trial_times = trials.stimOn_times[incl_trials]
probability_left = trials.probabilityLeft[incl_trials]
trial_blocks = (trials.probabilityLeft[incl_trials] == 0.2).astype(int)
# Get clusters in this brain region
region_clusters = combine_layers_cortex(clusters[PROBE]['acronym'])
clusters_in_region = clusters[PROBE].metrics.cluster_id[region_clusters == REGION]
# Select spikes and clusters
spks_region = spikes[PROBE].times[np.isin(spikes[PROBE].clusters, clusters_in_region)]
clus_region = spikes[PROBE].clusters[np.isin(spikes[PROBE].clusters,
clusters_in_region)]
# Decode block identity
decode_block = decode(spks_region, clus_region, trial_times, trial_blocks,
pre_time=PRE_TIME, post_time=POST_TIME,
classifier=DECODER, cross_validation='kfold-interleaved',
num_splits=5)
shuffle_block = decode(spks_region, clus_region, trial_times, trial_blocks,
pre_time=PRE_TIME, post_time=POST_TIME,
classifier=DECODER, cross_validation='kfold-interleaved',
num_splits=5, shuffle=True, iterations=ITERATIONS)
pseudo_block = decode(spks_region, clus_region, trial_times, trial_blocks,
pre_time=PRE_TIME, post_time=POST_TIME,
classifier=DECODER, cross_validation='kfold-interleaved',
num_splits=5, pseudo_blocks=True, iterations=ITERATIONS)
# %%
"""
figure_style()
print(
'%d out of %d drift neurons detected' %
(drift_neurons.shape[0], np.unique(clus_region).shape[0]))
spks_region = spks_region[~np.isin(clus_region, drift_neurons)]
clus_region = clus_region[~np.isin(clus_region, drift_neurons)]
# Check if there are enough neurons in this brain region
if np.unique(clus_region).shape[0] < MIN_NEURONS:
continue
# Decode
decode_result = decode(spks_region,
clus_region,
trial_times,
trial_ids,
pre_time=PRE_TIME,
post_time=POST_TIME,
classifier=clf,
cross_validation=VALIDATION,
num_splits=NUM_SPLITS)
# Estimate chance level
if CHANCE_LEVEL == 'phase-rand':
decode_chance = decode(spks_region,
clus_region,
trial_times,
trial_ids,
pre_time=PRE_TIME,
post_time=POST_TIME,
classifier=clf,
cross_validation=VALIDATION,
# Select spikes and clusters
spks_region = spikes[probe].times[np.isin(spikes[probe].clusters,
clusters_in_region)]
clus_region = spikes[probe].clusters[np.isin(spikes[probe].clusters,
clusters_in_region)]
if len(spks_region) == 0:
continue
# Decode block identity
decode_block = decode(spks_region,
clus_region,
trial_times,
trial_blocks,
pre_time=PRE_TIME,
post_time=POST_TIME,
classifier=DECODER,
cross_validation=VALIDATION,
num_splits=5)
pseudo_block = decode(spks_region,
clus_region,
trial_times,
trial_blocks,
pre_time=PRE_TIME,
post_time=POST_TIME,
classifier=DECODER,
cross_validation=VALIDATION,
num_splits=5,
pseudo_blocks=True,
region_clusters)]
clus_region = spikes[probe].clusters[np.isin(spikes[probe].clusters,
region_clusters)]
if len(region_clusters) <= N_NEURONS:
continue
# Decode over time
decode_time = pd.DataFrame()
shuffle_time = pd.DataFrame()
for j, win_center in enumerate(WIN_CENTERS):
print('Decoding window [%d of %d]' % (j + 1, WIN_CENTERS.shape[0]))
decode_result = decode(spks_region,
clus_region,
stimon_times,
stimon_blocks,
pre_time=-win_center + (WIN_SIZE / 2),
post_time=win_center + (WIN_SIZE / 2),
classifier='bayes',
cross_validation='kfold',
n_neurons=N_NEURONS,
iterations=ITERATIONS)
decode_time = decode_time.append(pd.DataFrame({
'f1':
decode_result['f1'],
'accuracy':
decode_result['accuracy'] * 100,
'auroc':
decode_result['auroc'],
'win_center':
win_center,
'session':
'%s_%s' % (ses[i]['subject'], ses[i]['start_time'][:10])
spikes.clusters, clusters.metrics.cluster_id[clusters_to_use])]
spikes.clusters = spikes.clusters[np.isin(
spikes.clusters, clusters.metrics.cluster_id[clusters_to_use])]
cluster_ids = clusters.metrics.cluster_id[clusters_to_use]
# %% Do decoding
print('Decoding whether the stimulus was on the left or the right..')
stim_times = trials.goCue_times
stim_sides = np.isnan(trials.contrastLeft).astype(int)
# Decode left vs right stimulus from a 1 second window after stimulus onset using default settings:
# Naive Bayes classifier with 5-fold cross-validation
decode_result = decode(spikes.times,
spikes.clusters,
stim_times,
stim_sides,
pre_time=0,
post_time=1)
# Get the accuracy over chance
print('\nNaive Bayes with 5-fold cross-validation')
print('Performance: %.2f%% correct [chance level: %.2f%%]' %
(decode_result['accuracy'] * 100,
((stim_sides.sum() / stim_sides.shape[0]) * 100)))
# Decode stimulus side using a subset of 50 random neurons drawn 300 times
decode_result = decode(spikes.times,
spikes.clusters,
stim_times,
stim_sides,
pre_time=0,
region_clusters = combine_layers_cortex(clusters[PROBE]['acronym'])
clusters_in_region = clusters[PROBE].metrics.cluster_id[region_clusters ==
REGION]
# Select spikes and clusters
spks_region = spikes[PROBE].times[np.isin(spikes[PROBE].clusters,
clusters_in_region)]
clus_region = spikes[PROBE].clusters[np.isin(spikes[PROBE].clusters,
clusters_in_region)]
# Decode block identity
decode_5fold = decode(spks_region,
clus_region,
trial_times,
trial_blocks,
pre_time=PRE_TIME,
post_time=POST_TIME,
classifier=DECODER,
cross_validation='kfold',
num_splits=5)
shuffle_5fold = decode(spks_region,
clus_region,
trial_times,
trial_blocks,
pre_time=PRE_TIME,
post_time=POST_TIME,
classifier=DECODER,
cross_validation='kfold',
num_splits=5,
shuffle=True,
spikes.clusters = spikes.clusters[np.isin(
spikes.clusters, clusters.metrics.cluster_id[
clusters.metrics.ks2_label == 'good'])]
clusters.channels = clusters.channels[clusters.metrics.ks2_label == 'good']
clusters.depths = clusters.depths[clusters.metrics.ks2_label == 'good']
cluster_ids = clusters.metrics.cluster_id[clusters.metrics.ks2_label == 'good']
# Get trial vectors
incl_trials = (trials.probabilityLeft > 0.55) | (trials.probabilityLeft < 0.45)
trial_times = trials.goCue_times[incl_trials]
probability_left = trials.probabilityLeft[incl_trials]
trial_blocks = (trials.probabilityLeft[incl_trials] > 0.55).astype(int)
# Decode block identity
bayes_kfold = decode(spikes.times, spikes.clusters, trial_times, trial_blocks,
pre_time=PRE_TIME, post_time=POST_TIME,
classifier='bayes', cross_validation='kfold')
bayes_block = decode(spikes.times, spikes.clusters, trial_times, trial_blocks,
pre_time=PRE_TIME, post_time=POST_TIME,
classifier='bayes', cross_validation='block', prob_left=probability_left)
bayes_loo = decode(spikes.times, spikes.clusters, trial_times, trial_blocks,
pre_time=PRE_TIME, post_time=POST_TIME,
classifier='bayes', cross_validation='leave-one-out')
forest_kfold = decode(spikes.times, spikes.clusters, trial_times, trial_blocks,
pre_time=PRE_TIME, post_time=POST_TIME,
classifier='forest', cross_validation='kfold')
forest_block = decode(spikes.times, spikes.clusters, trial_times, trial_blocks,
pre_time=PRE_TIME, post_time=POST_TIME,
classifier='forest', cross_validation='block', prob_left=probability_left)
forest_loo = decode(spikes.times, spikes.clusters, trial_times, trial_blocks,
pre_time=PRE_TIME, post_time=POST_TIME,
for i, n_neurons in enumerate(N_NEURONS):
print('Decoding from groups of %d neurons [%d of %d]' %
(n_neurons, i + 1, len(N_NEURONS)))
for j in range(N_NEURON_PICK):
# Subselect neurons
use_neurons = np.random.choice(clusters_in_region,
n_neurons,
replace=False)
# Decode
decode_result = decode(spks_region[np.isin(clus_region, use_neurons)],
clus_region[np.isin(clus_region, use_neurons)],
trial_times,
trial_ids,
pre_time=PRE_TIME,
post_time=POST_TIME,
classifier=DECODER,
cross_validation=VALIDATION,
num_splits=NUM_SPLITS)
decode_subselects = decode_subselects.append(
pd.DataFrame(index=[decode_subselects.shape[0] + 1],
data={
'accuracy': decode_result['accuracy'],
'null': 'Original',
'n_neurons': n_neurons
}))
if CHANCE_LEVEL == 'shuffle':
decode_chance = decode(spks_region[np.isin(clus_region,
use_neurons)],
clus_region[np.isin(clus_region,