if VALIDATION[6:] == 'interleaved':
y_pred, y_probs = decode(this_pop_vector, trial_ids,
NUM_SPLITS, True)
else:
y_pred, y_probs = decode(this_pop_vector, trial_ids,
NUM_SPLITS, False)
# Calculate performance metrics and confusion matrix
decode_result.loc[k, 'accuracy'] = accuracy_score(
trial_ids, y_pred)
decode_result.loc[k, 'f1'] = f1_score(trial_ids, y_pred)
decode_result.loc[k, 'auroc'] = roc_auc_score(
trial_ids[~np.isnan(y_probs)], y_probs[~np.isnan(y_probs)])
# Decode pseudo session
pseudo_trials = generate_pseudo_session(trials)
incl_pseudo_trials = balanced_trial_set(pseudo_trials)
# Select activity matrix and trial ids for this iteration
this_pseudo_pop_vector = pop_vector[incl_pseudo_trials]
pseudo_trial_ids = (trials.probabilityLeft[incl_pseudo_trials]
== 0.2).astype(int)
if VALIDATION[6:] == 'interleaved':
y_pred, y_probs = decode(this_pseudo_pop_vector,
pseudo_trial_ids, NUM_SPLITS,
True)
else:
y_pred, y_probs = decode(this_pseudo_pop_vector,
pseudo_trial_ids, NUM_SPLITS,
False)
these_priors = priors[j][:trials.shape[0]][incl_trials]
pred_prior, pred_prior_train = regress(
population_activity,
these_priors,
cross_validation=cv,
return_training=True)
r_prior[n] = pearsonr(these_priors, pred_prior)[0]
r_prior_train[n] = pearsonr(these_priors,
pred_prior_train)[0]
# Estimate chance level
r_pseudo, r_train_pseudo = np.empty(ITERATIONS), np.empty(
ITERATIONS)
for k in range(ITERATIONS):
if 'prior-stimside' in TARGET:
pseudo_trials = generate_pseudo_session(
trials, generate_choices=False)
pseudo_trials['choice'] = np.nan
elif 'prior-prevaction' in TARGET:
pseudo_trials = generate_pseudo_session(
trials, generate_choices=True)
stim_side, stimuli, actions, prob_left = utils.format_data(
pseudo_trials)
p_priors = model.compute_prior(
np.array(actions),
np.array(stimuli),
np.array(stim_side),
parameter_type='posterior_mean')[0]
# Decode prior
p_pred_prior, p_pred_prior_train = regress(
population_activity,