def uncertainty_vs_last_rewarded(f_behavior, max_duration=5000):
##start by getting the trial data for this session
trial_data = ptr.get_full_trials(f_behavior, max_duration=max_duration)
##now determine which trials were rewarded or unrewarded
last_rew_idx, last_unrew_idx = ptr.split_by_last_outcome(trial_data)
##compute the hidden markov model
model_data = mf.fit_models_from_trial_data(trial_data)
##return the confidence, belief states
uncertainty = mf.uncertainty_from_trial_data(trial_data)
##split into last rewarded and last unrewarded belief states
last_rew_uncertainty = uncertainty[last_rew_idx]
last_unrew_uncertainty = uncertainty[last_unrew_idx]
return last_rew_uncertainty, last_unrew_uncertainty
def get_behavior_data(f_behavior, n_back=3, max_duration=5000):
global trial_lut
##start by parsing the trials
trial_data = ptr.get_full_trials(f_behavior, max_duration=max_duration)
##get the session number for this session
session_num = ptr.get_session_number(f_behavior)
##pre-allocation
n_trials = len(trial_data.index)
##how many features we have depends on the length of our
##action/reward history
features = ['training_day', 'trial_number']
for i in range(n_back):
features.append('action-' + str(i + 1))
features.append('outcome-' + str(i + 1))
features.append('interaction-' + str(i + 1))
##create the data arrays
y = pd.DataFrame(index=np.arange(n_trials), columns=['value', 'action'])
X = pd.DataFrame(index=np.arange(n_trials), columns=features)
"""
Now parse each trial using the following values:
reward/no reward: 1 or 0
upper lever/lower lever: 2 or 1
"""
for t in range(n_trials):
##get the trial data for this trial
trial = trial_data.loc[t]
##fill out the outcomes array first
y['value'][t] = trial_lut[trial['action']]
y['action'][t] = trial['action']
X['training_day'][t] = session_num
X['trial_number'][t] = t
for i in range(n_back):
if t > n_back:
X['action-' +
str(i + 1)][t] = trial_lut[trial_data['action'][t - (i + 1)]]
X['outcome-' +
str(i + 1)][t] = trial_lut[trial_data['outcome'][t -
(i + 1)]]
X['interaction-' +
str(i + 1)][t] = trial_lut[trial_data['action'][t - (
i + 1)]] * trial_lut[trial_data['outcome'][t - (i + 1)]]
else:
X['action-' + str(i + 1)][t] = 0
X['outcome-' + str(i + 1)][t] = 0
X['interaction-' + str(i + 1)][t] = 0
return y, X
def get_session_data(f_behavior, max_duration=5000):
trial_data = ptr.get_full_trials(f_behavior,
pad=[400, 400],
max_duration=max_duration)
n_trials = trial_data.index.size
actions = np.zeros(n_trials)
outcomes = np.zeros(n_trials)
first_block = trial_data['context'][0]
upper_levers = np.where(trial_data['action'] == 'upper_lever')[0]
lower_levers = np.where(trial_data['action'] == 'lower_lever')[0]
rewarded = np.where(trial_data['outcome'] == 'rewarded_poke')[0]
unrewarded = np.where(trial_data['outcome'] == 'unrewarded_poke')[0]
actions[upper_levers] = 2
actions[lower_levers] = 1
outcomes[rewarded] = 1
outcomes[unrewarded] = 0
ctx = np.asarray(trial_data['context'] == 'upper_rewarded').astype(int)
switch_times = np.where(np.diff(ctx) != 0)[0]
return actions, outcomes, switch_times, first_block
def get_session_meta(f_behavior, max_duration=5000):
##start by parsing the data
data = ptr.get_full_trials(f_behavior, max_duration=max_duration)
metadata = {
'unrewarded': np.where(data['outcome'] == 'unrewarded_poke')[0],
'rewarded': np.where(data['outcome'] == 'rewarded_poke')[0],
'upper_lever': np.where(data['action'] == 'upper_lever')[0],
'lower_lever': np.where(data['action'] == 'lower_lever')[0],
'upper_context': np.where(data['context'] == 'upper_rewarded')[0],
'lower_context': np.where(data['context'] == 'lower_rewarded')[0],
}
trial_info = ptr.parse_trial_data(data)
metadata['n_blocks'] = trial_info['n_blocks']
metadata['block_lengths'] = trial_info['block_lengths']
metadata['first_block'] = data['context'][0]
metadata['mean_block_len'] = np.mean(trial_info['block_lengths'])
# metadata['reward_rate'] = np.mean([len(trial_info['upper_correct_rewarded'])/(
# len(trial_info['upper_correct_rewarded'])+len(trial_info['upper_correct_unrewarded'])),
# len(trial_info['lower_correct_rewarded'])/(len(trial_info['lower_correct_rewarded'])+len(
# trial_info['lower_correct_unrewarded']))])
return metadata
def session_trial_durations(f_behavior, max_duration=5000):
data = ptr.get_full_trials(f_behavior, max_duration=max_duration)
return np.asarray(data['outcome_ts'] - data['action_ts']).astype(int)