if __name__ == "__main__":
#################
n_dims = 3
frame_rate = 500.
#################
# Import data from mapping experiment
mapping_space_directory = os.path.join(mapping_experiment.subdirs['analysis'], 'behaviour_space')
eigenfish = np.load(os.path.join(mapping_space_directory, 'eigenfish.npy'))
mean_tail, std_tail = np.load(os.path.join(mapping_space_directory, 'tail_statistics.npy'))
exemplar_info = pd.read_csv(os.path.join(mapping_experiment.subdirs['analysis'], 'exemplars.csv'),
index_col='bout_index',
dtype={'ID': str, 'video_code': str})
exemplar_info = exemplar_info[exemplar_info['clean']]
exemplars = BoutData.from_directory(exemplar_info, mapping_experiment.subdirs['kinematics'],
check_tail_lengths=False, tail_columns_only=True)
exemplars = exemplars.map(eigenfish, whiten=True, mean=mean_tail, std=std_tail)
exemplars = exemplars.list_bouts(values=True, ndims=n_dims)
# Set paths
output_directory = create_folder(experiment.subdirs['analysis'], 'distance_matrices')
# Import experiment bouts
experiment_bouts = import_bouts(experiment.directory)
experiment_bouts = experiment_bouts.map(eigenfish, whiten=True, mean=mean_tail, std=std_tail)
# Compute distance matrices
print_heading('CALCULATING DISTANCE MATRICES')
distances = {}
analysis_times = []
timer = Timer()
import pandas as pd
if __name__ == "__main__":
frame_rate = 500.
n_dims = 3
bout_indices = np.load(paths['bout_indices'])
bouts_df = pd.read_csv(paths['bouts'],
dtype={
'ID': str,
'video_code': str
})
bouts_df = bouts_df.loc[bout_indices]
bouts = BoutData.from_directory(bouts_df,
experiment.subdirs['kinematics'],
tail_columns_only=True,
check_tail_lengths=False)
eigenfish = np.load(paths['eigenfish'])
mean_tail, std_tail = np.load(paths['tail_statistics'])
transformed = bouts.map(eigenfish,
whiten=True,
mean=mean_tail,
std=std_tail)
transformed_bouts = transformed.list_bouts(values=True, ndims=n_dims)
print_heading('CALCULATING DISTANCE MATRIX - NORMAL')
distance_matrix = calculate_distance_matrix(transformed_bouts,
fs=frame_rate,
flip=False)
strike_frames['start']).values / 500.
example_lengths = (strike_frames['end'] -
strike_frames['start']).values / 500.
for key, label in cluster_labels.iteritems():
strike_times = example_times[strike_frames['strike_cluster'] == label]
kde = KernelDensity(0.01).fit(strike_times.reshape(-1, 1))
density = np.exp(
kde.score_samples(bins)) * 0.01 * len(strike_times) / float(
len(example_times))
strike_info[key]['capture_times'] = strike_times
strike_info[key]['capture_times_kde'] = density
# Import kinematic data for example strikes
example_strike_frames = BoutData.from_directory(
strike_frames,
experiment.subdirs['kinematics'],
check_tail_lengths=False,
tail_columns_only=False)
example_capture_frames = example_strike_frames.list_bouts()
example_capture_frames = [
bout.loc[:, 'tip'].values for bout in example_capture_frames
]
# ================
# COMPOSITE FIGURE
# ================
# Make axes
fig = plt.figure(figsize=(1.854, 3))
gs = gridspec.GridSpec(4, 1)
ax1 = fig.add_subplot(gs[0])
if __name__ == "__main__":
# =========
# Open data
# =========
capture_strike_info = pd.read_csv(
os.path.join(experiment.subdirs['analysis'], 'capture_strikes', 'capture_strikes.csv'),
index_col=0, dtype={'ID': str, 'video_code': str})
eigenfish = np.load(os.path.join(experiment.subdirs['analysis'], 'behaviour_space', 'eigenfish.npy'))
mean, std = np.load(os.path.join(experiment.subdirs['analysis'], 'behaviour_space', 'tail_statistics.npy'))
# =======================================================
# Load bout data for capture strikes and project onto PCs
# =======================================================
capture_strikes = BoutData.from_directory(capture_strike_info, experiment.subdirs['kinematics'],
check_tail_lengths=False, tail_columns_only=True)
transformed_strikes = capture_strikes.map(eigenfish, whiten=True, mean=mean, std=std)
transformed_strikes = transformed_strikes.list_bouts(values=True, ndims=3)
# =========================================================
# Split strikes by cluster and flip to be in same direction
# =========================================================
clustered = [[], []]
for l, strike in zip(capture_strike_info['strike_cluster'].values, transformed_strikes):
truncated = strike[12:37]
truncated = truncated * np.sign(truncated[:10, 1][np.abs(truncated[:10, 1]).argmax()])
clustered[l].append(truncated)
cluster_mean_traces = [np.mean(traces, axis=0) for traces in clustered]
# ==============================================================
# Align all traces to cluster average and calculate new averages
if __name__ == "__main__":
# Load the exemplar bout data
exemplars = pd.read_csv(os.path.join(experiment.subdirs['analysis'],
'exemplars.csv'),
index_col='bout_index',
dtype={
'ID': str,
'video_code': str
})
exemplars = exemplars[exemplars['clean']]
labels = exemplars['module'].values
exemplar_data = BoutData.from_directory(exemplars,
experiment.subdirs['kinematics'],
check_tail_lengths=False,
tail_columns_only=False)
eigenfish = np.load(
os.path.join(experiment.subdirs['analysis'], 'behaviour_space',
'eigenfish.npy'))
tail_stats = np.load(
os.path.join(experiment.subdirs['analysis'], 'behaviour_space',
'tail_statistics.npy'))
exemplar_kinematics = BoutData(exemplar_data.data.loc[:, 'k0':'k49'],
exemplar_data.metadata)
exemplars_transformed = exemplar_kinematics.map(eigenfish, True,
tail_stats[0],
tail_stats[1])
mapped_bouts = pd.read_csv(paths['mapped_bouts'],
index_col='bout_index',
dtype={
'ID': str,
'video_code': str
})
n_states = len(mapped_bouts['state'].unique())
cluster_indices = [
np.where(mapped_bouts['state'].values == l)[0]
for l in np.arange(n_states)
]
mapped_bouts_data = BoutData.from_directory(
mapped_bouts,
experiment.subdirs['kinematics'],
tail_columns_only=False,
check_tail_lengths=False)
all_bouts = mapped_bouts_data.list_bouts()
# Turn angles
turn_angles = np.array(
[bout['angular_velocity'].sum() for bout in all_bouts])
turn_angles = np.degrees(
np.abs(turn_angles)
) / 500. # ignore direction, convert to degrees, account for frame rate
turn_angles = np.array([
np.median(turn_angles[idxs]) for idxs in cluster_indices
]) # take cluster medians
# Maximum angular velocity