader(args = args, |
| pytorch_custom_dset = test_dset, |
| shuffle = False, |
| collate_fn = collator) |
| out['test_dl'] = test_dl |
| |
| if not only_test: |
| training_dl = init_dataloader(args = args, |
| pytorch_custom_dset = training_dset, |
| shuffle = True, |
| collate_fn = collator) |
| out['training_dl'] = training_dl |
| |
| return out |
|
|
