def form_string_all_trials_perf(self, translated_trial_matrix):
"""Form a string with side perf and anova for all trials"""
side2perf_all = count_hits_by_type_from_trials_info(
translated_trial_matrix, split_key='rewside')
string_perf_by_side = self.form_string_perf_by_side(side2perf_all)
if len(translated_trial_matrix
) > self.cached_anova_len2 or self.cached_anova_text2 == '':
numericated_trial_matrix = TrialMatrix.numericate_trial_matrix(
translated_trial_matrix)
anova_stats = TrialMatrix.run_anova(numericated_trial_matrix)
self.cached_anova_text2 = anova_stats
self.cached_anova_len2 = len(translated_trial_matrix)
else:
anova_stats = self.cached_anova_text2
return 'All: ' + string_perf_by_side + '. Biases: ' + anova_stats
def form_string_all_trials_perf(self, translated_trial_matrix):
"""Form a string with side perf and anova for all trials"""
side2perf_all = count_hits_by_type_from_trials_info(
translated_trial_matrix,
split_key='rewside')
string_perf_by_side = self.form_string_perf_by_side(side2perf_all)
if len(translated_trial_matrix) > self.cached_anova_len2 or self.cached_anova_text2 == '':
numericated_trial_matrix = TrialMatrix.numericate_trial_matrix(
translated_trial_matrix)
anova_stats = TrialMatrix.run_anova(numericated_trial_matrix)
self.cached_anova_text2 = anova_stats
self.cached_anova_len2 = len(translated_trial_matrix)
else:
anova_stats = self.cached_anova_text2
return 'All: ' + string_perf_by_side + '. Biases: ' + anova_stats
def form_string_recent_trials_perf(self, translated_trial_matrix):
"""Form a string with side perf and anova for recent trials
cached in cached_anova_text3 and cached_anova_len3
"""
side2perf = count_hits_by_type_from_trials_info(
translated_trial_matrix.iloc[-60:], split_key='rewside')
string_perf_by_side = self.form_string_perf_by_side(side2perf)
if len(translated_trial_matrix) > self.cached_anova_len3 or self.cached_anova_text3 == '':
numericated_trial_matrix = TrialMatrix.numericate_trial_matrix(
translated_trial_matrix.iloc[-60:])
anova_stats = TrialMatrix.run_anova(numericated_trial_matrix)
self.cached_anova_text3 = anova_stats
self.cached_anova_len3 = len(translated_trial_matrix)
else:
anova_stats = self.cached_anova_text3
return 'Recent: ' + string_perf_by_side + '. Biases: ' + anova_stats
def form_string_recent_trials_perf(self, translated_trial_matrix):
"""Form a string with side perf and anova for recent trials
cached in cached_anova_text3 and cached_anova_len3
"""
side2perf = count_hits_by_type_from_trials_info(
translated_trial_matrix.iloc[-60:], split_key='rewside')
string_perf_by_side = self.form_string_perf_by_side(side2perf)
if len(translated_trial_matrix
) > self.cached_anova_len3 or self.cached_anova_text3 == '':
numericated_trial_matrix = TrialMatrix.numericate_trial_matrix(
translated_trial_matrix.iloc[-60:])
anova_stats = TrialMatrix.run_anova(numericated_trial_matrix)
self.cached_anova_text3 = anova_stats
self.cached_anova_len3 = len(translated_trial_matrix)
else:
anova_stats = self.cached_anova_text3
return 'Recent: ' + string_perf_by_side + '. Biases: ' + anova_stats
def form_string_unforced_trials_perf(self, translated_trial_matrix):
"""Exactly the same as form_string_all_trials_perf, except that:
We drop all trials where bad is True.
We use cached_anova_len1 and cached_anova_text1 instead of 2.
"""
side2perf = count_hits_by_type_from_trials_info(
translated_trial_matrix[~translated_trial_matrix.bad],
split_key='rewside')
string_perf_by_side = self.form_string_perf_by_side(side2perf)
if len(translated_trial_matrix) > self.cached_anova_len1 or self.cached_anova_text1 == '':
numericated_trial_matrix = TrialMatrix.numericate_trial_matrix(
translated_trial_matrix[~translated_trial_matrix.bad])
anova_stats = TrialMatrix.run_anova(numericated_trial_matrix)
self.cached_anova_text1 = anova_stats
self.cached_anova_len1 = len(translated_trial_matrix)
else:
anova_stats = self.cached_anova_text1
return 'UF: ' + string_perf_by_side + '. Biases: ' + anova_stats
def form_string_unforced_trials_perf(self, translated_trial_matrix):
"""Exactly the same as form_string_all_trials_perf, except that:
We drop all trials where bad is True.
We use cached_anova_len1 and cached_anova_text1 instead of 2.
"""
side2perf = count_hits_by_type_from_trials_info(
translated_trial_matrix[~translated_trial_matrix.bad],
split_key='rewside')
string_perf_by_side = self.form_string_perf_by_side(side2perf)
if len(translated_trial_matrix
) > self.cached_anova_len1 or self.cached_anova_text1 == '':
numericated_trial_matrix = TrialMatrix.numericate_trial_matrix(
translated_trial_matrix[~translated_trial_matrix.bad])
anova_stats = TrialMatrix.run_anova(numericated_trial_matrix)
self.cached_anova_text1 = anova_stats
self.cached_anova_len1 = len(translated_trial_matrix)
else:
anova_stats = self.cached_anova_text1
return 'UF: ' + string_perf_by_side + '. Biases: ' + anova_stats
def choose_scheduler_main_body(self, translated_trial_matrix):
# Main body of session
this_trial = len(translated_trial_matrix)
# Do nothing if we've changed recently
if this_trial < self.last_changed_trial + self.n_trials_sticky:
return
# Check whether we've had at least 10 random in the last 50
recents = translated_trial_matrix['isrnd'].values[-self.
n_trials_recent_win:]
recent_randoms = recents.sum()
if len(recents) == self.n_trials_recent_win and \
recent_randoms < self.n_trials_recent_random_thresh:
# Set to occasional random
self.current_sub_scheduler = self.sub_schedulers['RandomStim']
self.last_changed_trial = this_trial
self.params['status'] = 'randchk' + str(this_trial)
return
# Run the anova on all trials (used for checking for stay bias)
numericated_trial_matrix = TrialMatrix.numericate_trial_matrix(
translated_trial_matrix)
#~ recent_ntm = numericated_trial_matrix.iloc[
#~ -self.n_trials_recent_for_side_bias:]
aov_res = TrialMatrix._run_anova(numericated_trial_matrix)
if aov_res is None:
self.current_sub_scheduler = self.sub_schedulers['RandomStim']
self.last_changed_trial = this_trial
self.params['status'] = 'an_none' + str(this_trial)
return
# Also calculate the side bias in all recent trials
recent_ttm = translated_trial_matrix.iloc[
-self.n_trials_recent_for_side_bias:]
# Take the largest significant bias
# Actually, better to take the diff of perf between sides for forced
# side. Although this is a bigger issue than unexplainable variance
# shouldn't be interpreted.
side2perf_all = TrialMatrix.count_hits_by_type(recent_ttm,
split_key='rewside')
if 'left' in side2perf_all and 'right' in side2perf_all:
lperf = side2perf_all['left'][0] / float(side2perf_all['left'][1])
rperf = side2perf_all['right'][0] / float(
side2perf_all['right'][1])
sideperf_diff = rperf - lperf
else:
sideperf_diff = 0
# Decide whether stay, side, or neither bias is critical
if aov_res['pvals']['p_prevchoice'] < 0.05 and aov_res['fit'][
'fit_prevchoice'] > 0:
# Stay bias
self.last_changed_trial = this_trial
self.params['status'] = 'antistay' + str(this_trial)
self.current_sub_scheduler = self.sub_schedulers[
'ForcedAlternation']
elif np.abs(sideperf_diff) > .25:
# Side bias
self.last_changed_trial = this_trial
self.params['status'] = 'antiside' + str(this_trial)
self.current_sub_scheduler = self.sub_schedulers['ForcedSide']
if sideperf_diff > 0:
self.current_sub_scheduler.params['side'] = 'left'
else:
self.current_sub_scheduler.params['side'] = 'right'
else:
# No bias
self.last_changed_trial = this_trial
self.params['status'] = 'good' + str(this_trial)
self.current_sub_scheduler = self.sub_schedulers['RandomStim']
def choose_scheduler_main_body(self, translated_trial_matrix):
# Main body of session
this_trial = len(translated_trial_matrix)
# Do nothing if we've changed recently
if this_trial < self.last_changed_trial + self.n_trials_sticky:
return
# Check whether we've had at least 10 random in the last 50
recents = translated_trial_matrix['isrnd'].values[
-self.n_trials_recent_win:]
recent_randoms = recents.sum()
if len(recents) == self.n_trials_recent_win and \
recent_randoms < self.n_trials_recent_random_thresh:
# Set to occasional random
self.current_sub_scheduler = self.sub_schedulers['RandomStim']
self.last_changed_trial = this_trial
self.params['status'] = 'randchk' + str(this_trial)
return
# Run the anova on all trials (used for checking for stay bias)
numericated_trial_matrix = TrialMatrix.numericate_trial_matrix(
translated_trial_matrix)
#~ recent_ntm = numericated_trial_matrix.iloc[
#~ -self.n_trials_recent_for_side_bias:]
aov_res = TrialMatrix._run_anova(numericated_trial_matrix)
if aov_res is None:
self.current_sub_scheduler = self.sub_schedulers['RandomStim']
self.last_changed_trial = this_trial
self.params['status'] = 'an_none' + str(this_trial)
return
# Also calculate the side bias in all recent trials
recent_ttm = translated_trial_matrix.iloc[
-self.n_trials_recent_for_side_bias:]
# Take the largest significant bias
# Actually, better to take the diff of perf between sides for forced
# side. Although this is a bigger issue than unexplainable variance
# shouldn't be interpreted.
side2perf_all = TrialMatrix.count_hits_by_type(
recent_ttm, split_key='rewside')
if 'left' in side2perf_all and 'right' in side2perf_all:
lperf = side2perf_all['left'][0] / float(side2perf_all['left'][1])
rperf = side2perf_all['right'][0] / float(side2perf_all['right'][1])
sideperf_diff = rperf - lperf
else:
sideperf_diff = 0
# Decide whether stay, side, or neither bias is critical
if aov_res['pvals']['p_prevchoice'] < 0.05 and aov_res['fit']['fit_prevchoice'] > 0:
# Stay bias
self.last_changed_trial = this_trial
self.params['status'] = 'antistay' + str(this_trial)
self.current_sub_scheduler = self.sub_schedulers['ForcedAlternation']
elif np.abs(sideperf_diff) > .25:
# Side bias
self.last_changed_trial = this_trial
self.params['status'] = 'antiside' + str(this_trial)
self.current_sub_scheduler = self.sub_schedulers['ForcedSide']
if sideperf_diff > 0:
self.current_sub_scheduler.params['side'] = 'left'
else:
self.current_sub_scheduler.params['side'] = 'right'
else:
# No bias
self.last_changed_trial = this_trial
self.params['status'] = 'good' + str(this_trial)
self.current_sub_scheduler = self.sub_schedulers['RandomStim']