run.py 19.3 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
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
152
153
    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:
            BrainMapperModel = BrainMapperUNet3D(network_parameters)

154
155
        BrainMapperModel.reset_parameters()

156
157
        optimizer = torch.optim.Adam

158
159
160
161
        solver = Solver(model=BrainMapperModel,
                        device=misc_parameters['device'],
                        number_of_classes=network_parameters['number_of_classes'],
                        experiment_name=training_parameters['experiment_name'],
162
                        optimizer= optimizer,
163
                        optimizer_arguments={'lr': training_parameters['learning_rate'],
164
165
166
167
                                             'betas': training_parameters['optimizer_beta'],
                                             'eps': training_parameters['optimizer_epsilon'],
                                             'weight_decay': training_parameters['optimizer_weigth_decay']
                                             },
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
                        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))

191
        del train_data, validation_data, train_loader, validation_loader, BrainMapperModel, solver, optimizer
192
193
194
195
196
197
198
199
        torch.cuda.empty_cache()

        return validation_loss

    if data_parameters['k_fold'] is None:

        _ = _train_runner(data_parameters, training_parameters,
                          network_parameters, misc_parameters)
200

201
    else:
202
        print("Training initiated using K-fold Cross Validation!")
203
        k_fold_losses = []
204

205
        for k in range(data_parameters['k_fold']):
206

207
            print("K-fold Number: {}".format(k+1))  
208

209
            data_parameters['train_list'] = os.path.join(
210
                data_parameters['data_folder_name'], 'train' + str(k+1)+'.txt')
211
            data_parameters['validation_list'] = os.path.join(
212
213
214
                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")
215

216
            validation_loss = _train_runner(
217
                data_parameters, training_parameters, network_parameters, misc_parameters)
218

219
            k_fold_losses.append(validation_loss)
220

221
222
223
        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)))
224

Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
225

226
def evaluate_score(training_parameters, network_parameters, misc_parameters, evaluation_parameters):
227
228
229
230
231
232
233
234
235
236
237
    """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'
        }

238
        network_parameters (dict): Contains information relevant parameters
239
240
241
        network_parameters= {
            'number_of_classes': 1
        }
242

243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
        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'
        }
    """

261
    # TODO - NEED TO UPDATE THE DATA FUNCTIONS!
262

263
264
265
    logWriter = LogWriter(number_of_classes=network_parameters['number_of_classes'],
                          logs_directory=misc_parameters['logs_directory'],
                          experiment_name=training_parameters['experiment_name']
266
                          )
267

268
    prediction_output_path = os.path.join(misc_parameters['experiments_directory'],
269
270
271
272
                                          training_parameters['experiment_name'],
                                          evaluation_parameters['saved_predictions_directory']
                                          )

273
    _ = evaluations.evaluate_dice_score(trained_model_path=evaluation_parameters['trained_model_path'],
274
275
276
277
278
279
280
281
282
283
                                        number_of_classes=network_parameters['number_of_classes'],
                                        data_directory=evaluation_parameters['data_directory'],
                                        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
    )
284
285
286

    logWriter.close()

287

288
289
def evaluate_mapping(mapping_evaluation_parameters):
    """Mapping Evaluator
290

291
292
    This function passes through the network an input and generates the rsfMRI outputs.
    This function allows the user to either use one or two or three paths.
293

294
295
296
297
    The convention for the different model paths is as follows:
    - model1: coronal
    - model2: axial
    - model3: saggital
298

299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
    However, this convention can be changed either bellow or the settings file.

    Args:
        mapping_evaluation_parameters (dict): Dictionary of parameters useful during mapping evaluation.
        mapping_evaluation_parameters = {
            'trained_model1_path': 'path/to/model1'
            'trained_model2_path': 'path/to/model2'
            'trained_model3_path': 'path/to/model3'
            'data_directory': 'path/to/data'
            'data_list': 'path/to/datalist.txt/
            'orientation1': 'coronal'
            'orientation2': 'axial'
            'orientation3': 'sagittal'
            'prediction_output_path': 'directory-of-saved-predictions'
            'batch_size': 2
            'device': 0
            'exit_on_error': True
            'number_of_paths': 3
        }

    """
    trained_model1_path = mapping_evaluation_parameters['trained_model1_path']
    trained_model2_path = mapping_evaluation_parameters['trained_model2_path']
    trained_model3_path = mapping_evaluation_parameters['trained_model3_path']
    data_directory = mapping_evaluation_parameters['data_directory']
    data_list = mapping_evaluation_parameters['data_list']
    orientation1 = mapping_evaluation_parameters['orientation1']
    orientation2 = mapping_evaluation_parameters['orientation2']
327
    orientation3 = mapping_evaluation_parameters['orientation3']
328
329
    prediction_output_path = mapping_evaluation_parameters['prediction_output_path']
    batch_size = mapping_evaluation_parameters['batch_size']
330
    device = mapping_evaluation_parameters['device']
331
332
333
    exit_on_error = mapping_evaluation_parameters['exit_on_error']

    if mapping_evaluation_parameters['number_of_paths'] == 1:
334
335
336
337
338
339
340
341
        evaluations.evaluate_single_path(trained_model1_path,
                                         data_directory,
                                         data_list,
                                         orientation1,
                                         prediction_output_path,
                                         batch_size,
                                         device=device,
                                         exit_on_error=exit_on_error)
342
343
    elif mapping_evaluation_parameters['number_of_paths'] == 2:
        evaluations.evaluate_two_paths(trained_model1_path,
344
345
346
347
348
349
350
351
352
                                       trained_model2_path,
                                       data_directory,
                                       data_list,
                                       orientation1,
                                       orientation2,
                                       prediction_output_path,
                                       batch_size,
                                       device=device,
                                       exit_on_error=exit_on_error)
353
354
    elif mapping_evaluation_parameters['number_of_paths'] == 3:
        evaluations.evaluate_all_paths(trained_model1_path,
355
356
357
358
359
360
361
362
363
364
365
366
367
368
                                       trained_model2_path,
                                       trained_model3_path,
                                       data_directory,
                                       data_list,
                                       orientation1,
                                       orientation2,
                                       orientation3,
                                       prediction_output_path,
                                       batch_size,
                                       device=device,
                                       exit_on_error=exit_on_error)


def delete_files(folder):
369
    """ Clear Folder Contents
370

371
372
373
374
    Function which clears contents (like experiments or logs)

    Args:
        folder (str): Name of folders whose conents is to be deleted
375

376
377
    Returns:
        None
378

379
380
    Raises:
        Exception: Any error
381
    """
382

383
384
385
386
387
388
389
390
391
392
    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)

393
394

if __name__ == '__main__':
395
    parser = argparse.ArgumentParser()
396
397
398
399
    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')
400
401
402
403
404
405
406
407
408
409

    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']

410
411
412
413
    # Here we shuffle the data!

    if data_parameters['data_split_flag'] == True:
        if data_parameters['use_data_file'] == True:
414
415
416
            data_test_train_validation_split(data_parameters['data_folder_name'],
                                             data_parameters['test_percentage'],
                                             data_parameters['subject_number'],
417
                                             data_directory=data_parameters['data_directory'],
418
419
420
                                             data_file=data_parameters['data_file'],
                                             K_fold=data_parameters['k_fold']
                                             )
421
        else:
422
423
424
425
426
427
            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']
                                             )
428
        update_shuffling_flag('settings.ini')
429

430
431
    if arguments.mode == 'train':
        train(data_parameters, training_parameters,
432
              network_parameters, misc_parameters)
433
434
435
436
437

    # NOTE: THE EVAL FUNCTIONS HAVE NOT YET BEEN DEBUGGED (16/04/20)

    elif arguments.mode == 'evaluate-score':
        evaluate_score(training_parameters,
438
                       network_parameters, misc_parameters, evaluation_parameters)
439
440
441
442
    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)
443
        else:
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
            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')