def logreg_results_to_pandas(common_molids_cache=False):
"""Collects all the results in disk and place them in record-format in a pandas dataframe.
Allows convenient reporting, grouping and filtering of results.
"""
results = ResultInDisk.collect_results_under_dir(MALARIA_LOGREGS_EXPERIMENT_ROOT,
factory=malaria_result_factory)
# --- molids cache
molids_cache = None
if common_molids_cache:
rf_lab, rf_amb, rf_unl, rf_scr = malaria_logreg_fpt_providers(None)
# Labelled molids
lab_molids = rf_lab.ids()
amb_molids = rf_amb.ids() # To prioritize confirmatory tests on labelled data
# Unlabelled molids
unl_molids = rf_unl.ids()
scr_molids = rf_scr.ids()
# Let's avoid the need to reread them...
molids_cache = {
'lab': lab_molids,
'amb': amb_molids,
'unl': unl_molids,
'scr': scr_molids
}
results_dict_of_dicts = {}
for result in results:
if common_molids_cache:
result.ids_cache = molids_cache # dodgy, rework with a copying constructor
rdict = copy(result.info())
rdict['result'] = result
rdict['class_weight'] = 'uniform' if rdict['class_weight'] is None else rdict['class_weight']
# Some more ad-hoc keys for the model
rdict['num_present_folds'] = result.num_present_folds()
rdict['auc_mean'] = result.auc_mean()
rdict['enrichement5_mean'] = result.enrichement5_mean()
# Some more ad-hoc keys for the fingerprint folder
folder = result.fingerprint_folder()
rdict['folder_seed'] = int(folder.seed) if folder is not None else -1
rdict['folder_size'] = int(folder.fold_size) if folder is not None else 0
# Add this result to the data frame
results_dict_of_dicts[result.root_key()] = rdict
return DataFrame(results_dict_of_dicts).T
def trees_results_to_pandas(common_molids_cache=False):
"""Collects all the results in disk and place them in record-format in a pandas dataframe.
Allows convenient reporting, grouping and filtering of results.
"""
results = ResultInDisk.collect_results_under_dir(MALARIA_TREES_EXPERIMENT_ROOT,
factory=malaria_result_factory)
# --- molids cache
molids_cache = None
if common_molids_cache:
a_result = results[0]
# Labelled molids
lab_molids = a_result.ids('lab')
amb_molids = a_result.ids('amb') # To prioritize confirmatory tests on labelled data
# Unlabelled molids
unl_molids = a_result.ids('unl')
scr_molids = a_result.ids('scr')
# Let's avoid the need to reread them...
molids_cache = {
'lab': lab_molids,
'amb': amb_molids,
'unl': unl_molids,
'scr': scr_molids
}
results_dict_of_dicts = {}
for result in results:
if common_molids_cache:
result.ids_cache = molids_cache # dodgy, rework with a copying constructor
rdict = copy(result.info())
rdict['result'] = result
# Some more ad-hoc keys for the model
model_params = split_by(result.model_setup_id())[1]
rdict['model_num_trees'] = int(model_params['num_trees'])
rdict['model_seed'] = int(model_params['seed'])
rdict['model_type'] = 'ExtraTrees' if model_params['etc'] else 'RandomForest'
# Add this result to the data frame
results_dict_of_dicts[result.root_key()] = rdict
return DataFrame(results_dict_of_dicts).T
def clean_results_pre_infojson_bug_fix():
results = ResultInDisk.collect_results_under_dir(MALARIA_LOGREGS_EXPERIMENT_ROOT,
factory=malaria_result_factory)
bad_results = [res for res in results if not op.isfile(op.join(res.eval_dir, 'info.json'))]
for res in bad_results:
info('Bye %s' % res.eval_dir)
def trees_molids(dset='lab'):
# No need to do this on a per-result basis because
# atm we are warranted that they are the same accross all evaluations.
a_result = ResultInDisk.collect_results_under_dir(MALARIA_TREES_EXPERIMENT_ROOT)[0]
return a_result.ids(dset=dset)
