def calc_trial_fitness_df(trial):
'''
Calculate the trial fitness df by aggregating from the collected session_data_dict (session_fitness_df's).
Adds a consistency dimension to fitness vector.
'''
trial_fitness_data = {}
try:
all_session_fitness_df = pd.concat(list(trial.session_data_dict.values()))
except ValueError as e:
logger.exception('Sessions failed, no data to analyze. Check stack trace above')
for aeb in util.get_df_aeb_list(all_session_fitness_df):
aeb_fitness_df = all_session_fitness_df.loc[:, aeb]
aeb_fitness_sr = aeb_fitness_df.mean()
consistency = calc_consistency(aeb_fitness_df)
aeb_fitness_sr = aeb_fitness_sr.append(pd.Series({'consistency': consistency}))
aeb_fitness_df = pd.DataFrame([aeb_fitness_sr], index=[trial.index])
aeb_fitness_df = aeb_fitness_df.reindex(FITNESS_COLS, axis=1)
trial_fitness_data[aeb] = aeb_fitness_df
# form multi_index df, then take mean across all bodies
trial_fitness_df = pd.concat(trial_fitness_data, axis=1)
mean_fitness_df = calc_mean_fitness(trial_fitness_df)
trial_fitness_df = mean_fitness_df
trial_fitness = calc_fitness(mean_fitness_df)
logger.info(f'Trial mean fitness: {trial_fitness}\n{mean_fitness_df}')
return trial_fitness_df
def save_session_df(session_data, prepath, info_space):
'''Save session_df, and if is in eval mode, modify it and save with append'''
filepath = f'{prepath}_session_df.csv'
if util.get_lab_mode() in ('enjoy', 'eval'):
ckpt = util.find_ckpt(info_space.eval_model_prepath)
epi = int(re.search('epi(\d+)', ckpt)[1])
totalt = int(re.search('totalt(\d+)', ckpt)[1])
session_df = pd.concat(session_data, axis=1)
eval_session_df = pd.DataFrame(data=[session_df.mean()])
for aeb in util.get_df_aeb_list(eval_session_df):
eval_session_df.loc[:, aeb + ('epi', )] = epi
eval_session_df.loc[:, aeb + ('total_t', )] = totalt
# if eval, save with append mode
header = not os.path.exists(filepath)
with open(filepath, 'a') as f:
eval_session_df.to_csv(f, header=header)
else:
session_df = pd.concat(session_data, axis=1)
util.write(session_df, filepath)
def calc_trial_fitness_df(trial):
'''Calculate the trial fitness df'''
trial_fitness_data = {}
all_session_fitness_df = pd.concat(
list(trial.session_fitness_df_dict.values()))
for aeb in util.get_df_aeb_list(all_session_fitness_df):
aeb_df = all_session_fitness_df.loc[:, aeb]
aeb_fitness_sr = aeb_df.mean()
consistency = calc_consistency(aeb_df.values)
aeb_fitness_sr = aeb_fitness_sr.append(
pd.Series({'consistency': consistency}))
aeb_fitness_df = pd.DataFrame([aeb_fitness_sr], index=[trial.index])
trial_fitness_data[aeb] = aeb_fitness_df
trial_fitness_df = pd.concat(trial_fitness_data, axis=1)
mean_fitness_df = trial_fitness_df.mean(axis=1,
level=3) # mean across all bodies
trial_fitness_df = mean_fitness_df # agg for trial level
trial_fitness = calc_fitness(mean_fitness_df)
logger.info(f'Trial mean fitness: {trial_fitness}\n{mean_fitness_df}')
return trial_fitness_df
def save_session_df(session_data, filepath, info_space):
'''Save session_df, and if is in eval mode, modify it and save with append'''
if util.in_eval_lab_modes():
ckpt = util.find_ckpt(info_space.eval_model_prepath)
epi = int(re.search('epi(\d+)', ckpt)[1])
totalt = int(re.search('totalt(\d+)', ckpt)[1])
session_df = pd.concat(session_data, axis=1)
mean_sr = session_df.mean()
mean_sr.name = totalt # set index to prevent all being the same
eval_session_df = pd.DataFrame(data=[mean_sr])
# set sr name too, to total_t
for aeb in util.get_df_aeb_list(eval_session_df):
eval_session_df.loc[:, aeb + ('epi',)] = epi
eval_session_df.loc[:, aeb + ('total_t',)] = totalt
# if eval, save with append mode
header = not os.path.exists(filepath)
with open(filepath, 'a') as f:
eval_session_df.to_csv(f, header=header)
else:
session_df = pd.concat(session_data, axis=1)
util.write(session_df, filepath)
def calc_trial_fitness_df(trial):
'''
Calculate the trial fitness df by aggregating from the collected session_data_dict (session_fitness_df's).
Adds a consistency dimension to fitness vector.
'''
trial_fitness_data = {}
all_session_fitness_df = pd.concat(list(trial.session_data_dict.values()))
for aeb in util.get_df_aeb_list(all_session_fitness_df):
aeb_df = all_session_fitness_df.loc[:, aeb]
aeb_fitness_sr = aeb_df.mean()
consistency = calc_consistency(aeb_df.values)
aeb_fitness_sr = aeb_fitness_sr.append(
pd.Series({'consistency': consistency}))
aeb_fitness_df = pd.DataFrame([aeb_fitness_sr], index=[trial.index])
trial_fitness_data[aeb] = aeb_fitness_df
# form multiindex df, then take mean across all bodies
trial_fitness_df = pd.concat(trial_fitness_data, axis=1)
mean_fitness_df = trial_fitness_df.mean(axis=1, level=3)
trial_fitness_df = mean_fitness_df
trial_fitness = calc_fitness(mean_fitness_df)
logger.info(f'Trial mean fitness: {trial_fitness}\n{mean_fitness_df}')
return trial_fitness_df
def calc_trial_fitness_df(trial):
'''
Calculate the trial fitness df by aggregating from the collected session_data_dict (session_fitness_df's).
Adds a consistency dimension to fitness vector.
'''
trial_fitness_data = {}
all_session_fitness_df = pd.concat(list(trial.session_data_dict.values()))
for aeb in util.get_df_aeb_list(all_session_fitness_df):
aeb_fitness_df = all_session_fitness_df.loc[:, aeb]
aeb_fitness_sr = aeb_fitness_df.mean()
consistency = calc_consistency(aeb_fitness_df)
aeb_fitness_sr = aeb_fitness_sr.append(pd.Series({'consistency': consistency}))
aeb_fitness_df = pd.DataFrame([aeb_fitness_sr], index=[trial.index])
aeb_fitness_df = aeb_fitness_df.reindex(FITNESS_COLS, axis=1)
trial_fitness_data[aeb] = aeb_fitness_df
# form multiindex df, then take mean across all bodies
trial_fitness_df = pd.concat(trial_fitness_data, axis=1)
mean_fitness_df = trial_fitness_df.mean(axis=1, level=3)
trial_fitness_df = mean_fitness_df
trial_fitness = calc_fitness(mean_fitness_df)
logger.info(f'Trial mean fitness: {trial_fitness}\n{mean_fitness_df}')
return trial_fitness_df
