rand = np.load(os.path.join(stimulus_map_directory, 'random_frames.npy'))
subset_averages = []
subset_counts = []
for i in np.arange(0, len(rand), 1000):
subset = rand[i:i + 1000].astype('float64')
subset_averages.append(subset.mean(axis=0))
subset_counts.append(float(len(subset)))
subset_averages = np.array(subset_averages)
subset_counts = np.array(subset_counts)
average = np.einsum('i,ijk->jk', subset_counts, subset_averages)
average /= subset_counts.sum()
np.save(os.path.join(stimulus_map_directory, 'random_average.npy'),
average)
# Compute histogram
histogram = np.zeros((250, 250))
for img in rand:
img_points, img_orientations = find_paramecia(img)
img_counts, x, y = np.histogram2d(
*img_points.T, bins=[np.arange(0, 251),
np.arange(0, 251)])
histogram += img_counts
np.save(os.path.join(stimulus_map_directory, 'random_histogram.npy'),
histogram)
print 'Time taken:', timer.convert_time(timer.stop())
# Compare first antisymmetric transition mode
a1_dot_a = np.abs(np.dot(USVa1[0, :, 0], USVa2[0, :, 0]))
a1_dot_b = np.abs(np.dot(USVa1[0, :, 0], USVa2[0, :, 1]))
a1_dot = max(a1_dot_a, a1_dot_b)
# Append to output
dot_products.append((s1_dot, s2_dot, a1_dot))
i += 1
return np.array(dot_products)
if __name__ == "__main__":
transition_directory = os.path.join(experiment.subdirs['analysis'],
'transitions')
control_matrices = np.load(
os.path.join(transition_directory, 'control',
'transition_matrices.npy'))
ablated_matrices = np.load(
os.path.join(transition_directory, 'ablated',
'transition_matrices.npy'))
timer = Timer()
timer.start()
dot_products = compare_transition_modes(ablated_matrices,
control_matrices,
exact=False,
n_permutations=100)
np.save(os.path.join(transition_directory, 'compare_control_ablated.npy'),
dot_products)
print timer.convert_time(timer.stop())
from behaviour_analysis.analysis.stimulus_mapping import BoutStimulusMapper
import os
import pandas as pd
if __name__ == "__main__":
for experiment in (blu, lak):
print_heading(os.path.basename(experiment.directory))
stimulus_map_directory = create_folder(experiment.subdirs['analysis'],
'stimulus_maps')
mapped_bouts = pd.read_csv(os.path.join(experiment.subdirs['analysis'],
'mapped_bouts.csv'),
index_col=0,
dtype={
'ID': str,
'video_code': str
})
# Calculate stimulus maps for each fish in parallel
timer = Timer()
timer.start()
mapper = BoutStimulusMapper(mapped_bouts,
experiment,
stimulus_map_directory,
n_threads=20)
mapper.run()
total_time = timer.stop()
print 'Total time:', timer.convert_time(total_time)
# Transform
transformed_strikes = capture_strike_data.map(eigenfish, whiten=True, mean=mean_tail, std=std_tail)
transformed_strikes = transformed_strikes.list_bouts(values=True, ndims=3)
# Truncate
truncated_strikes = [bout[12:37] for bout in transformed_strikes]
bw = 0.006 # 3 frames
# Calculate distance matrix
print_heading('CALCULATING CAPTURE STRIKE DISTANCE MATRIX')
timer = Timer()
timer.start()
D_normal = calculate_distance_matrix(truncated_strikes, bw=bw)
D_flipped = calculate_distance_matrix(truncated_strikes, bw=bw, flip=True)
D = np.min([D_normal, D_flipped], axis=0)
np.save(paths['distance_matrix'], D)
time_taken = timer.stop()
print 'Time taken: {}'.format(timer.convert_time(time_taken))
# Perform embedding
D = squareform(D)
np.random.seed(1992)
isomap = IsomapPrecomputed(n_neighbors=5, n_components=2)
isomapped_strikes = isomap.fit_transform(D)
np.save(paths['isomapped_strikes'], isomapped_strikes)
# Cluster
cluster_labels = KMeans(2).fit_predict(isomapped_strikes)
capture_strikes['cluster_label'] = cluster_labels
capture_strikes.to_csv(paths['capture_strikes'], index=True)