def action_predictions_list():
'''
shows a table with the list of predictions
'''
# get de model repo path
predictions_path = pathlib.Path(utils.predictions_repository_path())
if predictions_path.is_dir() is False:
return False, 'the predictions repository path does not exist. Please run "flame -c config".'
# get directories in model repo path
dirs = [x for x in predictions_path.iterdir() if x.is_dir()]
result = []
iresult = []
# iterate models
for d in dirs:
#label is retrieved from the directory name
label = d.parts[-1]
#metainfo is extracted from prediction-meta picke
with open(d.joinpath('prediction-meta.pkl'), 'rb') as handle:
endpoint = pickle.load(handle)
version = pickle.load(handle)
ifile = pickle.load(handle)
time = pickle.load(handle)
timestamp = pickle.load(handle)
try:
modelID = pickle.load(handle)
except:
modelID = '*legacy*'
# ifile is simplified to avoid discossing the repository
ifile = os.path.basename(ifile)
# ensemble models are hidden
if label[0:8] == 'ensemble':
continue
# add as a tupla
iresult.append((label, endpoint, version, time, ifile, modelID))
# format as a text line for reverse date sorting and printing
line = f'{label:10} {endpoint:15} {version} {time} {ifile} {modelID}'
result.append((timestamp, line))
result.sort(reverse=True, key=getdate)
[print(i[1]) for i in result]
return True, iresult
def action_predictions_result (label, output='text'):
'''
try to retrieve the prediction result with the label used as argument
returns
- (False, Null) if it there is no directory or the predictions
pickle files cannot be found
- (True, object) with the results otherwyse
'''
# get de model repo path
predictions_path = pathlib.Path(utils.predictions_repository_path())
label_path = predictions_path.joinpath(label)
if not os.path.isdir(label_path):
if output != 'text':
return False, {'code':0, 'message': f'directory {label_path} not found'}
print (f'directory {label_path} not found')
return False, None
result_path = label_path.joinpath('prediction-results.pkl')
if not result_path.is_file():
if output != 'text':
return False, {'code':0, 'message': f'predictions not found for {label} directory'}
print (f'predictions not found for {label} directory')
return False, None
iconveyor = Conveyor()
with open(result_path, 'rb') as handle:
success, message = iconveyor.load(handle)
if not success:
if output != 'text':
return False, {'code':1, 'message': f'error reading prediction results with message {message}'}
print (f'error reading prediction results with message {message}')
return False, None
# console output
print_prediction_result(('obj_num','number of objects',iconveyor.getVal('obj_num')))
if iconveyor.isKey('external-validation'):
for val in iconveyor.getVal('external-validation'):
print_prediction_result (val)
if iconveyor.isKey('values'):
for i in range (iconveyor.getVal('obj_num')):
print (iconveyor.getVal('obj_nam')[i], '\t', float("{0:.4f}".format(iconveyor.getVal('values')[i])))
# return iconveyor
return True, iconveyor
def action_predictions_remove(label):
'''
try to remove the prediction result with the label used as argument
returns
- (False, message) if it there is no directory or the removal failed
- (True, OK) removal succeeded
'''
# get de model repo path
predictions_path = pathlib.Path(utils.predictions_repository_path())
label_path = predictions_path.joinpath(label)
if not os.path.isdir(label_path):
return (False, f'directory {label_path} not found')
try:
shutil.rmtree(label_path)
except Exception as e:
return (False, f'failed to remove {label_path} with error: {e}')
return (True, 'OK')
def run_apply(self):
''' Process the results of apply.
The ouput generated by the prediction are:
1. prediction-results.pkl and prediction-meta.pkl
2. console output
2. molecular descriptors file in TSV format [optional]
3. results file in TSV format [optional]
'''
# if len(self.conveyor.getMain()) == 0:
# self.conveyor.setError('Unable to find main prediction')
if 'ghost' not in self.format:
opath = utils.predictions_repository_path()
if os.path.isdir(opath):
opath = os.path.join(opath, self.label)
if not os.path.isdir(opath):
os.mkdir(opath)
###
# 1. prediction-results.pkl and prediction-meta.pkl
###
# Save conveyor from prediction only if confidential is False
if not self.param.getVal(
'confidential') and 'ghost' not in self.format:
results_pkl_path = os.path.join(opath, 'prediction-results.pkl')
meta_pkl_path = os.path.join(opath, 'prediction-meta.pkl')
LOG.info(f'saving model results to: {opath}')
# dump conveyor
with open(results_pkl_path, 'wb') as handle:
self.conveyor.save(handle)
# dump metainfo
with open(meta_pkl_path, 'wb') as handle:
pickle.dump(self.conveyor.getMeta('endpoint'), handle)
pickle.dump(self.conveyor.getMeta('version'), handle)
pickle.dump(self.conveyor.getMeta('input_file'), handle)
now = datetime.now()
pickle.dump(now.strftime("%d/%m/%Y %H:%M:%S"), handle)
pickle.dump(datetime.timestamp(now), handle)
pickle.dump(self.conveyor.getMeta('modelID'), handle)
pickle.dump(self.conveyor.getWarningMessage(), handle)
pickle.dump(self.conveyor.getErrorMessage(), handle)
####
# ! in case of error, print and return error
####
if self.conveyor.getError():
error_msg = self.conveyor.getErrorMessage()
return False, error_msg
####
# 2. console output
####
#print (self.results)
if 'ghost' not in self.format:
self.print_result(('obj_num', 'number of objects',
self.conveyor.getVal('obj_num')))
if self.conveyor.isKey('external-validation'):
for val in self.conveyor.getVal('external-validation'):
self.print_result(val)
if self.conveyor.isKey('values'):
nams = self.conveyor.getVal('obj_nam')
vals = self.conveyor.getVal('values')
for i in range(self.conveyor.getVal('obj_num')):
# print (self.conveyor.getVal('obj_nam')[i], '\t' , float("{0:.4f}".format(self.conveyor.getVal('values')[i])))
print(f'{nams[i]}\t{vals[i]:.4f}')
###
# 3. molecular descriptors file in TSV format [optional]
###
if 'ghost' not in self.format:
if self.param.getVal('output_md'):
self._output_md()
###
# 4. results file in TSV format [optional]
###
if 'ghost' not in self.format and 'TSV' in self.format:
LOG.info('writting results to TSV file "output.tsv"')
# label and smiles
key_list = ['obj_nam']
if self.conveyor.isKey('SMILES'):
key_list.append('SMILES')
# main result
key_list += self.conveyor.getMain()
# add all object type results
# manifest = self.results['manifest']
# for item in manifest:
# if item['dimension'] == 'objs' and item['key'] not in key_list:
# key_list.append(item['key'])
key_obj = self.conveyor.objectKeys()
for i in key_obj:
if i not in key_list:
key_list.append(i)
with open('output.tsv', 'w') as fo:
header = ''
for label in key_list:
header += label + '\t'
fo.write(header + '\n')
obj_num = int(self.conveyor.getVal('obj_num'))
for i in range(obj_num):
line = ''
for key in key_list:
ikey = self.conveyor.getVal(key)
if ikey is None:
line += '-\t'
continue
if i >= len(ikey):
val = None
else:
val = ikey[i]
if val is None:
line += '-'
else:
if isinstance(val, float):
line += "%.4f" % val
else:
line += str(val)
line += '\t'
fo.write(line + '\n')
return True, self.conveyor
def config(path: str=None) -> bool:
"""Configures model repository.
Loads config.yaml and writes a correct model repository path
with the path provided by the user or a default from appdirs
if the path is not provided.
"""
# ---- CLI interface -----
if path is None: # set default
default_models_path = Path(appdirs.user_data_dir('models', 'flame'))
default_spaces_path = Path(appdirs.user_data_dir('spaces', 'flame'))
default_predictions_path = Path(appdirs.user_data_dir('predictions', 'flame'))
print(f'Setting model, space and predictions repositories (default) to {default_models_path}, {default_spaces_path} and {default_predictions_path}'
'\nWould you like to continue?(y/n)')
if ask_user():
if default_models_path.exists() or default_spaces_path.exists() or default_predictions_path.exists():
print(f'These paths already exists. '
'Would you like to set them anyway?(y/n)')
if ask_user():
utils.set_repositories(default_models_path, default_spaces_path, default_predictions_path)
else:
print('aborting...')
return False
else: # models_path doesn't exists
default_models_path.mkdir(parents=True)
default_spaces_path.mkdir(parents=True)
default_predictions_path.mkdir(parents=True)
utils.set_repositories(default_models_path, default_spaces_path, default_predictions_path)
print(f'model repository set to {default_models_path}')
print(f'space repository set to {default_spaces_path}')
print(f'predictions repository set to {default_predictions_path}')
return True
else:
print('aborting...')
return False
else: # path input by user
in_path = Path(path).expanduser()
in_path_models = Path.joinpath(in_path,'models')
in_path_spaces = Path.joinpath(in_path,'spaces')
in_path_predictions = Path.joinpath(in_path,'predictions')
current_models_path = Path(utils.model_repository_path())
current_spaces_path = Path(utils.space_repository_path())
current_predictions_path = Path(utils.predictions_repository_path())
if in_path_models == current_models_path and in_path_spaces == current_spaces_path and in_path_predictions == current_predictions_path:
print(f'{in_path_models} already is model repository path')
print(f'{in_path_spaces} already is space repository path')
print(f'{in_path_predictions} already is predictions repository path')
return False
elif not (in_path_models.exists() and in_path_spaces.exists() and in_path_predictions.exists()):
print("paths doesn't exists. Would you like to create it?(y/n)")
if ask_user():
if not in_path_models.exists():
in_path_models.mkdir(parents=True)
if not in_path_spaces.exists():
in_path_spaces.mkdir(parents=True)
if not in_path_predictions.exists():
in_path_predictions.mkdir(parents=True)
utils.set_repositories(in_path_models, in_path_spaces, in_path_predictions)
else:
print('aborting...')
return False
else: # in_path exists
utils.set_repositories(in_path_models, in_path_spaces, in_path_predictions)
print(f'space repository set to {in_path_spaces}')
print(f'model repository set to {in_path_models}')
print(f'predictions repository set to {in_path_predictions}')
return True
def run_apply(self):
''' Process the results of apply.
The ouput generated by the prediction are:
1. console output
2. molecular descriptors file in TSV format [optional]
3. results file in TSV format [optional]
4. this function return results in JSON format [optional]
'''
if len(self.conveyor.getMain()) == 0:
self.conveyor.setError('Unable to find main prediction')
return
opath = utils.predictions_repository_path()
if os.path.isdir(opath):
opath = os.path.join(opath, self.label)
if not os.path.isdir(opath):
os.mkdir(opath)
# the ouput generated by the prediction are:
#
# 1. console output
# 2. molecular descriptors file in TSV format [optional]
# 3. results file in TSV format [optional]
# 4. this function return results in JSON format [optional]
####
# 1. console output
####
#print (self.results)
self.print_result(
('obj_num', 'number of objects', self.conveyor.getVal('obj_num')))
if self.conveyor.isKey('external-validation'):
for val in self.conveyor.getVal('external-validation'):
self.print_result(val)
if self.conveyor.isKey('values'):
for i in range(self.conveyor.getVal('obj_num')):
print(
self.conveyor.getVal('obj_nam')[i], '\t',
float("{0:.4f}".format(self.conveyor.getVal('values')[i])))
###
# 2. molecular descriptors file in TSV format [optional]
###
if self.param.getVal('output_md'):
self._output_md()
###
# 3. results file in TSV format [optional]
###
if 'TSV' in self.format:
LOG.info('writting results to TSV file "output.tsv"')
# label and smiles
key_list = ['obj_nam']
if self.conveyor.isKey('SMILES'):
key_list.append('SMILES')
# main result
key_list += self.conveyor.getMain()
# add all object type results
# manifest = self.results['manifest']
# for item in manifest:
# if item['dimension'] == 'objs' and item['key'] not in key_list:
# key_list.append(item['key'])
key_obj = self.conveyor.objectKeys()
for i in key_obj:
if i not in key_list:
key_list.append(i)
with open('output.tsv', 'w') as fo:
header = ''
for label in key_list:
header += label + '\t'
fo.write(header + '\n')
obj_num = int(self.conveyor.getVal('obj_num'))
for i in range(obj_num):
line = ''
for key in key_list:
ikey = self.conveyor.getVal(key)
if ikey is None:
line += '-\t'
continue
if i >= len(ikey):
val = None
else:
val = ikey[i]
if val is None:
line += '-'
else:
if isinstance(val, float):
line += "%.4f" % val
else:
line += str(val)
line += '\t'
fo.write(line + '\n')
# the function returns "True, output". output can be empty or a JSON
output = ''
###
# 4. this function return results in JSON format [optional]
###
# returns a JSON with the prediction results
if 'JSON' in self.format:
if self.param.getVal('input_type') == 'model_ensemble':
output = self.conveyor.getJSON(xdata=True)
else:
output = self.conveyor.getJSON()
#print (self.conveyor.getJSON())
# Save conveyor from prediction only if confidential is False
if not self.param.getVal('confidential'):
if not os.path.isdir(opath):
return True, output
results_pkl_path = os.path.join(opath, 'prediction-results.pkl')
meta_pkl_path = os.path.join(opath, 'prediction-meta.pkl')
LOG.info('saving model results to: {}'.format(opath))
# dump conveyor
with open(results_pkl_path, 'wb') as handle:
self.conveyor.save(handle)
# dump metainfo
with open(meta_pkl_path, 'wb') as handle:
pickle.dump(self.conveyor.getMeta('endpoint'), handle)
pickle.dump(self.conveyor.getMeta('version'), handle)
pickle.dump(self.conveyor.getMeta('input_file'), handle)
now = datetime.now()
pickle.dump(now.strftime("%d/%m/%Y %H:%M:%S"), handle)
pickle.dump(datetime.timestamp(now), handle)
# print (self.conveyor.getMeta('endpoint'))
# print ('saving version',self.conveyor.getMeta('version'))
# print (self.conveyor.getMeta('input_file'))
# now = datetime.now()
# print (now.strftime("%d/%m/%Y %H:%M:%S"))
return True, output
