def testSearchSimilarPatterns(self):
template_start_time = 1
template_duration = 2
template_end_time = template_start_time + template_duration
template = np.ones((n_leads, template_duration * sampling_freq))
template_start_index = template_start_time * sampling_freq
template_end_index = template_end_time * sampling_freq
self.base_data[:, template_start_index:template_end_index] = template
target_start_time = 5
target_end_time = target_start_time + template_duration
target_start_index = target_start_time * sampling_freq
target_end_index = target_end_time * sampling_freq
self.base_data[:, target_start_index:target_end_index] = template
patterns_found = similarity.SearchSimilarPatterns(self.base_data,
template_start_time,
template_duration,
sampling_freq,
top_n=3)
self.assertLen(patterns_found, 3)
target_similar_pattern = similarity_pb2.SimilarPattern()
target_similar_pattern.score = 1
target_similar_pattern.start_time = target_start_time
target_similar_pattern.duration = template_duration
self.assertIn(target_similar_pattern, patterns_found)
def testSearchSimilarPatterns(self):
template_start_time = 1
template_duration = 2
template = np.ones((n_leads, template_duration * sampling_freq))
set_slice_value(self.base_data, template, template_start_time,
sampling_freq)
target_start_time = 5
set_slice_value(self.base_data, template, target_start_time,
sampling_freq)
patterns_found = similarity.SearchSimilarPatterns(self.base_data,
template_start_time,
template_duration,
[],
sampling_freq,
top_n=3)
self.assertLen(patterns_found, 3)
target_similar_pattern = similarity_pb2.SimilarPattern()
target_similar_pattern.score = 1
target_similar_pattern.start_time = target_start_time
target_similar_pattern.duration = template_duration
self.assertIn(target_similar_pattern, patterns_found)
def SearchSimilarPatterns(full_data,
window_start,
window_duration,
seen_events,
sampling_freq,
top_n=5,
merge_close_results=False,
merge_threshold=1):
"""Searches similar patterns for a target window in a 2d array.
Args:
full_data: numpy array that holds the full data to analyze.
Must have shape (n_channels, n_data_points).
window_start: the start of the window in seconds.
window_duration: the duration of the window in seconds.
seen_events: array of TimeSpan marking events already seen.
sampling_freq: sampling frequency used in the data.
top_n: Amount of similar results to return.
merge_close_results: Boolean indicating if merge between near results should
be performed.
merge_threshold: Amount of seconds to use as merge_threshold.
Returns:
Array of SimilarPatterns, holding the most similar patterns found.
"""
window_start_index = int(sampling_freq * window_start)
window_end_index = window_start_index + int(
sampling_freq * window_duration)
window_data = full_data[:, window_start_index:window_end_index]
_, n_samples = window_data.shape
if window_end_index <= window_start_index or n_samples == 0:
raise ValueError(
'Window must have positive duration: found %s-%s, (%s samples)' %
(window_end_index, window_start_index, n_samples))
scores = cv2.matchTemplate(full_data, window_data, cv2.TM_CCORR_NORMED)[0]
similar_patterns = []
for index, score in enumerate(scores):
sim_pattern = similarity_pb2.SimilarPattern()
sim_pattern.score = score
sim_pattern.duration = window_duration
sim_pattern.start_time = float(index) / sampling_freq
similar_patterns.append(sim_pattern)
similar_patterns = _FilterSeenResults(similar_patterns, seen_events)
top_results = _GetTopNonOverlappingResults(similar_patterns, top_n,
merge_close_results,
merge_threshold)
return top_results
def SearchSimilarPatterns(full_data,
window_start,
window_duration,
seen_events,
sampling_freq=200,
top_n=5):
"""Searches similar patterns for a target window in a 2d array.
Args:
full_data: numpy array that holds the full data to analyze.
Must have shape (n_channels, n_data_points).
window_start: the start of the window in seconds.
window_duration: the duration of the window in seconds.
seen_events: array of TimeSpan marking events already seen.
sampling_freq: sampling frequency used in the data.
top_n: Amount of similar results to return.
Returns:
Array of SimilarPattern proto objects, holding the most similar patterns
found.
"""
window_start_index = int(sampling_freq * window_start)
window_duration_index = int(sampling_freq * window_duration)
window_end_index = window_start_index + window_duration_index
window_data = full_data[:, window_start_index:window_end_index]
_, n_samples = window_data.shape
if window_end_index <= window_start_index or n_samples == 0:
raise ValueError(
'Window must have positive duration: found %s-%s, (%s samples)' %
(window_end_index, window_start_index, n_samples))
sims = cv2.matchTemplate(full_data, window_data, cv2.TM_CCORR_NORMED)[0]
sims = list(enumerate(sims))
sims = _FilterOverlappedResults(sims, seen_events, sampling_freq)
sims = sorted(sims, key=lambda x: x[1], reverse=True)
sims = _GetTopNonOverlappingResults(sims, top_n, window_duration_index)
sim_patterns = []
for index, score in sims:
sim_pattern = similarity_pb2.SimilarPattern()
sim_pattern.score = score
sim_pattern.duration = window_duration
sim_pattern.start_time = float(index) / sampling_freq
sim_patterns.append(sim_pattern)
return sim_patterns
def _MergePatterns(base_pattern, patterns):
"""Merges a list of patterns into a base pattern.
Keeps the longest span possible and the maximum score.
Args:
base_pattern: a SimilarPattern instance.
patterns: array of SimilarPatterns.
Returns:
A SimilarPattern with the merged information.
"""
start_time = base_pattern.start_time
end_time = base_pattern.start_time + base_pattern.duration
score = base_pattern.score
for other_pattern in patterns:
start_time = min(start_time, other_pattern.start_time)
end_time = max(end_time, other_pattern.start_time + other_pattern.duration)
score = max(score, other_pattern.score)
merged_pattern = similarity_pb2.SimilarPattern()
merged_pattern.start_time = start_time
merged_pattern.duration = end_time - start_time
merged_pattern.score = score
return merged_pattern
