def get_data_source_batch_summary(self, data_source_batch_name):
failed_docs_directory = self.load_config.failed_docs_directory(
data_source_batch_name)
loaded_docs_directory = self.load_config.loaded_docs_directory(
data_source_batch_name)
updated_ids = []
indexed_ids = []
failed_ids = []
for data_loader_batch_name in os.listdir(failed_docs_directory):
data_loader_batch_path = os.path.join(
failed_docs_directory, data_loader_batch_name)
if os.path.isfile(data_loader_batch_path) and data_loader_batch_name.startswith(DATA_LOADER_BATCH_PREFIX):
batch_failed_docs = file_utils.load_file(
failed_docs_directory, data_loader_batch_name)
failed_ids.extend(batch_failed_docs.keys())
for data_loader_batch_name in os.listdir(loaded_docs_directory):
data_loader_batch_path = os.path.join(
loaded_docs_directory, data_loader_batch_name)
if os.path.isfile(data_loader_batch_path) and data_loader_batch_name.startswith(DATA_LOADER_BATCH_PREFIX):
data_loader_batch = file_utils.load_file(
loaded_docs_directory, data_loader_batch_name)
updated_ids.extend(data_loader_batch['updated_ids'])
indexed_ids.extend(data_loader_batch['indexed_ids'])
summary = dict()
summary['updated_ids'] = updated_ids
summary['indexed_ids'] = indexed_ids
summary['failed_ids'] = failed_ids
return summary
def get_loaded_doc_count(self):
updated_ids = []
indexed_ids = []
data_source_directory = self.load_config.data_source_directory()
for data_source_batch_name in os.listdir(data_source_directory):
data_source_batch_path = os.path.join(data_source_directory,
data_source_batch_name)
if not os.path.isfile(data_source_batch_path
) and data_source_batch_name.startswith(
DATA_SOURCE_BATCH_PREFIX
) and 'old' not in data_source_batch_name:
self.load_config.log(LOG_LEVEL_TRACE,
'Processing data source batch:',
data_source_batch_path)
for data_loader_batch_name in os.listdir(
data_source_batch_path):
data_loader_batch_path = os.path.join(
data_source_batch_path, data_loader_batch_name)
if os.path.isfile(data_loader_batch_path
) and data_loader_batch_name.startswith(
DATA_LOADER_BATCH_PREFIX):
data_loader_batch = file_utils.load_file(
data_source_batch_path, data_loader_batch_name)
updated_ids.extend(data_loader_batch['updated_ids'])
indexed_ids.extend(data_loader_batch['indexed_ids'])
return len(updated_ids) + len(indexed_ids)
def get_progress(self):
data_source_directory = self.load_config.data_source_directory()
stats_file_name = self.data_source_stats_file_name()
if self.data_source_stats is None:
self.data_source_stats = file_utils.load_file(
data_source_directory, stats_file_name)
row_count = 0
unique_ids = 0
# if 'row_count' in self.data_source_stats:
# row_count = self.data_source_stats['row_count']
if 'unique_ids' in self.data_source_stats:
unique_ids = self.data_source_stats['unique_ids']
docs_loaded = self.get_loaded_doc_count()
if unique_ids > 0:
self.load_config.log(LOG_LEVEL_INFO, 'docs loaded', docs_loaded,
'unique_ids', unique_ids)
progress = (docs_loaded / float(unique_ids)) * 100
self.data_source_stats['progress'] = progress
file_utils.save_file(data_source_directory, stats_file_name,
self.data_source_stats)
return progress
return -1
def get_failed_docs_files(self, data_source_name=None):
data_source_batches = {}
if data_source_name is not None:
data_source_directory = self.data_source_directory(
data_source_name=data_source_name)
dsb = file_utils.load_file(data_source_directory,
DATA_SOURCE_BATCHES_FILE)
if data_source_name not in data_source_batches:
data_source_batches[data_source_name] = []
for data_source_batch in dsb:
data_source_batches[data_source_name].append(data_source_batch)
else:
generated_files_directory = self.generated_files_directory()
print 'generated_files_directory', generated_files_directory
for name in os.listdir(generated_files_directory):
print 'data_source_name', name
path = os.path.join(generated_files_directory, name)
if os.path.isdir(path):
data_source_directory = self.data_source_directory(
data_source_name=name)
print 'data_source_directory', data_source_directory
dsb = file_utils.load_file(data_source_directory,
DATA_SOURCE_BATCHES_FILE)
if name not in data_source_batches:
data_source_batches[name] = []
for data_source_batch in dsb:
data_source_batches[name].append(data_source_batch)
failed_docs_files = []
for data_source_name in data_source_batches:
print 'data_source_name', data_source_name
for data_source_batch in data_source_batches[data_source_name]:
failed_docs_directory = self.failed_docs_directory(
data_source_batch_name=data_source_batch,
data_source_name=data_source_name)
print 'failed_docs_directory', failed_docs_directory
for name in os.listdir(failed_docs_directory):
# print 'failed_doc_file', name
path = os.path.join(failed_docs_directory, name)
print 'failed_doc_file', path
if os.path.isfile(path) and name.endswith('.json'):
failed_docs_files.append(path)
return failed_docs_files
def load_stats(self):
self.load_config.log(LOG_LEVEL_INFO, 'Loading stats...')
data_source_directory = self.load_config.data_source_directory()
stats = file_utils.load_file(data_source_directory, 'stats.json')
if 'total_rows' in stats:
self.total_rows = stats['total_rows']
if 'total_ids' in stats:
self.total_ids = stats['total_ids']
return stats
def get_data_source_batch_summary(self, data_source_batch_name):
data_source_batch_directory = self.load_config.data_source_batch_directory(
data_source_batch_name)
data_source_directory = self.load_config.data_source_directory()
updated_ids = []
indexed_ids = []
failed_ids = []
skipped_ids = []
for data_loader_batch_name in os.listdir(data_source_batch_directory):
data_loader_batch_path = os.path.join(data_source_batch_directory,
data_loader_batch_name)
if os.path.isfile(data_loader_batch_path
) and data_loader_batch_name.startswith(
DATA_LOADER_BATCH_PREFIX):
data_loader_batch = file_utils.load_file(
data_source_batch_directory, data_loader_batch_name)
updated_ids.extend(data_loader_batch['updated_ids'])
indexed_ids.extend(data_loader_batch['indexed_ids'])
failed_ids.extend(data_loader_batch['failed_ids'])
skipped_ids.extend(data_loader_batch['skipped_ids'])
summary = dict()
summary['updated_ids'] = updated_ids
summary['indexed_ids'] = indexed_ids
summary['failed_ids'] = failed_ids
summary['skipped_ids'] = skipped_ids
data_source_batch_summary = file_utils.load_file(
data_source_directory, data_source_batch_name + '.json')
for key in data_source_batch_summary:
summary[key] = data_source_batch_summary[key]
return summary
def get_failed_ids(self):
self.load_config.log(LOG_LEVEL_INFO, 'Fetching failed ids...')
failed_ids = {}
data_source_directory = self.load_config.data_source_directory()
for name in os.listdir(data_source_directory):
file_path = os.path.join(data_source_directory, name)
if os.path.isfile(file_path) and name.startswith(DATA_SOURCE_BATCH_PREFIX):
batch_info = file_utils.load_file(data_source_directory, name)
batch_failed_ids = batch_info['failed_ids']
batch_failed_ids = dict.fromkeys(batch_failed_ids, None)
failed_ids.update(batch_failed_ids)
self.load_config.log(LOG_LEVEL_INFO, 'Failed ids', len(failed_ids))
return failed_ids
def get_processed_indices(self):
self.load_config.log(LOG_LEVEL_INFO, 'Fetching processed ids...')
processed_indices = {}
data_source_directory = self.load_config.data_source_directory()
for name in os.listdir(data_source_directory):
file_path = os.path.join(data_source_directory, name)
if os.path.isfile(file_path) and name.startswith(DATA_SOURCE_BATCH_PREFIX):
batch_info = file_utils.load_file(data_source_directory, name)
if 'processed_indices' in batch_info:
batch_processed_indices = batch_info['processed_indices']
batch_processed_indices = dict.fromkeys(batch_processed_indices, None)
processed_indices.update(batch_processed_indices)
self.load_config.log(
LOG_LEVEL_INFO, 'Processed indices', len(processed_indices))
return processed_indices
def get_failed_ids(self, data_loader_batch_directory):
failed_ids = {}
for name in os.listdir(data_loader_batch_directory):
file_path = os.path.join(data_loader_batch_directory, name)
if os.path.isfile(file_path) and name.startswith("failed_docs_"):
self.load_config.log(LOG_LEVEL_TRACE, 'processing file:',
file_path)
failed_docs = file_utils.load_file(data_loader_batch_directory,
name)
for _id in failed_docs:
failed_ids[_id] = failed_docs[_id]
self.load_config.log(LOG_LEVEL_DEBUG, 'Failed ids', len(failed_ids))
return failed_ids
def list_failed_docs(self, data_source_batch_directory):
# print '...Processing', data_loader_batch_directory
for name in os.listdir(data_source_batch_directory):
file_path = os.path.join(data_source_batch_directory, name)
if os.path.isfile(file_path) and name.startswith("failed_docs_"):
failed_docs = file_utils.load_file(data_source_batch_directory,
name)
print file_path, '- Failed docs', len(failed_docs)
if len(failed_docs) > 0:
a = raw_input('List docs? (y/n)')
if a.lower() in ['y', 'yes']:
for _id in failed_docs:
reason = failed_docs[_id]['reason']
print 'Doc:', _id
print 'Reason', reason
c = raw_input('Continue?')
if c.lower() in ['n', 'no']:
break
def get_processed_rows(self):
data_source_directory = self.load_config.data_source_directory()
processed_rows = 0
for name in os.listdir(data_source_directory):
file_path = os.path.join(data_source_directory, name)
if os.path.isfile(file_path) and name.startswith(
DATA_SOURCE_BATCH_PREFIX):
self.load_config.log(LOG_LEVEL_DEBUG, 'processing file:',
file_path)
batch_data = file_utils.load_file(data_source_directory, name)
start_index = batch_data['start_index']
row_count = batch_data['row_count']
end_index = start_index + row_count
if end_index > processed_rows:
processed_rows = end_index
return processed_rows
def count_rows(self):
data_source_directory = self.load_config.data_source_directory()
stats_file_name = self.data_source_stats_file_name()
self.data_source_stats = file_utils.load_file(data_source_directory,
stats_file_name)
if self.data_source_stats is None or len(self.data_source_stats) == 0:
self.count = 0
self.data_source_batch = {}
self.data_source.process_rows(0, self.count_row)
self.load_config.log(LOG_LEVEL_INFO, 'Total rows:', self.count)
self.load_config.log(LOG_LEVEL_INFO, 'Total ids:',
len(self.data_source_batch))
self.data_source_stats = {
'row_count': self.count,
'unique_ids': len(self.data_source_batch)
}
file_utils.save_file(data_source_directory, stats_file_name,
self.data_source_stats)
def get_failed_ids(self, data_source_batch_name):
failed_docs_directory = self.load_config.failed_docs_directory(
data_source_batch_name)
failed_ids = {}
for name in os.listdir(failed_docs_directory):
file_path = os.path.join(failed_docs_directory, name)
if os.path.isfile(file_path) and name.startswith(
DATA_LOADER_BATCH_PREFIX):
self.load_config.log(LOG_LEVEL_TRACE, 'processing file:',
file_path)
batch_failed_docs = file_utils.load_file(
failed_docs_directory, name)
for _id in batch_failed_docs:
failed_ids[_id] = 0
self.load_config.log(LOG_LEVEL_DEBUG, 'Failed ids', len(failed_ids))
return failed_ids
def get_combined_data_source_summary(self):
updated_ids = {}
indexed_ids = {}
failed_ids = {}
data_source_directory = self.load_config.data_source_directory()
data_source_batches = self.load_data_source_batches()
for data_source_batch_name in data_source_batches:
data_source_batch_summary = self.get_data_source_batch_summary(
data_source_batch_name)
# Remove duplicates across batches
for _id in data_source_batch_summary['updated_ids']:
updated_ids[_id] = 0
for _id in data_source_batch_summary['indexed_ids']:
indexed_ids[_id] = 0
for _id in data_source_batch_summary['failed_ids']:
failed_ids[_id] = 0
# failed_ids can be present in updated_ids or indexed_ids arrays.
# filter those out
filtered_failed_ids = {}
for _id in failed_ids:
if _id not in updated_ids and _id not in indexed_ids:
filtered_failed_ids[_id] = 0
# Load the data source stats
data_source_summary = file_utils.load_file(
data_source_directory, 'stats.json')
summary = dict()
summary['total_rows'] = data_source_summary['total_rows']
summary['total_ids'] = data_source_summary['total_ids']
summary['updated_ids'] = updated_ids
summary['indexed_ids'] = indexed_ids
summary['failed_ids'] = filtered_failed_ids
return summary
def get_loaded_ids(self, data_loader_batch_directory):
loaded_ids = {}
for name in os.listdir(data_loader_batch_directory):
file_path = os.path.join(data_loader_batch_directory, name)
if os.path.isfile(file_path) and name.startswith(
DATA_LOADER_BATCH_PREFIX):
self.load_config.log(LOG_LEVEL_TRACE, 'processing file:',
file_path)
batch_data = file_utils.load_file(data_loader_batch_directory,
name)
updated_ids = batch_data['updated_ids']
indexed_ids = batch_data['indexed_ids']
for _id in updated_ids:
loaded_ids[_id] = 0
for _id in indexed_ids:
loaded_ids[_id] = 0
self.load_config.log(LOG_LEVEL_DEBUG, 'Loaded ids', len(loaded_ids))
return loaded_ids
def __init__(
self,
index_filenames,
window_length=1, # in seconds
shift=0.5, # in seconds
timesteps=19, # in seconds
sampling_rate=256, # in Hz
batch_size=10,
do_standard_scaling=True,
in_training_mode=False,
balance_batches=False,
patient_id=None):
'''
Constructor to create objects of the class **RawDataProcessor** and load all the data.
Future implementations will pave the way to load data from files on-the-fly in order to allow work
with large enough datasets.
Parameters
----------
:param self:
Reference to the current object.
:param list index_filenames:
List of index filenames from which to load data filenames.
:param int window_length:
Length of the window (in seconds) to slide through the signal.
:param int shift:
Number of seconds to shift the window to get each sample.
:param int timesteps:
Number of time steps for the sequences.
:param int sampling_rate:
Sampling rate of the signals, in Hz.
:param int batch_size:
Size of the batch to use.
:param boolean do_standard_scaling:
Flag to indicate whether to scale features to zero
mean and unit variance.
:param boolean in_training_mode:
Flag to indicate whether the process is in training mode.
:param boolean balance_batches:
Flag to indicate whether to balance the batches.
:param string patient_id:
String to indicate the patient id. It is used to save the statistics file.
'''
#
if patient_id is None:
raise Exception('You have to specify a patient id, i.e. "chb01"')
self.sampling_rate = sampling_rate
self.batch_size = batch_size
self.do_standard_scaling = do_standard_scaling
self.in_training_mode = in_training_mode
self.balance_batches = balance_batches
self.patient_id = patient_id
#
self.window_length = window_length * sampling_rate
self.sample_shift = int(sampling_rate * shift) # Half of sample length
self.timesteps = timesteps
self.num_channels = 23
#
self.input_shape = None
self.num_batches = 0
#
self.filenames = list()
for fname in index_filenames:
with open(fname, 'r') as f:
for l in f:
if l[0] != '#':
self.filenames.append(l.strip() + '.edf.pbz2')
f.close()
#
# List to store the data separated by files
self.data = list()
# List to store the labels associated at each sample separated by files
self.labels = list()
self.num_seizures = 0
num_ictal = 0
num_interictal = 0
print("Loading EDF signals...")
for i in tqdm(range(len(self.filenames))):
d_p = load_file(self.filenames[i],
exclude_seizures=False,
do_preemphasis=False,
separate_seizures=True,
verbose=0)
len_file = 0
self.data.append([])
self.labels.append([])
for p, label in d_p:
self.data[-1].append(p) # p.tolist() # Append data
self.labels[-1] += [label] * len(p)
if label == 1:
self.num_seizures += 1
num_ictal += len(p)
elif label == 0:
num_interictal += len(p)
len_file += len(p)
#
#print(self.data[-1][0].shape)
self.data[-1] = numpy.concatenate(self.data[-1], axis=0)
#print(self.data[-1].shape)
#
# Iterate over the data list and generate indices for the sequences
# We will treat each recording as independent from the others
self.file_indexes = list()
for i in range(len(self.data)):
limit = len(self.data[i]) // self.sample_shift
limit -= self.timesteps + 1
limit = (limit * self.sample_shift) # - self.sample_shift
self.file_indexes.append(
numpy.arange(0, limit, step=self.sample_shift).tolist())
#
num_sequences_per_class = [0, 0]
# Sequences indexes
if self.balance_batches:
# The batches will have the same samples for each class
# List for the indices of class 0
self.sequences_indexes_c0 = list()
# List for the indices of class 1
self.sequences_indexes_c1 = list()
for fi in range(len(self.file_indexes)):
for t in self.file_indexes[fi]:
label = self.labels[fi][t + (self.timesteps + 1) *
self.sample_shift - 1]
if label == 0:
self.sequences_indexes_c0.append((fi, t))
elif label == 1:
self.sequences_indexes_c1.append((fi, t))
#
#
num_sequences_per_class[0] = len(self.sequences_indexes_c0)
num_sequences_per_class[1] = len(self.sequences_indexes_c1)
num_sequences = num_sequences_per_class[
0] + num_sequences_per_class[1]
self.num_batches = num_sequences // self.batch_size
if num_sequences % self.batch_size != 0:
self.num_batches += 1
#
else:
# Just one list for all of the sequences. No balancing
self.sequences_indexes = list()
for fi in range(len(self.file_indexes)):
for t in self.file_indexes[fi]:
label = self.labels[fi][t + (self.timesteps + 1) *
self.sample_shift - 1]
num_sequences_per_class[label] += 1
self.sequences_indexes.append((fi, t, label))
#
#
num_sequences = len(self.sequences_indexes)
self.num_batches = num_sequences // self.batch_size
if num_sequences % self.batch_size != 0:
self.num_batches += 1
#
self.input_shape = (self.window_length, )
print('Signals loaded!')
print(
'\n-----------------------------------------------------------\n')
print(f'Number of seizures available: {self.num_seizures}')
#if self.num_seizures < 3:
# raise Exception('Not enough seizures, please try other patient.')
print(
f'Number of samples (not sequences): {num_ictal + num_interictal}')
print(
f'Interictal samples: {num_interictal} ({(num_interictal / (num_ictal + num_interictal) * 100):.2f} %)'
)
print(
f'Ictal samples: {num_ictal} ({(num_ictal / (num_ictal + num_interictal) * 100):.2f} %)'
)
print(f'\nNumber of sequences: {num_sequences}')
print(
f'Interictal sequences: {num_sequences_per_class[0]} ({num_sequences_per_class[0] / num_sequences * 100.0 :.2f}%)'
)
print(
f'Ictal sequences: {num_sequences_per_class[1]} ({num_sequences_per_class[1] / num_sequences * 100.0 :.2f}%)'
)
print(f'Number of batches: {self.num_batches}')
print(
'\n-----------------------------------------------------------\n')
# Standard scaling
if self.do_standard_scaling:
os.makedirs('stats', exist_ok=True)
if self.in_training_mode:
print('Calculating statistics to scale the data...')
means = []
counts = []
stddevs = []
for p in tqdm(self.data):
#p = numpy.array(p)
#print(p.shape)
means.append(p.mean())
counts.append(len(p))
stddevs.append(p.std())
#
#
self.mean = sum([m * c for m, c in zip(means, counts)])
self.std = sum([s * c for s, c in zip(stddevs, counts)])
self.mean /= sum(counts)
self.std /= sum(counts)
#
array = numpy.array([self.mean, self.std])
numpy.save(f'stats/statistics_detection_raw_{patient_id}.npy',
array)
del means
del counts
del stddevs
#
else:
print('Loading statistics to scale the data...')
array = numpy.load(
f'stats/statistics_detection_raw_{patient_id}.npy')
self.mean = array[0]
self.std = array[1]
#
del array
gc.collect()
def load_data_source_batches(self):
data_source_directory = self.load_config.data_source_directory()
return file_utils.load_file(data_source_directory, DATA_SOURCE_BATCHES_FILE)
number_of_subsets = 20
size_of_subsets = 500
max_iterations = 100 # optimal value: 5000 - but extremely slow. Keep it to 100 for faster results
hc_count = 30
algorithms = [
["Constructive", greedy_constructive, None, 'red'],
["Hill Climb (%d iter.)" % max_iterations, hill_climb, [max_iterations], 'green'],
["MHC (%d iter. %d runs)" % (max_iterations, hc_count), multi_hc, [max_iterations, hc_count], 'blue'],
# ["Spectrum v1", spectrum_sort, None, 'violet'],
["Spectrum v2", spectrum_sort_shake, None, 'purple']
]
size, colours = load_file()
subsets = list(generate_subsets(colours, number_of_subsets, size_of_subsets))
times = []
for algorithm in algorithms:
algorithm_times = list(benchmark_algorithm(subsets, algorithm[1], algorithm[2]))
times.append(algorithm_times)
print("%s min/avg/max/std = %.2f/%.2f/%.2f/%.2f ms" % (
algorithm[0],
np.min(algorithm_times) * 1000,
np.average(algorithm_times) * 1000,
np.max(algorithm_times) * 1000,
np.std(algorithm_times) * 1000,
))
