run.py 17.7 KB
Newer Older
1
2
3
4
"""Brain Mapper Run File

Description:

5
6
7
8
9
    This file contains all the relevant functions for running BrainMapper.
    The network can be ran in one of these modes:
        - train
        - evaluate path
        - evaluate whole
10

11
    TODO: Might be worth adding some information on uncertaintiy estimation, later down the line
12

13
14
15
16
17
18
19
20
21
22
23
24
25
Usage:

    In order to run the network, in the terminal, the user needs to pass it relevant arguments:

        $ ./setup.sh
        $ source env/bin/activate
        $ python run.py --mode ...

    The arguments for mode are the following:

        mode=train # For training the model
        mode=evaluate-score # For evaluating the model score
        mode=evaluate-mapping # For evaluating the model mapping
26
27
        # For clearning the experiments and logs directories of the last experiment
        mode=clear-experiment
28
        mode=clear-all # For clearing all the files from the experiments and logs directories/
29
30
31

"""

32
import os
33
import shutil
34
35
import argparse
import logging
36
37
38
39
from settings import Settings

import torch
import torch.utils.data as data
Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
40
import numpy as np
41
42

from solver import Solver
43
from BrainMapperUNet import BrainMapperUNet3D, BrainMapperResUNet3D, BrainMapperResUNet3Dshallow, BrainMapperCompResUNet3D
44
from utils.data_utils import get_datasets, data_test_train_validation_split, update_shuffling_flag, create_folder
45
46
import utils.data_evaluation_utils as evaluations
from utils.data_logging_utils import LogWriter
47
48
49
50
51

# Set the default floating point tensor type to FloatTensor

torch.set_default_tensor_type(torch.FloatTensor)

52

53
54
55
def load_data(data_parameters):
    """Dataset Loader

56
    This function loads the training and validation datasets.
57
58
59
60
61
62

    Args:
        data_parameters (dict): Dictionary containing relevant information for the datafiles.

    Returns:
        train_data (dataset object): Pytorch map-style dataset object, mapping indices to training data samples.
63
        validation_data (dataset object): Pytorch map-style dataset object, mapping indices to testing data samples.
64
65
66

    """
    print("Data is loading...")
67
    train_data, validation_data = get_datasets(data_parameters)
68
69
    print("Data has loaded!")
    print("Training dataset size is {}".format(len(train_data)))
70
    print("Validation dataset size is {}".format(len(validation_data)))
71

72
    return train_data, validation_data
73

74

75
76
def train(data_parameters, training_parameters, network_parameters, misc_parameters):
    """Training Function
77

78
    This function trains a given model using the provided training data.
79
    Currently, the data loaded is set to have multiple sub-processes.
80
81
    A high enough number of workers assures that CPU computations are efficiently managed, i.e. that the bottleneck is indeed the neural network's forward and backward operations on the GPU (and not data generation)
    Loader memory is also pinned, to speed up data transfer from CPU to GPU  by using the page-locked memory.
82
    Train data is also re-shuffled at each training epoch.
83
84

    Args:
85
86
87
88
89
        data_parameters (dict): Dictionary containing relevant information for the datafiles.

        training_parameters(dict): Dictionary containing relevant hyperparameters for training the network.
        training_parameters = {
            'training_batch_size': 5
90
            'validation_batch_size: 5
91
92
93
94
95
96
97
98
99
100
101
102
103
            'use_pre_trained': False
            'pre_trained_path': 'pre_trained/path'
            'experiment_name': 'experiment_name'
            'learning_rate': 1e-4
            'optimizer_beta': (0.9, 0.999)
            'optimizer_epsilon': 1e-8
            'optimizer_weigth_decay': 1e-5
            'number_of_epochs': 10
            'loss_log_period': 50
            'learning_rate_scheduler_step_size': 3
            'learning_rate_scheduler_gamma': 1e-1
            'use_last_checkpoint': True
            'final_model_output_file': 'path/to/model'
104
        }
105

106
        network_parameters (dict): Contains information relevant parameters
107

108
109
110
111
112
113
114
115
        misc_parameters (dict): Dictionary of aditional hyperparameters
        misc_parameters = {
            'save_model_directory': 'directory_name'
            'model_name': 'BrainMapper'
            'logs_directory': 'log-directory'
            'device': 1
            'experiments_directory': 'experiments-directory'
        }
116
117
    """

118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
    def _train_runner(data_parameters, training_parameters, network_parameters, misc_parameters):
        """Wrapper for the training operation

        This function wraps the training operation for the network

        Args:
            data_parameters (dict): Dictionary containing relevant information for the datafiles.
            training_parameters(dict): Dictionary containing relevant hyperparameters for training the network.
            network_parameters (dict): Contains information relevant parameters
            misc_parameters (dict): Dictionary of aditional hyperparameters

        """
        train_data, validation_data = load_data(data_parameters)

        train_loader = data.DataLoader(
            dataset=train_data,
            batch_size=training_parameters['training_batch_size'],
            shuffle=True,
            num_workers=4,
            pin_memory=True
        )

        validation_loader = data.DataLoader(
            dataset=validation_data,
            batch_size=training_parameters['validation_batch_size'],
            shuffle=False,
            num_workers=4,
            pin_memory=True
        )

        if training_parameters['use_pre_trained']:
            BrainMapperModel = torch.load(
                training_parameters['pre_trained_path'])
        else:
152
            # BrainMapperModel = BrainMapperUNet3D(network_parameters)
153
154
155
            # BrainMapperModel = BrainMapperResUNet3D(network_parameters)
            # BrainMapperModel = BrainMapperResUNet3Dshallow(network_parameters)
            BrainMapperModel = BrainMapperCompResUNet3D(network_parameters)
156

157

158
159
        BrainMapperModel.reset_parameters()

160
161
        optimizer = torch.optim.Adam

162
163
164
165
        solver = Solver(model=BrainMapperModel,
                        device=misc_parameters['device'],
                        number_of_classes=network_parameters['number_of_classes'],
                        experiment_name=training_parameters['experiment_name'],
Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
166
                        optimizer=optimizer,
167
                        optimizer_arguments={'lr': training_parameters['learning_rate'],
168
169
170
171
                                             'betas': training_parameters['optimizer_beta'],
                                             'eps': training_parameters['optimizer_epsilon'],
                                             'weight_decay': training_parameters['optimizer_weigth_decay']
                                             },
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
                        model_name=misc_parameters['model_name'],
                        number_epochs=training_parameters['number_of_epochs'],
                        loss_log_period=training_parameters['loss_log_period'],
                        learning_rate_scheduler_step_size=training_parameters[
                            'learning_rate_scheduler_step_size'],
                        learning_rate_scheduler_gamma=training_parameters['learning_rate_scheduler_gamma'],
                        use_last_checkpoint=training_parameters['use_last_checkpoint'],
                        experiment_directory=misc_parameters['experiments_directory'],
                        logs_directory=misc_parameters['logs_directory'],
                        checkpoint_directory=misc_parameters['checkpoint_directory']
                        )

        validation_loss = solver.train(train_loader, validation_loader)

        model_output_path = os.path.join(
            misc_parameters['save_model_directory'], training_parameters['final_model_output_file'])

        create_folder(misc_parameters['save_model_directory'])

        BrainMapperModel.save(model_output_path)

        print("Final Model Saved in: {}".format(model_output_path))

195
        del train_data, validation_data, train_loader, validation_loader, BrainMapperModel, solver, optimizer
196
197
198
199
200
201
202
203
        torch.cuda.empty_cache()

        return validation_loss

    if data_parameters['k_fold'] is None:

        _ = _train_runner(data_parameters, training_parameters,
                          network_parameters, misc_parameters)
204

205
    else:
206
        print("Training initiated using K-fold Cross Validation!")
207
        k_fold_losses = []
208

209
        for k in range(data_parameters['k_fold']):
210

Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
211
            print("K-fold Number: {}".format(k+1))
212

213
            data_parameters['train_list'] = os.path.join(
214
                data_parameters['data_folder_name'], 'train' + str(k+1)+'.txt')
215
            data_parameters['validation_list'] = os.path.join(
216
217
218
                data_parameters['data_folder_name'], 'validation' + str(k+1)+'.txt')
            training_parameters['final_model_output_file'] = training_parameters['final_model_output_file'].replace(
                ".pth.tar", str(k+1)+".pth.tar")
219

220
            validation_loss = _train_runner(
221
                data_parameters, training_parameters, network_parameters, misc_parameters)
222

223
            k_fold_losses.append(validation_loss)
224

225
226
227
        for k in range(data_parameters['k_fold']):
            print("K-fold Number: {} Loss: {}".format(k+1, k_fold_losses[k]))
        print("K-fold Cross Validation Avearge Loss: {}".format(np.mean(k_fold_losses)))
228

Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
229

230
def evaluate_score(training_parameters, network_parameters, misc_parameters, evaluation_parameters):
231
232
233
234
235
236
237
238
239
240
241
    """Mapping Score Evaluator

    This function evaluates a given trained model by calculating the it's dice score prediction.

    Args:

        training_parameters(dict): Dictionary containing relevant hyperparameters for training the network.
        training_parameters = {
            'experiment_name': 'experiment_name'
        }

242
        network_parameters (dict): Contains information relevant parameters
243
244
245
        network_parameters= {
            'number_of_classes': 1
        }
246

247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
        misc_parameters (dict): Dictionary of aditional hyperparameters
        misc_parameters = {
            'logs_directory': 'log-directory'
            'device': 1
            'experiments_directory': 'experiments-directory'
        }

        evaluation_parameters (dict): Dictionary of parameters useful during evaluation.
        evaluation_parameters = {
            'trained_model_path': 'path/to/model'
            'data_directory': 'path/to/data'
            'targets_directory': 'path/to/targets'
            'data_list': 'path/to/datalist.txt/
            'orientation': 'coronal'
            'saved_predictions_directory': 'directory-of-saved-predictions'
        }
    """

265
    # TODO - NEED TO UPDATE THE DATA FUNCTIONS!
266

267
268
269
    logWriter = LogWriter(number_of_classes=network_parameters['number_of_classes'],
                          logs_directory=misc_parameters['logs_directory'],
                          experiment_name=training_parameters['experiment_name']
270
                          )
271

272
    prediction_output_path = os.path.join(misc_parameters['experiments_directory'],
273
274
275
276
                                          training_parameters['experiment_name'],
                                          evaluation_parameters['saved_predictions_directory']
                                          )

277
    _ = evaluations.evaluate_dice_score(trained_model_path=evaluation_parameters['trained_model_path'],
278
279
                                        number_of_classes=network_parameters['number_of_classes'],
                                        data_directory=evaluation_parameters['data_directory'],
Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
280
281
282
283
284
285
                                        targets_directory=evaluation_parameters['targets_directory'],
                                        data_list=evaluation_parameters['data_list'],
                                        orientation=evaluation_parameters['orientation'],
                                        prediction_output_path=prediction_output_path,
                                        device=misc_parameters['device'],
                                        LogWriter=logWriter
286
    )
287
288
289

    logWriter.close()

290

291
292
def evaluate_mapping(mapping_evaluation_parameters):
    """Mapping Evaluator
293

294
295
296
297
298
    This function passes through the network an input and generates the rsfMRI outputs.

    Args:
        mapping_evaluation_parameters (dict): Dictionary of parameters useful during mapping evaluation.
        mapping_evaluation_parameters = {
299
            'trained_model_path': 'path/to/model'
300
            'data_directory': 'path/to/data'
301
            'mapping_data_file': 'path/to/file'
302
303
304
305
306
307
308
309
            'data_list': 'path/to/datalist.txt/
            'prediction_output_path': 'directory-of-saved-predictions'
            'batch_size': 2
            'device': 0
            'exit_on_error': True
        }

    """
310
    trained_model_path = mapping_evaluation_parameters['trained_model_path']
311
    data_directory = mapping_evaluation_parameters['data_directory']
312
    mapping_data_file = mapping_evaluation_parameters['mapping_data_file']
313
314
    data_list = mapping_evaluation_parameters['data_list']
    prediction_output_path = mapping_evaluation_parameters['prediction_output_path']
315
    device = mapping_evaluation_parameters['device']
316
    exit_on_error = mapping_evaluation_parameters['exit_on_error']
317
318
319
    brain_mask_path = mapping_evaluation_parameters['brain_mask_path']
    mean_mask_path = mapping_evaluation_parameters['mean_mask_path']
    mean_reduction = mapping_evaluation_parameters['mean_reduction']
320

321
    evaluations.evaluate_mapping(trained_model_path,
Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
322
323
324
325
                                 data_directory,
                                 mapping_data_file,
                                 data_list,
                                 prediction_output_path,
326
327
328
                                 brain_mask_path,
                                 mean_mask_path,
                                 mean_reduction,
Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
329
330
331
                                 device=device,
                                 exit_on_error=exit_on_error)

332
333

def delete_files(folder):
334
    """ Clear Folder Contents
335

336
337
338
339
    Function which clears contents (like experiments or logs)

    Args:
        folder (str): Name of folders whose conents is to be deleted
340

341
    """
342

343
344
345
346
347
348
349
350
351
352
    for object_name in os.listdir(folder):
        file_path = os.path.join(folder, object_name)
        try:
            if os.path.isfile(file_path):
                os.unlink(file_path)
            elif os.path.isdir(file_path):
                shutil.rmtree(file_path)
        except Exception as exception:
            print(exception)

353
354

if __name__ == '__main__':
355
    parser = argparse.ArgumentParser()
356
357
358
359
    parser.add_argument('--mode', '-m', required=True,
                        help='run mode, valid values are train or evaluate')
    parser.add_argument('--settings_path', '-sp', required=False,
                        help='optional argument, set path to settings_evaluation.ini')
360
361
362
363
364
365
366
367
368
369

    arguments = parser.parse_args()

    settings = Settings('settings.ini')
    data_parameters = settings['DATA']
    training_parameters = settings['TRAINING']
    network_parameters = settings['NETWORK']
    misc_parameters = settings['MISC']
    evaluation_parameters = settings['EVALUATION']

370
371
372
373
    # Here we shuffle the data!

    if data_parameters['data_split_flag'] == True:
        if data_parameters['use_data_file'] == True:
374
375
376
            data_test_train_validation_split(data_parameters['data_folder_name'],
                                             data_parameters['test_percentage'],
                                             data_parameters['subject_number'],
377
                                             data_directory=data_parameters['data_directory'],
378
379
380
                                             data_file=data_parameters['data_file'],
                                             K_fold=data_parameters['k_fold']
                                             )
381
        else:
382
383
384
385
386
387
            data_test_train_validation_split(data_parameters['data_folder_name'],
                                             data_parameters['test_percentage'],
                                             data_parameters['subject_number'],
                                             data_directory=data_parameters['data_directory'],
                                             K_fold=data_parameters['k_fold']
                                             )
388
        update_shuffling_flag('settings.ini')
389

390
391
    if arguments.mode == 'train':
        train(data_parameters, training_parameters,
392
              network_parameters, misc_parameters)
393
394

    # NOTE: THE EVAL FUNCTIONS HAVE NOT YET BEEN DEBUGGED (16/04/20)
395
    # NOTE: THE EVAL-MAPPING FUNCTION HAS BEEN DEBUGGED (28/04/20)
396
397
398

    elif arguments.mode == 'evaluate-score':
        evaluate_score(training_parameters,
399
                       network_parameters, misc_parameters, evaluation_parameters)
400
401
402
403
    elif arguments.mode == 'evaluate-mapping':
        logging.basicConfig(filename='evaluate-mapping-error.log')
        if arguments.settings_path is not None:
            settings_evaluation = Settings(arguments.settings_path)
404
        else:
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
            settings_evaluation = Settings('settings_evaluation.ini')
        mapping_evaluation_parameters = settings_evaluation['MAPPING']
        evaluate_mapping(mapping_evaluation_parameters)
    elif arguments.mode == 'clear-experiments':
        shutil.rmtree(os.path.join(
            misc_parameters['experiments_directory'], training_parameters['experiment_name']))
        shutil.rmtree(os.path.join(
            misc_parameters['logs_directory'], training_parameters['experiment_name']))
        print('Cleared the current experiments and logs directory successfully!')
    elif arguments.mode == 'clear-everything':
        delete_files(misc_parameters['experiments_directory'])
        delete_files(misc_parameters['logs_directory'])
        print('Cleared the current experiments and logs directory successfully!')
    else:
        raise ValueError(
            'Invalid mode value! Only supports: train, evaluate-score, evaluate-mapping, clear-experiments and clear-everything')