def save_task_data(data_loc, data):
path = os.path.join(data_loc,'Individual_Measures')
if not os.path.exists(path):
os.makedirs(path)
for exp_id in np.sort(data.experiment_exp_id.unique()):
print('Saving %s...' % exp_id)
extract_experiment(data,exp_id).to_csv(os.path.join(path, exp_id + '.csv.gz'), compression = 'gzip')
def save_task_data(data_loc, data):
path = os.path.join(data_loc, 'Individual_Measures')
if not os.path.exists(path):
os.makedirs(path)
for exp_id in np.sort(data.experiment_exp_id.unique()):
print('Saving %s...' % exp_id)
extract_experiment(data,
exp_id).to_csv(os.path.join(path,
exp_id + '.csv.gz'),
compression='gzip')
def get_items(data):
excluded_surveys = ['holt_laury_survey']
items = []
responses = []
responses_text = []
options = []
workers = []
item_nums = []
exps = []
for exp in data.experiment_exp_id.unique():
if 'survey' in exp and exp not in excluded_surveys:
survey = extract_experiment(data,exp)
try:
responses += list(survey.response.map(lambda x: float(x)))
except ValueError:
continue
items += list(survey.text)
responses_text += [str(i) for i in list(survey.response_text)]
options += list(survey.options)
workers += list(survey.worker_id)
item_nums += list(survey.question_num)
exps += [exp] * len(survey.text)
items_df = pd.DataFrame({'survey': exps, 'worker': workers, 'item_text': items, 'coded_response': responses,
'response_text': responses_text, 'options': options}, dtype = float)
items_df.loc[:,'item_num'] = [str(i).zfill(2) for i in item_nums]
items_df.loc[:,'item_ID'] = items_df['survey'] + '.' + items_df['item_num'].astype(str)
items_df=items_df[['worker','item_ID','coded_response','item_text','response_text','options','survey','item_num']]
return items_df
def quality_check_correction(data):
"""
This function corrects the issues with the stop signal tasks mentioned above
"""
for exp in [
'stop_signal', 'motor_selective_stop_signal',
'stim_selective_stop_signal'
]:
df = extract_experiment(data, exp)
rt_thresh = 200
acc_thresh = .6
missed_thresh = .25
response_thresh = .95
passed_rt = df.query('rt != -1 and SS_trial_type=="go"').groupby(
'worker_id').rt.median() >= rt_thresh
passed_miss = df.query('SS_trial_type=="go"').groupby(
'worker_id').rt.agg(lambda x: np.mean(x == -1)) < missed_thresh
passed_acc = df.query('rt != -1').groupby(
'worker_id').correct.mean() >= acc_thresh
passed_response = np.logical_not(
df.query('rt != -1').groupby('worker_id').key_press.agg(
lambda x: np.any(
pd.value_counts(x) > pd.value_counts(x).sum() *
response_thresh)))
passed_df = pd.concat(
[passed_rt, passed_acc, passed_miss, passed_response],
axis=1).fillna(False, inplace=False)
passed = passed_df.all(axis=1)
failed = passed[passed == False]
for subj in failed.index:
data.loc[(data.experiment_exp_id == exp) &
(data.worker_id == subj), 'passed_QC'] = False
for subj in passed.index:
data.loc[(data.experiment_exp_id == exp) &
(data.worker_id == subj), 'passed_QC'] = True
def get_average_variable(results, var):
'''Prints time taken for each experiment in minutes
'''
averages = {}
for exp in results.get_experiments():
data = extract_experiment(results,exp)
try:
average = data[var].mean()
except TypeError:
print("Cannot average %s" % (var))
averages[exp] = average
return averages
def get_average_variable(results, var):
'''Prints time taken for each experiment in minutes
'''
averages = {}
for exp in results.get_experiments():
data = extract_experiment(results, exp)
try:
average = data[var].mean()
except TypeError:
print("Cannot average %s" % (var))
averages[exp] = average
return averages
def get_items(data):
excluded_surveys = ['holt_laury_survey']
items = []
responses = []
responses_text = []
options = []
workers = []
item_nums = []
exps = []
for exp in data.experiment_exp_id.unique():
if 'survey' in exp and exp not in excluded_surveys:
survey = extract_experiment(data, exp)
try:
responses += list(survey.response.map(lambda x: float(x)))
except ValueError:
continue
items += list(survey.text)
responses_text += [str(i) for i in list(survey.response_text)]
options += list(survey.options)
workers += list(survey.worker_id)
item_nums += list(survey.question_num)
exps += [exp] * len(survey.text)
items_df = pd.DataFrame(
{
'survey': exps,
'worker': workers,
'item_text': items,
'coded_response': responses,
'response_text': responses_text,
'options': options
},
dtype=float)
items_df.loc[:, 'item_num'] = [str(i).zfill(2) for i in item_nums]
items_df.loc[:, 'item_ID'] = items_df['survey'] + '.' + items_df[
'item_num'].astype(str)
items_df = items_df[[
'worker', 'item_ID', 'coded_response', 'item_text', 'response_text',
'options', 'survey', 'item_num'
]]
return items_df
def quality_check_correction(data):
"""
This function corrects the issues with the stop signal tasks mentioned above
"""
for exp in ['stop_signal','motor_selective_stop_signal',
'stim_selective_stop_signal']:
df = extract_experiment(data, exp)
rt_thresh = 200
acc_thresh = .6
missed_thresh = .25
response_thresh = .95
passed_rt = df.query('rt != -1 and SS_trial_type=="go"').groupby('worker_id').rt.median() >= rt_thresh
passed_miss = df.query('SS_trial_type=="go"').groupby('worker_id').rt.agg(lambda x: np.mean(x == -1)) < missed_thresh
passed_acc = df.query('rt != -1').groupby('worker_id').correct.mean() >= acc_thresh
passed_response = np.logical_not(df.query('rt != -1').groupby('worker_id').key_press.agg(
lambda x: np.any(pd.value_counts(x) > pd.value_counts(x).sum()*response_thresh)))
passed_df = pd.concat([passed_rt,passed_acc,passed_miss,passed_response], axis = 1).fillna(False, inplace = False)
passed = passed_df.all(axis = 1)
failed = passed[passed == False]
for subj in failed.index:
data.loc[(data.experiment_exp_id == exp) & (data.worker_id == subj),'passed_QC'] = False
for subj in passed.index:
data.loc[(data.experiment_exp_id == exp) & (data.worker_id == subj),'passed_QC'] = True
def results_check(data, exp_id = None, worker = None, columns = ['correct', 'rt'], remove_practice = True, use_groups = True, plot = False, silent = False):
"""Outputs info for a basic data check on the results object. Uses data_check to group, describe and plot
dataframes. Function first filters the results object as specified,
loops through each experiment and worker contained in the results object, performs some basic dataframe manipulation
and runs data_check
:data: the data from an expanalysis Result object
:param experiment: a string or array of strings to select the experiment(s) before calculating basic stats
:param worker: a string or array of strings to select the worker(s) before calculating basic stats
:param columns: array of columns to subset summary statistics, if they exist
:param remove_practice: bool, default True. If True will remove any rows labeled "practice" in the "exp_stage" column, if it exists
:param use_groups: bool, default True. If True will lookup grouping variables using get_groupby for the experiment
:param silent: bool, default False. If True will not print output
:param plot: bool, default False: If True plots data using plot_groups
:return summary, p: summary data frame and plot object
"""
assert 'worker_id' in data.columns and 'experiment_exp_id' in data.columns, \
"Results data must have 'worker_id' and 'experiment_exp_id' in columns"
stats = {}
results = result_filter(data, exp_id = exp_id, worker = worker)
orig_plot = plot
orig_silent = silent
display = not silent or plot
if display:
print('******************************************************************************')
print('Input: Type "exit" to end, "skip" to skip to the next experiment, or hit enter to continue')
print('******************************************************************************')
for experiment in numpy.unique(results['experiment_exp_id']):
stats[experiment] = {}
if display:
print('******************************************************************************')
print(' Experiment: ', experiment)
print('******************************************************************************')
if use_groups:
groupby = get_groupby(experiment)
else:
groupby = []
experiment_df = extract_experiment(results, experiment)
for worker in pandas.unique(experiment_df['worker_id']):
if display:
print('******************************************************************************')
print(' Worker: ', worker)
print('******************************************************************************')
df = experiment_df.query('worker_id == "%s"' % worker)
summary, p = data_check(df, columns, remove_practice, groupby, silent, plot)
#add summary and plot to dictionary of summaries
stats[experiment]= {worker: {'summary': summary, 'plot': p}}
if not silent or plot:
input_text = input("Press Enter to continue...")
plt.close()
if input_text in ['skip', 'save']:
plot = False
silent = True
display = not silent or plot
elif input_text == 'exit':
break
if display:
if input_text not in ['exit', 'save']:
plot = orig_plot
silent = orig_silent
display = not silent or plot
elif input_text == 'exit':
break
return stats
def quality_check(data):
"""
Checks data to make sure each experiment passed some "gut check" measures
Used to exclude data on individual tasks or whole subjects if they fail
too many tasks.
NOTE: This function has an issue such that it inappropriately evaluates
stop signal tasks based on the number of missed responses. Rather than
changing the function (which would affect our samples which are already
determined) I am leaving it, and introducing a quality check correction
that will be performed after subjects are already rejected
"""
start_time = time()
rt_thresh_lookup = {
'angling_risk_task_always_sunny': 0,
'simple_reaction_time': 150
}
acc_thresh_lookup = {
'digit_span': 0,
'hierarchical_rule': 0,
'information_sampling_task': 0,
'probabilistic_selection': 0,
'ravens': 0,
'shift_task': 0,
'spatial_span': 0,
'tower_of_london': 0
}
missed_thresh_lookup = {
'information_sampling_task': 1,
'go_nogo': 1,
'tower_of_london': 2
}
response_thresh_lookup = {
'angling_risk_task_always_sunny': np.nan,
'columbia_card_task_cold': np.nan,
'discount_titrate': np.nan,
'digit_span': np.nan,
'go_nogo': .98,
'kirby': np.nan,
'simple_reaction_time': np.nan,
'spatial_span': np.nan,
}
templates = data.groupby('experiment_exp_id').experiment_template.unique()
data.loc[:,'passed_QC'] = True
for exp in data.experiment_exp_id.unique():
try:
if templates.loc[exp] == 'jspsych':
print('Running QC on ' + exp)
df = extract_experiment(data, exp)
rt_thresh = rt_thresh_lookup.get(exp,200)
acc_thresh = acc_thresh_lookup.get(exp,.6)
missed_thresh = missed_thresh_lookup.get(exp,.25)
response_thresh = response_thresh_lookup.get(exp,.95)
# special cases...
if exp == 'information_sampling_task':
df.groupby('worker_id').which_click_in_round.value_counts()
passed_response = df.groupby('worker_id').which_click_in_round.mean() > 2
passed_rt = pd.Series([True] * len(passed_response), index = passed_response.index)
passed_miss = pd.Series([True] * len(passed_response), index = passed_response.index)
passed_acc = pd.Series([True] * len(passed_response), index = passed_response.index)
elif exp == 'go_nogo':
passed_rt = df.query('rt != -1').groupby('worker_id').rt.median() >= rt_thresh
passed_miss = df.groupby('worker_id').rt.agg(lambda x: np.mean(x == -1)) < missed_thresh
df.correct = pd.to_numeric(df.correct)
passed_acc = df.groupby('worker_id').correct.mean() >= acc_thresh
passed_response = np.logical_not(df.groupby('worker_id').key_press.agg(
lambda x: np.any(pd.value_counts(x) > pd.value_counts(x).sum()*response_thresh)))
elif exp == 'psychological_refractory_period_two_choices':
passed_rt = (df.groupby('worker_id').median()[['choice1_rt','choice2_rt']] >= rt_thresh).all(axis = 1)
passed_acc = df.query('choice1_rt != -1').groupby('worker_id').choice1_correct.mean() >= acc_thresh
passed_miss = ((df.groupby('worker_id').choice1_rt.agg(lambda x: np.mean(x!=-1) >= missed_thresh)) \
+ (df.groupby('worker_id').choice2_rt.agg(lambda x: np.mean(x>-1) >= missed_thresh))) == 2
passed_response1 = np.logical_not(df.query('choice1_rt != -1').groupby('worker_id').choice1_key_press.agg(
lambda x: np.any(pd.value_counts(x) > pd.value_counts(x).sum()*response_thresh)))
passed_response2 = np.logical_not(df.query('choice2_rt != -1').groupby('worker_id').choice2_key_press.agg(
lambda x: np.any(pd.value_counts(x) > pd.value_counts(x).sum()*response_thresh)))
passed_response = np.logical_and(passed_response1,passed_response2)
elif exp == 'ravens':
passed_rt = df.query('rt != -1').groupby('worker_id').rt.median() >= rt_thresh
passed_acc = df.query('rt != -1').groupby('worker_id').correct.mean() >= acc_thresh
passed_response = np.logical_not(df.groupby('worker_id').stim_response.agg(
lambda x: np.any(pd.value_counts(x) > pd.value_counts(x).sum()*response_thresh)))
passed_miss = pd.Series([True] * len(passed_rt), index = passed_rt.index)
elif exp == 'tower_of_london':
passed_rt = df.groupby('worker_id').rt.median() >= rt_thresh
passed_acc = df.query('trial_id == "feedback"').groupby('worker_id').correct.mean() >= acc_thresh
# Labeling someone as "missing" too many problems if they don't make enough moves
passed_miss = (df.groupby(['worker_id','problem_id']).num_moves_made.max().reset_index().groupby('worker_id').mean() >= missed_thresh).num_moves_made
passed_response = pd.Series([True] * len(passed_rt), index = passed_rt.index)
elif exp == 'two_stage_decision':
passed_rt = (df.groupby('worker_id').median()[['rt_first','rt_second']] >= rt_thresh).all(axis = 1)
passed_miss = df.groupby('worker_id').trial_id.agg(lambda x: np.mean(x == 'incomplete_trial')) < missed_thresh
passed_acc = pd.Series([True] * len(passed_rt), index = passed_rt.index)
passed_response = pd.Series([True] * len(passed_rt), index = passed_rt.index)
passed_response1 = np.logical_not(df.query('rt_first != -1').groupby('worker_id').key_press_first.agg(
lambda x: np.any(pd.value_counts(x) > pd.value_counts(x).sum()*response_thresh)))
passed_response2 = np.logical_not(df.query('rt_second != -1').groupby('worker_id').key_press_second.agg(
lambda x: np.any(pd.value_counts(x) > pd.value_counts(x).sum()*response_thresh)))
passed_response = np.logical_and(passed_response1,passed_response2)
elif exp == 'writing_task':
passed_response = df.query('trial_id == "write"').groupby('worker_id').final_text.agg(lambda x: len(x[0]) > 100)
passed_acc = pd.Series([True] * len(passed_response), index = passed_response.index)
passed_rt = pd.Series([True] * len(passed_response), index = passed_response.index)
passed_miss = pd.Series([True] * len(passed_response), index = passed_response.index)
# everything else
else:
passed_rt = df.query('rt != -1').groupby('worker_id').rt.median() >= rt_thresh
passed_miss = df.groupby('worker_id').rt.agg(lambda x: np.mean(x == -1)) < missed_thresh
if 'correct' in df.columns:
df.correct = pd.to_numeric(df.correct)
passed_acc = df.query('rt != -1').groupby('worker_id').correct.mean() >= acc_thresh
else:
passed_acc = pd.Series([True] * len(passed_rt), index = passed_rt.index)
if 'mouse_click' in df.columns:
passed_response = np.logical_not(df.query('rt != -1').groupby('worker_id').mouse_click.agg(
lambda x: np.any(pd.value_counts(x) > pd.value_counts(x).sum()*response_thresh)))
elif 'key_press' in df.columns:
passed_response = np.logical_not(df.query('rt != -1').groupby('worker_id').key_press.agg(
lambda x: np.any(pd.value_counts(x) > pd.value_counts(x).sum()*response_thresh)))
passed_df = pd.concat([passed_rt,passed_acc,passed_miss,passed_response], axis = 1).fillna(False, inplace = False)
passed = passed_df.all(axis = 1)
failed = passed[passed == False]
for subj in failed.index:
data.loc[(data.experiment_exp_id == exp) & (data.worker_id == subj),'passed_QC'] = False
except AttributeError as e:
print('QC could not be run on experiment %s' % exp)
print(e)
finish_time = (time() - start_time)/60
print('Finished QC. Time taken: ' + str(finish_time))
DVs_valence = []
datasets.append((data,directory, DVs, DVs_valence))
# calculate DVs
for data,directory, DV_df, valence_df in datasets:
readme_lines = []
meta_dir = path.join(directory,'metadata')
reference_dir = path.join(directory,'references')
if not path.exists(meta_dir):
makedirs(meta_dir)
if not path.exists(reference_dir):
makedirs(reference_dir)
# save target datasets
print('Saving to %s...' % directory)
print('Saving target measures...')
demog_data = extract_experiment(data,'demographics_survey')
demog_data = process_demographics(demog_data, directory, meta_dir)
alcohol_drug_data = extract_experiment(data,'alcohol_drugs_survey')
alcohol_drug_data = process_alcohol_drug(alcohol_drug_data, directory, meta_dir)
health_data = extract_experiment(data,'k6_survey')
health_data = process_health(health_data, directory, meta_dir)
activity_level = DV_df.pop('leisure_time_activity_survey.activity_level')
# concatenate targets
target_data = pd.concat([demog_data, alcohol_drug_data,
health_data, activity_level], axis = 1)
target_data.to_csv(path.join(directory,'demographic_health.csv'))
# save items
items_df = get_items(data)
print('Saving items...')
subjectsxitems = items_df.pivot('worker','item_ID','coded_response')
# ensure there are the correct number of items
def quality_check(data):
"""
Checks data to make sure each experiment passed some "gut check" measures
Used to exclude data on individual tasks or whole subjects if they fail
too many tasks.
NOTE: This function has an issue such that it inappropriately evaluates
stop signal tasks based on the number of missed responses. Rather than
changing the function (which would affect our samples which are already
determined) I am leaving it, and introducing a quality check correction
that will be performed after subjects are already rejected
"""
start_time = time()
rt_thresh_lookup = {
'angling_risk_task_always_sunny': 0,
'simple_reaction_time': 150
}
acc_thresh_lookup = {
'digit_span': 0,
'hierarchical_rule': 0,
'information_sampling_task': 0,
'probabilistic_selection': 0,
'ravens': 0,
'shift_task': 0,
'spatial_span': 0,
'tower_of_london': 0
}
missed_thresh_lookup = {
'information_sampling_task': 1,
'go_nogo': 1,
'tower_of_london': 2
}
response_thresh_lookup = {
'angling_risk_task_always_sunny': np.nan,
'columbia_card_task_cold': np.nan,
'discount_titrate': np.nan,
'digit_span': np.nan,
'go_nogo': .98,
'kirby': np.nan,
'simple_reaction_time': np.nan,
'spatial_span': np.nan,
}
templates = data.groupby('experiment_exp_id').experiment_template.unique()
data.loc[:, 'passed_QC'] = True
for exp in data.experiment_exp_id.unique():
try:
if templates.loc[exp] == 'jspsych':
print('Running QC on ' + exp)
df = extract_experiment(data, exp)
rt_thresh = rt_thresh_lookup.get(exp, 200)
acc_thresh = acc_thresh_lookup.get(exp, .6)
missed_thresh = missed_thresh_lookup.get(exp, .25)
response_thresh = response_thresh_lookup.get(exp, .95)
# special cases...
if exp == 'information_sampling_task':
df.groupby('worker_id').which_click_in_round.value_counts()
passed_response = df.groupby(
'worker_id').which_click_in_round.mean() > 2
passed_rt = pd.Series([True] * len(passed_response),
index=passed_response.index)
passed_miss = pd.Series([True] * len(passed_response),
index=passed_response.index)
passed_acc = pd.Series([True] * len(passed_response),
index=passed_response.index)
elif exp == 'go_nogo':
passed_rt = df.query('rt != -1').groupby(
'worker_id').rt.median() >= rt_thresh
passed_miss = df.groupby('worker_id').rt.agg(
lambda x: np.mean(x == -1)) < missed_thresh
df.correct = pd.to_numeric(df.correct)
passed_acc = df.groupby(
'worker_id').correct.mean() >= acc_thresh
passed_response = np.logical_not(
df.groupby('worker_id').key_press.agg(lambda x: np.any(
pd.value_counts(x) > pd.value_counts(x).sum() *
response_thresh)))
elif exp == 'psychological_refractory_period_two_choices':
passed_rt = (df.groupby('worker_id').median()[[
'choice1_rt', 'choice2_rt'
]] >= rt_thresh).all(axis=1)
passed_acc = df.query('choice1_rt != -1').groupby(
'worker_id').choice1_correct.mean() >= acc_thresh
passed_miss = ((df.groupby('worker_id').choice1_rt.agg(lambda x: np.mean(x!=-1) >= missed_thresh)) \
+ (df.groupby('worker_id').choice2_rt.agg(lambda x: np.mean(x>-1) >= missed_thresh))) == 2
passed_response1 = np.logical_not(
df.query('choice1_rt != -1').groupby('worker_id').
choice1_key_press.agg(lambda x: np.any(
pd.value_counts(x) > pd.value_counts(x).sum() *
response_thresh)))
passed_response2 = np.logical_not(
df.query('choice2_rt != -1').groupby('worker_id').
choice2_key_press.agg(lambda x: np.any(
pd.value_counts(x) > pd.value_counts(x).sum() *
response_thresh)))
passed_response = np.logical_and(passed_response1,
passed_response2)
elif exp == 'ravens':
passed_rt = df.query('rt != -1').groupby(
'worker_id').rt.median() >= rt_thresh
passed_acc = df.query('rt != -1').groupby(
'worker_id').correct.mean() >= acc_thresh
passed_response = np.logical_not(
df.groupby('worker_id').stim_response.agg(
lambda x: np.any(
pd.value_counts(x) > pd.value_counts(x).sum() *
response_thresh)))
passed_miss = pd.Series([True] * len(passed_rt),
index=passed_rt.index)
elif exp == 'tower_of_london':
passed_rt = df.groupby(
'worker_id').rt.median() >= rt_thresh
passed_acc = df.query('trial_id == "feedback"').groupby(
'worker_id').correct.mean() >= acc_thresh
# Labeling someone as "missing" too many problems if they don't make enough moves
passed_miss = (df.groupby([
'worker_id', 'problem_id'
]).num_moves_made.max().reset_index().groupby(
'worker_id').mean() >= missed_thresh).num_moves_made
passed_response = pd.Series([True] * len(passed_rt),
index=passed_rt.index)
elif exp == 'two_stage_decision':
passed_rt = (df.groupby('worker_id').median()[[
'rt_first', 'rt_second'
]] >= rt_thresh).all(axis=1)
passed_miss = df.groupby('worker_id').trial_id.agg(
lambda x: np.mean(x == 'incomplete_trial'
)) < missed_thresh
passed_acc = pd.Series([True] * len(passed_rt),
index=passed_rt.index)
passed_response = pd.Series([True] * len(passed_rt),
index=passed_rt.index)
passed_response1 = np.logical_not(
df.query('rt_first != -1').groupby(
'worker_id').key_press_first.agg(lambda x: np.any(
pd.value_counts(x) > pd.value_counts(x).sum() *
response_thresh)))
passed_response2 = np.logical_not(
df.query('rt_second != -1').groupby(
'worker_id').key_press_second.agg(lambda x: np.any(
pd.value_counts(x) > pd.value_counts(x).sum() *
response_thresh)))
passed_response = np.logical_and(passed_response1,
passed_response2)
elif exp == 'writing_task':
passed_response = df.query('trial_id == "write"').groupby(
'worker_id').final_text.agg(lambda x: len(x[0]) > 100)
passed_acc = pd.Series([True] * len(passed_response),
index=passed_response.index)
passed_rt = pd.Series([True] * len(passed_response),
index=passed_response.index)
passed_miss = pd.Series([True] * len(passed_response),
index=passed_response.index)
# everything else
else:
passed_rt = df.query('rt != -1').groupby(
'worker_id').rt.median() >= rt_thresh
passed_miss = df.groupby('worker_id').rt.agg(
lambda x: np.mean(x == -1)) < missed_thresh
if 'correct' in df.columns:
df.correct = pd.to_numeric(df.correct)
passed_acc = df.query('rt != -1').groupby(
'worker_id').correct.mean() >= acc_thresh
else:
passed_acc = pd.Series([True] * len(passed_rt),
index=passed_rt.index)
if 'mouse_click' in df.columns:
passed_response = np.logical_not(
df.query('rt != -1').groupby(
'worker_id').mouse_click.agg(lambda x: np.any(
pd.value_counts(x) > pd.value_counts(x).
sum() * response_thresh)))
elif 'key_press' in df.columns:
passed_response = np.logical_not(
df.query('rt != -1').groupby(
'worker_id').key_press.agg(lambda x: np.any(
pd.value_counts(x) > pd.value_counts(x).
sum() * response_thresh)))
passed_df = pd.concat(
[passed_rt, passed_acc, passed_miss, passed_response],
axis=1).fillna(False, inplace=False)
passed = passed_df.all(axis=1)
failed = passed[passed == False]
for subj in failed.index:
data.loc[(data.experiment_exp_id == exp) &
(data.worker_id == subj), 'passed_QC'] = False
except AttributeError as e:
print('QC could not be run on experiment %s' % exp)
print(e)
finish_time = (time() - start_time) / 60
print('Finished QC. Time taken: ' + str(finish_time))
def results_check(data,
exp_id=None,
worker=None,
columns=['correct', 'rt'],
remove_practice=True,
use_groups=True,
plot=False,
silent=False):
"""Outputs info for a basic data check on the results object. Uses data_check to group, describe and plot
dataframes. Function first filters the results object as specified,
loops through each experiment and worker contained in the results object, performs some basic dataframe manipulation
and runs data_check
:data: the data from an expanalysis Result object
:param experiment: a string or array of strings to select the experiment(s) before calculating basic stats
:param worker: a string or array of strings to select the worker(s) before calculating basic stats
:param columns: array of columns to subset summary statistics, if they exist
:param remove_practice: bool, default True. If True will remove any rows labeled "practice" in the "exp_stage" column, if it exists
:param use_groups: bool, default True. If True will lookup grouping variables using get_groupby for the experiment
:param silent: bool, default False. If True will not print output
:param plot: bool, default False: If True plots data using plot_groups
:return summary, p: summary data frame and plot object
"""
assert 'worker_id' in data.columns and 'experiment_exp_id' in data.columns, \
"Results data must have 'worker_id' and 'experiment_exp_id' in columns"
stats = {}
results = result_filter(data, exp_id=exp_id, worker=worker)
orig_plot = plot
orig_silent = silent
display = not silent or plot
if display:
print(
'******************************************************************************'
)
print(
'Input: Type "exit" to end, "skip" to skip to the next experiment, or hit enter to continue'
)
print(
'******************************************************************************'
)
for experiment in numpy.unique(results['experiment_exp_id']):
stats[experiment] = {}
if display:
print(
'******************************************************************************'
)
print(' Experiment: ', experiment)
print(
'******************************************************************************'
)
if use_groups:
groupby = get_groupby(experiment)
else:
groupby = []
experiment_df = extract_experiment(results, experiment)
for worker in pandas.unique(experiment_df['worker_id']):
if display:
print(
'******************************************************************************'
)
print(' Worker: ', worker)
print(
'******************************************************************************'
)
df = experiment_df.query('worker_id == "%s"' % worker)
summary, p = data_check(df, columns, remove_practice, groupby,
silent, plot)
#add summary and plot to dictionary of summaries
stats[experiment] = {worker: {'summary': summary, 'plot': p}}
if not silent or plot:
input_text = input("Press Enter to continue...")
plt.close()
if input_text in ['skip', 'save']:
plot = False
silent = True
display = not silent or plot
elif input_text == 'exit':
break
if display:
if input_text not in ['exit', 'save']:
plot = orig_plot
silent = orig_silent
display = not silent or plot
elif input_text == 'exit':
break
return stats
DVs_valence = []
datasets.append((data,directory, DVs, DVs_valence))
# calculate DVs
for data,directory, DV_df, valence_df in datasets:
readme_lines = []
meta_dir = path.join(directory,'metadata')
reference_dir = path.join(directory,'references')
if not path.exists(meta_dir):
makedirs(meta_dir)
if not path.exists(reference_dir):
makedirs(reference_dir)
# save target datasets
print('Saving to %s...' % directory)
print('Saving target measures...')
demog_data = extract_experiment(data,'demographics_survey')
demog_data = process_demographics(demog_data, directory, meta_dir)
alcohol_drug_data = extract_experiment(data,'alcohol_drugs_survey')
alcohol_drug_data = process_alcohol_drug(alcohol_drug_data, directory, meta_dir)
health_data = extract_experiment(data,'k6_survey')
health_data = process_health(health_data, directory, meta_dir)
activity_level = DV_df.pop('leisure_time_activity_survey.activity_level')
# concatenate targets
target_data = pd.concat([demog_data, alcohol_drug_data,
health_data, activity_level], axis = 1)
target_data.to_csv(path.join(directory,'demographic_health.csv'))
# save items
items_df = get_items(data)
print('Saving items...')
subjectsxitems = items_df.pivot('worker','item_ID','coded_response')
# ensure there are the correct number of items
