def test_data_container_save_wrong_format(self):
data_sets = [{'name': 'dataset1', 'frame': pd.DataFrame(np.random.normal(size=(100, 2)),
columns=['A', 'B'])},
{'name': 'dataset2', 'frame': pd.DataFrame(np.random.normal(size=(120, 3)),
columns=['A', 'B', 'C'])}]
container = DataContainer(data_sets)
directory = 'temp_directory_container_save_wrong_format_xyz'
writer = DataWriter()
writer.write_experiment_output(directory,
container,
dataframe_names=['dataset1'],
file_format='html')
def test_data_container_save_wrong_format(self):
data_sets = [{'name': 'dataset1', 'frame': pd.DataFrame(np.random.normal(size=(100, 2)),
columns=['A', 'B'])},
{'name': 'dataset2', 'frame': pd.DataFrame(np.random.normal(size=(120, 3)),
columns=['A', 'B', 'C'])}]
container = DataContainer(data_sets)
directory = 'temp_directory_container_save_wrong_format_xyz'
writer = DataWriter()
writer.write_experiment_output(directory,
container,
dataframe_names=['dataset1'],
file_format='html')
def test_data_container_save_files_with_id(self):
data_sets = [{'name': 'dataset1', 'frame': pd.DataFrame(np.random.normal(size=(100, 2)),
columns=['A', 'B'])},
{'name': 'dataset2', 'frame': pd.DataFrame(np.random.normal(size=(120, 3)),
columns=['A', 'B', 'C'])}]
container = DataContainer(data_sets)
directory = 'temp_directory_save_files_with_id_xyz'
os.makedirs(directory, exist_ok=True)
writer = DataWriter('test')
for file_type in ['json', 'csv', 'xlsx']:
if file_type != 'json':
writer.write_experiment_output(directory,
container,
dataframe_names=['dataset1'],
file_format=file_type)
else:
writer.write_experiment_output(directory,
container,
new_names_dict={'dataset1': 'aaa'},
dataframe_names=['dataset1'],
file_format=file_type)
aaa_json = pd.read_json(os.path.join(directory, 'test_aaa.json'))
ds_1_csv = pd.read_csv(os.path.join(directory, 'test_dataset1.csv'))
ds_1_xls = pd.read_excel(os.path.join(directory, 'test_dataset1.xlsx'))
output_dir = os.listdir(directory)
rmtree(directory)
assert sorted(output_dir) == sorted(['test_aaa.json',
'test_dataset1.csv',
'test_dataset1.xlsx'])
assert_frame_equal(container.dataset1, aaa_json)
assert_frame_equal(container.dataset1, ds_1_csv)
assert_frame_equal(container.dataset1, ds_1_xls)
def write_fairness_results(fit_dictionary, fairness_container, group,
output_dir, experiment_id, file_format):
"""
Save the results of fairness analysis to disk.
Parameters
----------
fit_dictionary: dict
A dictionary of fitted models generated by ``get_fairness_analyses()``.
fairness_container: container.DataContainer
A data container with the results of fairness analysis generated by
``get_fairness_analyses()``.
group: str
The subgroup considered in this analysis.
output_dir: str
The directory where the results will be saved.
experiment_id: str
Experiment ID.
file_format: str
File format to use for data files.
"""
# let's first save model files and summaries
for model in fit_dictionary:
fit = fit_dictionary[model]
ols_file = join(output_dir,
'{}_{}_by_{}.ols'.format(experiment_id, model, group))
summary_file = join(
output_dir,
'{}_{}_by_{}_ols_summary.txt'.format(experiment_id, model, group))
with open(ols_file, 'wb') as olsf, open(summary_file, 'w') as summf:
pickle.dump(fit, olsf)
summf.write(str(fit.summary()))
# Now let's write out the content of the data container
writer = DataWriter(experiment_id)
writer.write_experiment_output(output_dir,
fairness_container,
file_format=file_format,
index=True)
def test_data_container_save_files_with_id(self):
data_sets = [{
'name':
'dataset1',
'frame':
pd.DataFrame(np.random.normal(size=(100, 2)), columns=['A', 'B'])
}, {
'name':
'dataset2',
'frame':
pd.DataFrame(np.random.normal(size=(120, 3)),
columns=['A', 'B', 'C'])
}]
container = DataContainer(data_sets)
directory = 'temp_directory_save_files_with_id_xyz'
os.makedirs(directory, exist_ok=True)
writer = DataWriter('test')
for file_type in ['json', 'csv', 'xlsx']:
if file_type != 'json':
writer.write_experiment_output(directory,
container,
dataframe_names=['dataset1'],
file_format=file_type)
else:
writer.write_experiment_output(
directory,
container,
new_names_dict={'dataset1': 'aaa'},
dataframe_names=['dataset1'],
file_format=file_type)
aaa_json = pd.read_json(os.path.join(directory, 'test_aaa.json'))
ds_1_csv = pd.read_csv(os.path.join(directory, 'test_dataset1.csv'))
ds_1_xls = pd.read_excel(os.path.join(directory, 'test_dataset1.xlsx'))
output_dir = os.listdir(directory)
rmtree(directory)
assert sorted(output_dir) == sorted(
['test_aaa.json', 'test_dataset1.csv', 'test_dataset1.xlsx'])
assert_frame_equal(container.dataset1, aaa_json)
assert_frame_equal(container.dataset1, ds_1_csv)
assert_frame_equal(container.dataset1, ds_1_xls)
def run_experiment(config_file_or_obj, output_dir):
"""
Run RSMTool experiment using the given configuration
file and generate all outputs in the given directory.
Parameters
----------
config_file_or_obj : str or Configuration
Path to the experiment configuration file.
Users can also pass a `Configuration` object that is in memory.
output_dir : str
Path to the experiment output directory.
Raises
------
ValueError
If any of the required fields are missing or ill-specified.
"""
logger = logging.getLogger(__name__)
# create the 'output' and the 'figure' sub-directories
# where all the experiment output such as the CSV files
# and the box plots will be saved
# Get absolute paths to output directories
csvdir = abspath(join(output_dir, 'output'))
figdir = abspath(join(output_dir, 'figure'))
reportdir = abspath(join(output_dir, 'report'))
featuredir = abspath(join(output_dir, 'feature'))
# Make directories, if necessary
makedirs(csvdir, exist_ok=True)
makedirs(figdir, exist_ok=True)
makedirs(reportdir, exist_ok=True)
# Allow users to pass Configuration object to the
# `config_file_or_obj` argument, rather than read from file
if not isinstance(config_file_or_obj, Configuration):
# Instantiate configuration parser object
parser = ConfigurationParser.get_configparser(config_file_or_obj)
configuration = parser.read_normalize_validate_and_process_config(
config_file_or_obj)
# get the directory where the configuration file lives
configpath = dirname(config_file_or_obj)
else:
configuration = config_file_or_obj
if configuration.filepath is not None:
configpath = dirname(configuration.filepath)
else:
configpath = getcwd()
logger.info('Saving configuration file.')
configuration.save(output_dir)
# Get output format
file_format = configuration.get('file_format', 'csv')
# Get DataWriter object
writer = DataWriter(configuration['experiment_id'])
# Get the paths and names for the DataReader
(file_names, file_paths_org) = configuration.get_names_and_paths(
['train_file', 'test_file', 'features', 'feature_subset_file'],
['train', 'test', 'feature_specs', 'feature_subset_specs'])
file_paths = DataReader.locate_files(file_paths_org, configpath)
# if there are any missing files after trying to locate
# all expected files, raise an error
if None in file_paths:
missing_file_paths = [
file_paths_org[idx] for idx, path in enumerate(file_paths)
if path is None
]
raise FileNotFoundError('The following files were not found: '
'{}'.format(repr(missing_file_paths)))
# Use the default converter for both train and test
converters = {
'train': configuration.get_default_converter(),
'test': configuration.get_default_converter()
}
logger.info('Reading in all data from files.')
# Initialize the reader
reader = DataReader(file_paths, file_names, converters)
data_container = reader.read()
logger.info('Preprocessing all features.')
# Initialize the processor
processor = FeaturePreprocessor()
(processed_config,
processed_container) = processor.process_data(configuration,
data_container)
# Rename certain frames with more descriptive names
# for writing out experiment files
rename_dict = {
'train_excluded': 'train_excluded_responses',
'test_excluded': 'test_excluded_responses',
'train_length': 'train_response_lengths',
'train_flagged': 'train_responses_with_excluded_flags',
'test_flagged': 'test_responses_with_excluded_flags'
}
logger.info('Saving training and test set data to disk.')
# Write out files
writer.write_experiment_output(
csvdir,
processed_container, [
'train_features', 'test_features', 'train_metadata',
'test_metadata', 'train_other_columns', 'test_other_columns',
'train_preprocessed_features', 'test_preprocessed_features',
'train_excluded', 'test_excluded', 'train_length',
'test_human_scores', 'train_flagged', 'test_flagged'
],
rename_dict,
file_format=file_format)
# Initialize the analyzer
analyzer = Analyzer()
(analyzed_config,
analyzed_container) = analyzer.run_data_composition_analyses_for_rsmtool(
processed_container, processed_config)
# Write out files
writer.write_experiment_output(csvdir,
analyzed_container,
file_format=file_format)
logger.info('Training {} model.'.format(processed_config['model_name']))
# Initialize modeler
modeler = Modeler()
modeler.train(processed_config, processed_container, csvdir, figdir,
file_format)
# Identify the features used by the model
selected_features = modeler.get_feature_names()
# Add selected features to processed configuration
processed_config['selected_features'] = selected_features
# Write out files
writer.write_feature_csv(featuredir,
processed_container,
selected_features,
file_format=file_format)
features_data_container = processed_container.copy()
# Get selected feature info, and write out to file
df_feature_info = features_data_container.feature_info.copy()
df_selected_feature_info = df_feature_info[df_feature_info['feature'].isin(
selected_features)]
selected_feature_dataset_dict = {
'name': 'selected_feature_info',
'frame': df_selected_feature_info
}
features_data_container.add_dataset(selected_feature_dataset_dict,
update=True)
writer.write_experiment_output(
csvdir,
features_data_container,
dataframe_names=['selected_feature_info'],
new_names_dict={'selected_feature_info': 'feature'},
file_format=file_format)
logger.info('Running analyses on training set.')
(train_analyzed_config,
train_analyzed_container) = analyzer.run_training_analyses(
processed_container, processed_config)
# Write out files
writer.write_experiment_output(csvdir,
train_analyzed_container,
reset_index=True,
file_format=file_format)
# Use only selected features for predictions
columns_for_prediction = ['spkitemid', 'sc1'] + selected_features
train_for_prediction = processed_container.train_preprocessed_features[
columns_for_prediction]
test_for_prediction = processed_container.test_preprocessed_features[
columns_for_prediction]
logged_str = 'Generating training and test set predictions'
logged_str += ' (expected scores).' if configuration[
'predict_expected_scores'] else '.'
logger.info(logged_str)
(pred_config, pred_data_container) = modeler.predict_train_and_test(
train_for_prediction, test_for_prediction, processed_config)
# Write out files
writer.write_experiment_output(
csvdir,
pred_data_container,
new_names_dict={'pred_test': 'pred_processed'},
file_format=file_format)
original_coef_file = join(
csvdir, '{}_coefficients.{}'.format(pred_config['experiment_id'],
file_format))
# If coefficients file exists, then generate
# scaled coefficients and save to file
if exists(original_coef_file):
logger.info('Scaling the coefficients and saving them to disk')
try:
# Scale coefficients, and return DataContainer w/ scaled coefficients
scaled_data_container = modeler.scale_coefficients(pred_config)
# Write out files to disk
writer.write_experiment_output(csvdir,
scaled_data_container,
file_format=file_format)
except AttributeError:
raise ValueError(
"It appears you are trying to save two different "
"experiments to the same directory using the same "
"ID. Please clear the content of the directory and "
"rerun both experiments using different "
"experiment IDs.")
# Add processed data_container frames to pred_data_container
new_pred_data_container = pred_data_container + processed_container
logger.info('Running prediction analyses.')
(pred_analysis_config,
pred_analysis_data_container) = analyzer.run_prediction_analyses(
new_pred_data_container, pred_config)
# Write out files
writer.write_experiment_output(csvdir,
pred_analysis_data_container,
reset_index=True,
file_format=file_format)
# Initialize reporter
reporter = Reporter()
# generate the report
logger.info('Starting report generation.')
reporter.create_report(processed_config, csvdir, figdir)
def compute_and_save_predictions(config_file_or_obj, output_file, feats_file=None):
"""
Run ``rsmpredict`` with given configuration file and generate
predictions (and, optionally, pre-processed feature values).
Parameters
----------
config_file_or_obj : str or configuration_parser.Configuration
Path to the experiment configuration file.
Users can also pass a `Configuration` object that is in memory.
output_dir : str
Path to the output directory for saving files.
feats_file (optional): str
Path to the output file for saving preprocessed feature values.
Raises
------
ValueError
If any of the required fields are missing or ill-specified.
"""
logger = logging.getLogger(__name__)
# Allow users to pass Configuration object to the
# `config_file_or_obj` argument, rather than read file
if not isinstance(config_file_or_obj, Configuration):
# Instantiate configuration parser object
parser = ConfigurationParser.get_configparser(config_file_or_obj)
config = parser.read_normalize_validate_and_process_config(config_file_or_obj,
context='rsmpredict')
# get the directory where the config file lives
configpath = dirname(config_file_or_obj)
else:
config = config_file_or_obj
if config.filepath is not None:
configpath = dirname(config.filepath)
else:
configpath = os.getcwd()
# get the experiment ID
experiment_id = config['experiment_id']
# Get output format
file_format = config.get('file_format', 'csv')
# Get DataWriter object
writer = DataWriter(experiment_id)
# get the input file containing the feature values
# for which we want to generate the predictions
input_features_file = DataReader.locate_files(config['input_features_file'], configpath)
if not input_features_file:
raise FileNotFoundError('Input file {} does not exist'
''.format(config['input_features_file']))
experiment_dir = DataReader.locate_files(config['experiment_dir'], configpath)
if not experiment_dir:
raise FileNotFoundError('The directory {} does not exist.'
''.format(config['experiment_dir']))
else:
experiment_output_dir = normpath(join(experiment_dir, 'output'))
if not exists(experiment_output_dir):
raise FileNotFoundError('The directory {} does not contain '
'the output of an rsmtool experiment.'.format(experiment_dir))
# find all the .model files in the experiment output directory
model_files = glob.glob(join(experiment_output_dir, '*.model'))
if not model_files:
raise FileNotFoundError('The directory {} does not contain any rsmtool models.'
''.format(experiment_output_dir))
experiment_ids = [splitext(basename(mf))[0] for mf in model_files]
if experiment_id not in experiment_ids:
raise FileNotFoundError('{} does not contain a model for the experiment "{}". '
'The following experiments are contained in this '
'directory: {}'.format(experiment_output_dir,
experiment_id,
experiment_ids))
# check that the directory contains outher required files
required_file_types = ['feature', 'postprocessing_params']
for file_type in required_file_types:
expected_file_name = "{}_{}.csv".format(experiment_id, file_type)
if not exists(join(experiment_output_dir, expected_file_name)):
raise FileNotFoundError('{} does not contain the required file '
'{} that was generated during the '
'original model training'.format(experiment_output_dir,
expected_file_name))
# model_files = glob.glob(join(experiment_output_dir, '*.model'))
# if not model_files:
# raise FileNotFoundError('The directory {} does not contain any rsmtool models. '
# ''.format(experiment_output_dir))
logger.info('Reading input files.')
feature_info = join(experiment_output_dir,
'{}_feature.csv'.format(experiment_id))
post_processing = join(experiment_output_dir,
'{}_postprocessing_params.csv'.format(experiment_id))
file_paths = [input_features_file, feature_info, post_processing]
file_names = ['input_features',
'feature_info',
'postprocessing_params']
converters = {'input_features': config.get_default_converter()}
# Initialize the reader
reader = DataReader(file_paths, file_names, converters)
data_container = reader.read(kwargs_dict={'feature_info': {'index_col': 0}})
# load the Modeler to generate the predictions
model = Modeler.load_from_file(join(experiment_output_dir,
'{}.model'.format(experiment_id)))
# Add the model to the configuration object
config['model'] = model
# Initialize the processor
processor = FeaturePreprocessor()
(processed_config,
processed_container) = processor.process_data(config,
data_container,
context='rsmpredict')
# save the pre-processed features to disk if we were asked to
if feats_file is not None:
logger.info('Saving pre-processed feature values to {}'.format(feats_file))
feats_dir = dirname(feats_file)
# create any directories needed for the output file
os.makedirs(feats_dir, exist_ok=True)
_, feats_filename = split(feats_file)
feats_filename, _ = splitext(feats_filename)
# Write out files
writer.write_experiment_output(feats_dir,
processed_container,
include_experiment_id=False,
dataframe_names=['features_processed'],
new_names_dict={'features_processed':
feats_filename},
file_format=file_format)
if (output_file.lower().endswith('.csv') or
output_file.lower().endswith('.xlsx')):
output_dir = dirname(output_file)
_, filename = split(output_file)
filename, _ = splitext(filename)
else:
output_dir = output_file
filename = 'predictions_with_metadata'
# create any directories needed for the output file
os.makedirs(output_dir, exist_ok=True)
# save the predictions to disk
logger.info('Saving predictions.')
# Write out files
writer.write_experiment_output(output_dir,
processed_container,
include_experiment_id=False,
dataframe_names=['predictions_with_metadata'],
new_names_dict={'predictions_with_metadata':
filename},
file_format=file_format)
# save excluded responses to disk
if not processed_container.excluded.empty:
# save the predictions to disk
logger.info('Saving excluded responses to {}'.format(join(output_dir,
'{}_excluded_responses.csv'
''.format(filename))))
# Write out files
writer.write_experiment_output(output_dir,
processed_container,
include_experiment_id=False,
dataframe_names=['excluded'],
new_names_dict={'excluded':
'{}_excluded_responses'
''.format(filename)},
file_format=file_format)
def run_evaluation(config_file_or_obj, output_dir):
"""
Run an `rsmeval` experiment using the given configuration
file and generate all outputs in the given directory.
Parameters
----------
config_file_or_obj : str or configuration_parser.Configuration
Path to the experiment configuration file.
Users can also pass a `Configuration` object that is in memory.
output_dir : str
Path to the experiment output directory.
Raises
------
ValueError
If any of the required fields are missing or ill-specified.
"""
logger = logging.getLogger(__name__)
# create the 'output' and the 'figure' sub-directories
# where all the experiment output such as the CSV files
# and the box plots will be saved
csvdir = abspath(join(output_dir, 'output'))
figdir = abspath(join(output_dir, 'figure'))
reportdir = abspath(join(output_dir, 'report'))
os.makedirs(csvdir, exist_ok=True)
os.makedirs(figdir, exist_ok=True)
os.makedirs(reportdir, exist_ok=True)
# Allow users to pass Configuration object to the
# `config_file_or_obj` argument, rather than read file
if not isinstance(config_file_or_obj, Configuration):
# Instantiate configuration parser object
parser = ConfigurationParser.get_configparser(config_file_or_obj)
configuration = parser.read_normalize_validate_and_process_config(
config_file_or_obj, context='rsmeval')
# get the directory where the configuration file lives
configpath = dirname(config_file_or_obj)
else:
configuration = config_file_or_obj
if configuration.filepath is not None:
configpath = dirname(configuration.filepath)
else:
configpath = os.getcwd()
logger.info('Saving configuration file.')
configuration.save(output_dir)
# Get output format
file_format = configuration.get('file_format', 'csv')
# Get DataWriter object
writer = DataWriter(configuration['experiment_id'])
# Make sure prediction file can be located
if not DataReader.locate_files(configuration['predictions_file'],
configpath):
raise FileNotFoundError('Error: Predictions file {} '
'not found.\n'.format(
configuration['predictions_file']))
scale_with = configuration.get('scale_with')
# scale_with can be one of the following:
# (a) None : the predictions are assumed to be 'raw' and should be used as is
# when computing the metrics; the names for the final columns are
# 'raw', 'raw_trim' and 'raw_trim_round'.
# (b) 'asis' : the predictions are assumed to be pre-scaled and should be used as is
# when computing the metrics; the names for the final columns are
# 'scale', 'scale_trim' and 'scale_trim_round'.
# (c) a CSV file : the predictions are assumed to be 'raw' and should be scaled
# before computing the metrics; the names for the final columns are
# 'scale', 'scale_trim' and 'scale_trim_round'.
# Check whether we want to do scaling
do_scaling = (scale_with is not None and scale_with != 'asis')
# The paths to files and names for data container properties
paths = ['predictions_file']
names = ['predictions']
# If we want to do scaling, get the scale file
if do_scaling:
# Make sure scale file can be located
scale_file_location = DataReader.locate_files(scale_with, configpath)
if not scale_file_location:
raise FileNotFoundError('Could not find scaling file {}.'
''.format(scale_file_location))
paths.append('scale_with')
names.append('scale')
# Get the paths, names, and converters for the DataReader
(file_names, file_paths) = configuration.get_names_and_paths(paths, names)
file_paths = DataReader.locate_files(file_paths, configpath)
converters = {'predictions': configuration.get_default_converter()}
logger.info('Reading predictions: {}.'.format(
configuration['predictions_file']))
# Initialize the reader
reader = DataReader(file_paths, file_names, converters)
data_container = reader.read()
logger.info('Preprocessing predictions.')
# Initialize the processor
processor = FeaturePreprocessor()
(processed_config,
processed_container) = processor.process_data(configuration,
data_container,
context='rsmeval')
logger.info('Saving pre-processed predictions and metadata to disk.')
writer.write_experiment_output(csvdir,
processed_container,
new_names_dict={
'pred_test': 'pred_processed',
'test_excluded':
'test_excluded_responses'
},
file_format=file_format)
# Initialize the analyzer
analyzer = Analyzer()
# do the data composition stats
(analyzed_config,
analyzed_container) = analyzer.run_data_composition_analyses_for_rsmeval(
processed_container, processed_config)
# Write out files
writer.write_experiment_output(csvdir,
analyzed_container,
file_format=file_format)
for_pred_data_container = analyzed_container + processed_container
# run the analyses on the predictions of the model`
logger.info('Running analyses on predictions.')
(pred_analysis_config,
pred_analysis_data_container) = analyzer.run_prediction_analyses(
for_pred_data_container, analyzed_config)
writer.write_experiment_output(csvdir,
pred_analysis_data_container,
reset_index=True,
file_format=file_format)
# Initialize reporter
reporter = Reporter()
# generate the report
logger.info('Starting report generation.')
reporter.create_report(processed_config, csvdir, figdir, context='rsmeval')
def run_experiment(config_file_or_obj,
output_dir):
"""
Run RSMTool experiment using the given configuration
file and generate all outputs in the given directory.
Parameters
----------
config_file_or_obj : str or Configuration
Path to the experiment configuration file.
Users can also pass a `Configuration` object that is in memory.
output_dir : str
Path to the experiment output directory.
Raises
------
ValueError
If any of the required fields are missing or ill-specified.
"""
logger = logging.getLogger(__name__)
# create the 'output' and the 'figure' sub-directories
# where all the experiment output such as the CSV files
# and the box plots will be saved
# Get absolute paths to output directories
csvdir = abspath(join(output_dir, 'output'))
figdir = abspath(join(output_dir, 'figure'))
reportdir = abspath(join(output_dir, 'report'))
featuredir = abspath(join(output_dir, 'feature'))
# Make directories, if necessary
makedirs(csvdir, exist_ok=True)
makedirs(figdir, exist_ok=True)
makedirs(reportdir, exist_ok=True)
# Allow users to pass Configuration object to the
# `config_file_or_obj` argument, rather than read from file
if not isinstance(config_file_or_obj, Configuration):
# Instantiate configuration parser object
parser = ConfigurationParser.get_configparser(config_file_or_obj)
configuration = parser.read_normalize_validate_and_process_config(config_file_or_obj)
# get the directory where the configuration file lives
configpath = dirname(config_file_or_obj)
else:
configuration = config_file_or_obj
if configuration.filepath is not None:
configpath = dirname(configuration.filepath)
else:
configpath = getcwd()
logger.info('Saving configuration file.')
configuration.save(output_dir)
# Get output format
file_format = configuration.get('file_format', 'csv')
# Get DataWriter object
writer = DataWriter(configuration['experiment_id'])
# Get the paths and names for the DataReader
(file_names,
file_paths_org) = configuration.get_names_and_paths(['train_file', 'test_file',
'features',
'feature_subset_file'],
['train', 'test',
'feature_specs',
'feature_subset_specs'])
file_paths = DataReader.locate_files(file_paths_org, configpath)
# if there are any missing files after trying to locate
# all expected files, raise an error
if None in file_paths:
missing_file_paths = [file_paths_org[idx] for idx, path in enumerate(file_paths)
if path is None]
raise FileNotFoundError('The following files were not found: '
'{}'.format(repr(missing_file_paths)))
# Use the default converter for both train and test
converters = {'train': configuration.get_default_converter(),
'test': configuration.get_default_converter()}
logger.info('Reading in all data from files.')
# Initialize the reader
reader = DataReader(file_paths, file_names, converters)
data_container = reader.read()
logger.info('Preprocessing all features.')
# Initialize the processor
processor = FeaturePreprocessor()
(processed_config,
processed_container) = processor.process_data(configuration,
data_container)
# Rename certain frames with more descriptive names
# for writing out experiment files
rename_dict = {'train_excluded': 'train_excluded_responses',
'test_excluded': 'test_excluded_responses',
'train_length': 'train_response_lengths',
'train_flagged': 'train_responses_with_excluded_flags',
'test_flagged': 'test_responses_with_excluded_flags'}
logger.info('Saving training and test set data to disk.')
# Write out files
writer.write_experiment_output(csvdir,
processed_container,
['train_features',
'test_features',
'train_metadata',
'test_metadata',
'train_other_columns',
'test_other_columns',
'train_preprocessed_features',
'test_preprocessed_features',
'train_excluded',
'test_excluded',
'train_length',
'test_human_scores',
'train_flagged',
'test_flagged'],
rename_dict,
file_format=file_format)
# Initialize the analyzer
analyzer = Analyzer()
(analyzed_config,
analyzed_container) = analyzer.run_data_composition_analyses_for_rsmtool(processed_container,
processed_config)
# Write out files
writer.write_experiment_output(csvdir,
analyzed_container,
file_format=file_format)
logger.info('Training {} model.'.format(processed_config['model_name']))
# Initialize modeler
modeler = Modeler()
modeler.train(processed_config,
processed_container,
csvdir,
figdir,
file_format)
# Identify the features used by the model
selected_features = modeler.get_feature_names()
# Add selected features to processed configuration
processed_config['selected_features'] = selected_features
# Write out files
writer.write_feature_csv(featuredir,
processed_container,
selected_features,
file_format=file_format)
features_data_container = processed_container.copy()
# Get selected feature info, and write out to file
df_feature_info = features_data_container.feature_info.copy()
df_selected_feature_info = df_feature_info[df_feature_info['feature'].isin(selected_features)]
selected_feature_dataset_dict = {'name': 'selected_feature_info',
'frame': df_selected_feature_info}
features_data_container.add_dataset(selected_feature_dataset_dict,
update=True)
writer.write_experiment_output(csvdir,
features_data_container,
dataframe_names=['selected_feature_info'],
new_names_dict={'selected_feature_info': 'feature'},
file_format=file_format)
logger.info('Running analyses on training set.')
(train_analyzed_config,
train_analyzed_container) = analyzer.run_training_analyses(processed_container,
processed_config)
# Write out files
writer.write_experiment_output(csvdir,
train_analyzed_container,
reset_index=True,
file_format=file_format)
# Use only selected features for predictions
columns_for_prediction = ['spkitemid', 'sc1'] + selected_features
train_for_prediction = processed_container.train_preprocessed_features[columns_for_prediction]
test_for_prediction = processed_container.test_preprocessed_features[columns_for_prediction]
logged_str = 'Generating training and test set predictions'
logged_str += ' (expected scores).' if configuration['predict_expected_scores'] else '.'
logger.info(logged_str)
(pred_config,
pred_data_container) = modeler.predict_train_and_test(train_for_prediction,
test_for_prediction,
processed_config)
# Write out files
writer.write_experiment_output(csvdir,
pred_data_container,
new_names_dict={'pred_test': 'pred_processed'},
file_format=file_format)
original_coef_file = join(csvdir, '{}_coefficients.{}'.format(pred_config['experiment_id'],
file_format))
# If coefficients file exists, then generate
# scaled coefficients and save to file
if exists(original_coef_file):
logger.info('Scaling the coefficients and saving them to disk')
try:
# Scale coefficients, and return DataContainer w/ scaled coefficients
scaled_data_container = modeler.scale_coefficients(pred_config)
# Write out files to disk
writer.write_experiment_output(csvdir,
scaled_data_container,
file_format=file_format)
except AttributeError:
raise ValueError("It appears you are trying to save two different "
"experiments to the same directory using the same "
"ID. Please clear the content of the directory and "
"rerun both experiments using different "
"experiment IDs.")
# Add processed data_container frames to pred_data_container
new_pred_data_container = pred_data_container + processed_container
logger.info('Running prediction analyses.')
(pred_analysis_config,
pred_analysis_data_container) = analyzer.run_prediction_analyses(new_pred_data_container,
pred_config)
# Write out files
writer.write_experiment_output(csvdir,
pred_analysis_data_container,
reset_index=True,
file_format=file_format)
# Initialize reporter
reporter = Reporter()
# generate the report
logger.info('Starting report generation.')
reporter.create_report(processed_config,
csvdir,
figdir)
def compute_and_save_predictions(config_file_or_obj,
output_file,
feats_file=None):
"""
Run ``rsmpredict`` with given configuration file and generate
predictions (and, optionally, pre-processed feature values).
Parameters
----------
config_file_or_obj : str or configuration_parser.Configuration
Path to the experiment configuration file.
Users can also pass a `Configuration` object that is in memory.
output_dir : str
Path to the output directory for saving files.
feats_file (optional): str
Path to the output file for saving preprocessed feature values.
Raises
------
ValueError
If any of the required fields are missing or ill-specified.
"""
logger = logging.getLogger(__name__)
# Allow users to pass Configuration object to the
# `config_file_or_obj` argument, rather than read file
if not isinstance(config_file_or_obj, Configuration):
# Instantiate configuration parser object
parser = ConfigurationParser.get_configparser(config_file_or_obj)
config = parser.read_normalize_validate_and_process_config(
config_file_or_obj, context='rsmpredict')
# get the directory where the config file lives
configpath = dirname(config_file_or_obj)
else:
config = config_file_or_obj
if config.filepath is not None:
configpath = dirname(config.filepath)
else:
configpath = os.getcwd()
# get the experiment ID
experiment_id = config['experiment_id']
# Get output format
file_format = config.get('file_format', 'csv')
# Get DataWriter object
writer = DataWriter(experiment_id)
# get the input file containing the feature values
# for which we want to generate the predictions
input_features_file = DataReader.locate_files(
config['input_features_file'], configpath)
if not input_features_file:
raise FileNotFoundError('Input file {} does not exist'
''.format(config['input_features_file']))
experiment_dir = DataReader.locate_files(config['experiment_dir'],
configpath)
if not experiment_dir:
raise FileNotFoundError('The directory {} does not exist.'
''.format(config['experiment_dir']))
else:
experiment_output_dir = normpath(join(experiment_dir, 'output'))
if not exists(experiment_output_dir):
raise FileNotFoundError(
'The directory {} does not contain '
'the output of an rsmtool experiment.'.format(experiment_dir))
# find all the .model files in the experiment output directory
model_files = glob.glob(join(experiment_output_dir, '*.model'))
if not model_files:
raise FileNotFoundError(
'The directory {} does not contain any rsmtool models.'
''.format(experiment_output_dir))
experiment_ids = [splitext(basename(mf))[0] for mf in model_files]
if experiment_id not in experiment_ids:
raise FileNotFoundError(
'{} does not contain a model for the experiment "{}". '
'The following experiments are contained in this '
'directory: {}'.format(experiment_output_dir, experiment_id,
experiment_ids))
# check that the directory contains outher required files
required_file_types = ['feature', 'postprocessing_params']
for file_type in required_file_types:
expected_file_name = "{}_{}.csv".format(experiment_id, file_type)
if not exists(join(experiment_output_dir, expected_file_name)):
raise FileNotFoundError('{} does not contain the required file '
'{} that was generated during the '
'original model training'.format(
experiment_output_dir,
expected_file_name))
# model_files = glob.glob(join(experiment_output_dir, '*.model'))
# if not model_files:
# raise FileNotFoundError('The directory {} does not contain any rsmtool models. '
# ''.format(experiment_output_dir))
logger.info('Reading input files.')
feature_info = join(experiment_output_dir,
'{}_feature.csv'.format(experiment_id))
post_processing = join(
experiment_output_dir,
'{}_postprocessing_params.csv'.format(experiment_id))
file_paths = [input_features_file, feature_info, post_processing]
file_names = ['input_features', 'feature_info', 'postprocessing_params']
converters = {'input_features': config.get_default_converter()}
# Initialize the reader
reader = DataReader(file_paths, file_names, converters)
data_container = reader.read(
kwargs_dict={'feature_info': {
'index_col': 0
}})
# load the Modeler to generate the predictions
model = Modeler.load_from_file(
join(experiment_output_dir, '{}.model'.format(experiment_id)))
# Add the model to the configuration object
config['model'] = model
# Initialize the processor
processor = FeaturePreprocessor()
(processed_config,
processed_container) = processor.process_data(config,
data_container,
context='rsmpredict')
# save the pre-processed features to disk if we were asked to
if feats_file is not None:
logger.info(
'Saving pre-processed feature values to {}'.format(feats_file))
feats_dir = dirname(feats_file)
# create any directories needed for the output file
os.makedirs(feats_dir, exist_ok=True)
_, feats_filename = split(feats_file)
feats_filename, _ = splitext(feats_filename)
# Write out files
writer.write_experiment_output(
feats_dir,
processed_container,
include_experiment_id=False,
dataframe_names=['features_processed'],
new_names_dict={'features_processed': feats_filename},
file_format=file_format)
if (output_file.lower().endswith('.csv')
or output_file.lower().endswith('.xlsx')):
output_dir = dirname(output_file)
_, filename = split(output_file)
filename, _ = splitext(filename)
else:
output_dir = output_file
filename = 'predictions_with_metadata'
# create any directories needed for the output file
os.makedirs(output_dir, exist_ok=True)
# save the predictions to disk
logger.info('Saving predictions.')
# Write out files
writer.write_experiment_output(
output_dir,
processed_container,
include_experiment_id=False,
dataframe_names=['predictions_with_metadata'],
new_names_dict={'predictions_with_metadata': filename},
file_format=file_format)
# save excluded responses to disk
if not processed_container.excluded.empty:
# save the predictions to disk
logger.info('Saving excluded responses to {}'.format(
join(output_dir, '{}_excluded_responses.csv'
''.format(filename))))
# Write out files
writer.write_experiment_output(output_dir,
processed_container,
include_experiment_id=False,
dataframe_names=['excluded'],
new_names_dict={
'excluded':
'{}_excluded_responses'
''.format(filename)
},
file_format=file_format)
