def check_scaled_coefficients(source, experiment_id, file_format='csv'):
"""
Check that the predictions generated using scaled
coefficients match the scaled scores. Raises an
AssertionError if they do not.
Parameters
----------
source : str
Path to the source directory on disk.
experiment_id : str
The experiment ID.
file_format : str, optional
The format of the output files.
Defaults to 'csv'.
"""
preprocessed_test_file = join('test_outputs',
source,
'output',
'{}_test_preprocessed_features.{}'.format(experiment_id,
file_format))
scaled_coefficients_file = join('test_outputs',
source,
'output',
'{}_coefficients_scaled.{}'.format(experiment_id,
file_format))
predictions_file = join('test_outputs',
source,
'output',
'{}_pred_processed.{}'.format(experiment_id,
file_format))
postprocessing_params_file = join('test_outputs',
source,
'output',
'{}_postprocessing_params.{}'.format(experiment_id,
file_format))
postproc_params = DataReader.read_from_file(postprocessing_params_file).loc[0]
df_preprocessed_test_data = DataReader.read_from_file(preprocessed_test_file)
df_old_predictions = DataReader.read_from_file(predictions_file)
df_old_predictions = df_old_predictions[['spkitemid', 'sc1', 'scale']]
# create fake skll objects with new coefficients
df_coef = DataReader.read_from_file(scaled_coefficients_file)
learner = Modeler.create_fake_skll_learner(df_coef)
modeler = Modeler.load_from_learner(learner)
# generate new predictions and rename the prediction column to 'scale'
df_new_predictions = modeler.predict(df_preprocessed_test_data,
postproc_params['trim_min'],
postproc_params['trim_max'])
df_new_predictions.rename(columns={'raw': 'scale'}, inplace=True)
# check that new predictions match the scaled old predictions
assert_frame_equal(df_new_predictions.sort_index(axis=1),
df_old_predictions.sort_index(axis=1),
check_exact=False,
check_less_precise=True)
def check_subgroup_outputs(output_dir, experiment_id, subgroups, file_format='csv'):
"""
Check to make sure that the subgroup outputs
look okay. Raise an AssertionError if they do not.
Parameters
----------
output_dir : str
Path to the `output` experiment output directory for a test.
experiment_id : str
The experiment ID.
subgroups : list of str
List of column names that contain grouping
information.
file_format : str, optional
The format of the output files.
Defaults to 'csv'.
"""
train_preprocessed_file = join(output_dir,
'{}_train_metadata.{}'.format(experiment_id,
file_format))
train_preprocessed = DataReader.read_from_file(train_preprocessed_file, index_col=0)
test_preprocessed_file = join(output_dir,
'{}_test_metadata.{}'.format(experiment_id,
file_format))
test_preprocessed = DataReader.read_from_file(test_preprocessed_file,
index_col=0)
for group in subgroups:
ok_(group in train_preprocessed.columns)
ok_(group in test_preprocessed.columns)
# check that the total sum of N per category matches the total N
# in data composition and the total N categories matches what is
# in overall data composition
file_data_composition_all = join(output_dir,
'{}_data_composition.{}'.format(experiment_id,
file_format))
df_data_composition_all = DataReader.read_from_file(file_data_composition_all)
for group in subgroups:
file_composition_by_group = join(output_dir,
'{}_data_composition_by_{}.{}'.format(experiment_id,
group,
file_format))
composition_by_group = DataReader.read_from_file(file_composition_by_group)
for partition in ['Training', 'Evaluation']:
partition_info = df_data_composition_all.loc[df_data_composition_all['partition'] ==
partition]
summation = sum(composition_by_group['{} set'
''.format(partition)])
ok_(summation == partition_info.iloc[0]['responses'])
length = len(composition_by_group.loc[composition_by_group['{} set'
''.format(partition)] != 0])
ok_(length == partition_info.iloc[0][group])
def check_scaled_coefficients(source, experiment_id, file_format='csv'):
"""
Check that the predictions generated using scaled
coefficients match the scaled scores. Raises an
AssertionError if they do not.
Parameters
----------
source : str
Path to the source directory on disk.
experiment_id : str
The experiment ID.
file_format : str, optional
The format of the output files.
Defaults to 'csv'.
"""
preprocessed_test_file = join(
'test_outputs', source, 'output',
'{}_test_preprocessed_features.{}'.format(experiment_id, file_format))
scaled_coefficients_file = join(
'test_outputs', source, 'output',
'{}_coefficients_scaled.{}'.format(experiment_id, file_format))
predictions_file = join(
'test_outputs', source, 'output',
'{}_pred_processed.{}'.format(experiment_id, file_format))
postprocessing_params_file = join(
'test_outputs', source, 'output',
'{}_postprocessing_params.{}'.format(experiment_id, file_format))
postproc_params = DataReader.read_from_file(
postprocessing_params_file).loc[0]
df_preprocessed_test_data = DataReader.read_from_file(
preprocessed_test_file)
df_old_predictions = DataReader.read_from_file(predictions_file)
df_old_predictions = df_old_predictions[['spkitemid', 'sc1', 'scale']]
# create fake skll objects with new coefficients
df_coef = DataReader.read_from_file(scaled_coefficients_file)
learner = Modeler.create_fake_skll_learner(df_coef)
modeler = Modeler.load_from_learner(learner)
# generate new predictions and rename the prediction column to 'scale'
df_new_predictions = modeler.predict(df_preprocessed_test_data,
postproc_params['trim_min'],
postproc_params['trim_max'])
df_new_predictions.rename(columns={'raw': 'scale'}, inplace=True)
# check that new predictions match the scaled old predictions
assert_frame_equal(df_new_predictions.sort_index(axis=1),
df_old_predictions.sort_index(axis=1),
check_exact=False,
check_less_precise=True)
def check_read_from_file(self, extension):
"""Test whether ``read_from_file()`` works as expected."""
name = TestDataReader.make_file_from_ext(self.df_train, extension)
# now read in the file using `read_data_file()`
df_read = DataReader.read_from_file(name,
converters={'id': str, 'candidate': str})
# Make sure we get rid of the file at the end,
# at least if we get to this point (i.e. no errors raised)
self.filepaths.append(name)
assert_frame_equal(self.df_train, df_read)
def check_read_from_file(self, extension):
"""
Test whether the ``read_from_file()`` method works as expected.
"""
name = TestDataReader.make_file_from_ext(self.df_train, extension)
# now read in the file using `read_data_file()`
df_read = DataReader.read_from_file(name,
converters={'id': str, 'candidate': str})
# Make sure we get rid of the file at the end,
# at least if we get to this point (i.e. no errors raised)
self.filepaths.append(name)
assert_frame_equal(self.df_train, df_read)
def check_subgroup_outputs(output_dir,
experiment_id,
subgroups,
file_format='csv'):
"""
Check to make sure that the subgroup outputs
look okay. Raise an AssertionError if they do not.
Parameters
----------
output_dir : str
Path to the `output` experiment output directory for a test.
experiment_id : str
The experiment ID.
subgroups : list of str
List of column names that contain grouping
information.
file_format : str, optional
The format of the output files.
Defaults to 'csv'.
"""
train_preprocessed_file = join(
output_dir, '{}_train_metadata.{}'.format(experiment_id, file_format))
train_preprocessed = DataReader.read_from_file(train_preprocessed_file,
index_col=0)
test_preprocessed_file = join(
output_dir, '{}_test_metadata.{}'.format(experiment_id, file_format))
test_preprocessed = DataReader.read_from_file(test_preprocessed_file,
index_col=0)
for group in subgroups:
ok_(group in train_preprocessed.columns)
ok_(group in test_preprocessed.columns)
# check that the total sum of N per category matches the total N
# in data composition and the total N categories matches what is
# in overall data composition
file_data_composition_all = join(
output_dir, '{}_data_composition.{}'.format(experiment_id,
file_format))
df_data_composition_all = DataReader.read_from_file(
file_data_composition_all)
for group in subgroups:
file_composition_by_group = join(
output_dir,
'{}_data_composition_by_{}.{}'.format(experiment_id, group,
file_format))
composition_by_group = DataReader.read_from_file(
file_composition_by_group)
for partition in ['Training', 'Evaluation']:
partition_info = df_data_composition_all.loc[
df_data_composition_all['partition'] == partition]
summation = sum(composition_by_group['{} set'
''.format(partition)])
ok_(summation == partition_info.iloc[0]['responses'])
length = len(composition_by_group.loc[
composition_by_group['{} set'
''.format(partition)] != 0])
ok_(length == partition_info.iloc[0][group])
def check_file_output(file1, file2, file_format='csv'):
"""
Check if two experiment files have values that are
the same to within three decimal places. Raises an
AssertionError if they are not.
Parameters
----------
file1 : str
Path to the first file.
file2 : str
Path to the second files.
file_format : str, optional
The format of the output files.
Defaults to 'csv'.
"""
# make sure that the main id columns are read as strings since
# this may affect merging in custom notebooks
string_columns = ['spkitemid', 'candidate']
converter_dict = {column: str for column in string_columns}
df1 = DataReader.read_from_file(file1, converters=converter_dict)
df2 = DataReader.read_from_file(file2, converters=converter_dict)
# if the first column is numeric, just force the index to string;
# however, if it is non-numeric, set it as the index and then
# force it to string. We do this to ensure string indices are
# preserved as such
for df in [df1, df2]:
if np.issubdtype(df[df.columns[0]].dtype, np.number):
df.index = df.index.map(str)
else:
df.index = df[df.columns[0]]
df.index = df.index.map(str)
# sort all the indices alphabetically
df1.sort_index(inplace=True)
df2.sort_index(inplace=True)
# convert any integer columns to floats in either data frame
for df in [df1, df2]:
for c in df.columns:
if df[c].dtype == np.int64:
df[c] = df[c].astype(np.float64)
# do the same for indices
for df in [df1, df2]:
if df.index.dtype == np.int64:
df.index = df.index.astype(np.float64)
# for pca and factor correlations convert all values to absolutes
# because the sign may not always be the same
if (file1.endswith('pca.{}'.format(file_format))
or file1.endswith('factor_correlations.{}'.format(file_format))):
for df in [df1, df2]:
msk = df.dtypes == np.float64
df.loc[:, msk] = df.loc[:, msk].abs()
try:
assert_frame_equal(df1.sort_index(axis=1),
df2.sort_index(axis=1),
check_exact=False,
check_less_precise=True)
except AssertionError as e:
message = e.args[0]
new_message = 'File {} - {}'.format(basename(file1), message)
e.args = (new_message, )
raise
def load_rsmtool_output(self, filedir, figdir, experiment_id, prefix, groups_eval):
"""
Function to load all of the outputs of an rsmtool experiment.
For each type of output, we first check whether the file exists
to allow comparing experiments with different sets of outputs.
Parameters
----------
filedir : str
Path to the directory containing output files.
figdir : str
Path to the directory containing output figures.
experiment_id : str
Original ``experiment_id`` used to generate the output files.
prefix: str
Must be set to ``scale`` or ``raw``. Indicates whether the score
is scaled or not.
groups_eval: list
List of subgroup names used for subgroup evaluation.
Returns
-------
files : dict
A dictionary with outputs converted to pandas data
frames. If a particular type of output did not exist for the
experiment, its value will be an empty data frame.
figs: dict
A dictionary with experiment figures.
"""
file_format = get_output_directory_extension(filedir, experiment_id)
files = defaultdict(pd.DataFrame)
figs = {}
# feature distributions and the inter-feature correlations
feature_train_file = join(filedir, '{}_train_features.{}'.format(experiment_id,
file_format))
if exists(feature_train_file):
files['df_train_features'] = DataReader.read_from_file(feature_train_file)
feature_distplots_file = join(figdir, '{}_distrib.svg'.format(experiment_id))
if exists(feature_distplots_file):
figs['feature_distplots'] = feature_distplots_file
feature_cors_file = join(filedir, '{}_cors_processed.{}'.format(experiment_id,
file_format))
if exists(feature_cors_file):
files['df_feature_cors'] = DataReader.read_from_file(feature_cors_file, index_col=0)
# df_scores
scores_file = join(filedir, '{}_pred_processed.{}'.format(experiment_id,
file_format))
if exists(scores_file):
df_scores = DataReader.read_from_file(scores_file, converters={'spkitemid': str})
files['df_scores'] = df_scores[['spkitemid', 'sc1', prefix]]
# model coefficients if present
betas_file = join(filedir, '{}_betas.{}'.format(experiment_id,
file_format))
if exists(betas_file):
files['df_coef'] = DataReader.read_from_file(betas_file, index_col=0)
files['df_coef'].index.name = None
# read in the model fit files if present
model_fit_file = join(filedir, '{}_model_fit.{}'.format(experiment_id,
file_format))
if exists(model_fit_file):
files['df_model_fit'] = DataReader.read_from_file(model_fit_file)
# human human agreement
consistency_file = join(filedir, '{}_consistency.{}'.format(experiment_id,
file_format))
# load if consistency file is present
if exists(consistency_file):
df_consistency = DataReader.read_from_file(consistency_file, index_col=0)
files['df_consistency'] = df_consistency
# degradation
degradation_file = join(filedir, "{}_degradation.{}".format(experiment_id,
file_format))
# load if degradation file is present
if exists(degradation_file):
df_degradation = DataReader.read_from_file(degradation_file, index_col=0)
files['df_degradation'] = df_degradation
# disattenuated correlations
dis_corr_file = join(filedir, "{}_disattenuated_correlations.{}".format(experiment_id,
file_format))
# load if disattenuated correlations is present
if exists(dis_corr_file):
df_dis_corr = DataReader.read_from_file(dis_corr_file, index_col=0)
# we only use the row for raw_trim or scale_trim score
files['df_disattenuated_correlations'] = df_dis_corr.loc[['{}_trim'.format(prefix)]]
# read in disattenuated correlations by group
for group in groups_eval:
group_dis_corr_file = join(filedir,
'{}_disattenuated_correlations_by_{}.{}'.format(experiment_id,
group,
file_format))
if exists(group_dis_corr_file):
df_dis_cor_group = DataReader.read_from_file(group_dis_corr_file, index_col=0)
files['df_disattenuated_correlations_by_{}'.format(group)] = df_dis_cor_group
files['df_disattenuated_correlations_by_{}_overview'.format(group)] = self.make_summary_stat_df(df_dis_cor_group)
# true score evaluations
true_score_eval_file = join(filedir, "{}_true_score_eval.{}".format(experiment_id,
file_format))
# load true score evaluations if present
if exists(true_score_eval_file):
df_true_score_eval = DataReader.read_from_file(true_score_eval_file, index_col=0)
# we only use the row for raw_trim or scale_trim score
files['df_true_score_eval'] = df_true_score_eval.loc[['{}_trim'.format(prefix)]]
# use the raw columns or the scale columns depending on the prefix
existing_eval_cols = (_df_eval_columns_existing_raw if prefix == 'raw'
else _df_eval_columns_existing_scale)
rename_dict = raw_rename_dict if prefix == 'raw' else scale_rename_dict
# read in the short version of the evaluation metrics for all data
short_metrics_list = ["N", "Adj. Agmt.(br)", "Agmt.(br)", "K(br)",
"Pearson(b)", "QWK(b)", "R2(b)", "RMSE(b)"]
eval_file_short = join(filedir, '{}_eval_short.{}'.format(experiment_id, file_format))
if exists(eval_file_short):
df_eval = DataReader.read_from_file(eval_file_short, index_col=0)
(rename_dict_new,
existing_eval_cols_new,
short_metrics_list_new,
_) = self._modify_eval_columns_to_ensure_version_compatibilty(df_eval,
rename_dict,
existing_eval_cols,
short_metrics_list)
df_eval = df_eval[existing_eval_cols_new]
df_eval = df_eval.rename(columns=rename_dict_new)
files['df_eval'] = df_eval[short_metrics_list_new]
files['df_eval'].index.name = None
eval_file = join(filedir, '{}_eval.{}'.format(experiment_id, file_format))
if exists(eval_file):
files['df_eval_for_degradation'] = DataReader.read_from_file(eval_file, index_col=0)
# read in the evaluation metrics by subgroup, if we are asked to
for group in groups_eval:
group_eval_file = join(filedir, '{}_eval_by_{}.{}'.format(experiment_id,
group,
file_format))
if exists(group_eval_file):
df_eval = DataReader.read_from_file(group_eval_file, index_col=0)
(rename_dict_new,
existing_eval_cols_new,
short_metrics_list_new,
smd_name
) = self._modify_eval_columns_to_ensure_version_compatibilty(df_eval,
rename_dict,
existing_eval_cols,
short_metrics_list,
raise_warnings=False)
# if `SMD` is being used, rather than `DSM`, we print a note for the user; we don't
# want to go so far as to raise a warning, but we do want to give the user some info
if smd_name == 'SMD':
warnings.warn("The subgroup evaluations in `{}` use 'SMD'. Please note "
"that newer versions of RSMTool (7.0 or greater) use 'DSM' with subgroup "
"evaluations. For additional details on how these metrics "
"differ, see the RSMTool documentation. Comparisons with experiments "
"using SMD for subgroup calculations will be deprecated in the next major "
"release.".format(group_eval_file), category=DeprecationWarning)
df_eval = df_eval[existing_eval_cols_new]
df_eval = df_eval.rename(columns=rename_dict_new)
files['df_eval_by_{}'.format(group)] = df_eval[short_metrics_list_new]
files['df_eval_by_{}'.format(group)].index.name = None
series = files['df_eval_by_{}'.format(group)]
files['df_eval_by_{}_overview'.format(group)] = self.make_summary_stat_df(series)
# set the ordering of mean/SD/SMD statistics
files['df_eval_by_{}_m_sd'.format(group)] = df_eval[['N', 'H1 mean',
'H1 SD', 'score mean(br)',
'score SD(br)',
'score mean(b)',
'score SD(b)',
'{}(br)'.format(smd_name),
'{}(b)'.format(smd_name)]]
files['df_eval_by_{}_m_sd'.format(group)].index.name = None
# read in the partial correlations vs. score for all data
pcor_score_file = join(filedir, '{}_pcor_score_all_data.{}'.format(experiment_id,
file_format))
if exists(pcor_score_file):
files['df_pcor_sc1'] = DataReader.read_from_file(pcor_score_file, index_col=0)
files['df_pcor_sc1_overview'] = self.make_summary_stat_df(files['df_pcor_sc1'])
# read in the partial correlations by subgroups, if we are asked to
for group in groups_eval:
group_pcor_file = join(filedir, '{}_pcor_score_by_{}.{}'.format(experiment_id,
group,
file_format))
if exists(group_pcor_file):
files['df_pcor_sc1_by_{}'
''.format(group)] = DataReader.read_from_file(group_pcor_file,
index_col=0)
series = files['df_pcor_sc1_by_{}'.format(group)]
files['df_pcor_sc1_{}_overview'.format(group)] = self.make_summary_stat_df(series)
# read in the marginal correlations vs. score for all data
mcor_score_file = join(filedir, '{}_margcor_score_all_data.{}'.format(experiment_id,
file_format))
if exists(mcor_score_file):
files['df_mcor_sc1'] = DataReader.read_from_file(mcor_score_file, index_col=0)
files['df_mcor_sc1_overview'] = self.make_summary_stat_df(files['df_mcor_sc1'])
# read in the partial correlations by subgroups, if we are asked to
for group in groups_eval:
group_mcor_file = join(filedir,
'{}_margcor_score_by_{}.{}'.format(experiment_id,
group,
file_format))
if exists(group_mcor_file):
files['df_mcor_sc1_by_{}'
''.format(group)] = DataReader.read_from_file(group_mcor_file,
index_col=0)
series = files['df_mcor_sc1_by_{}'.format(group)]
files['df_mcor_sc1_{}_overview'.format(group)] = self.make_summary_stat_df(series)
pca_file = join(filedir, '{}_pca.{}'.format(experiment_id, file_format))
if exists(pca_file):
files['df_pca'] = DataReader.read_from_file(pca_file, index_col=0)
files['df_pcavar'] = DataReader.read_from_file(join(filedir,
'{}_pcavar.{}'.format(experiment_id,
file_format)),
index_col=0)
descriptives_file = join(filedir, '{}_feature_descriptives.{}'.format(experiment_id,
file_format))
if exists(descriptives_file):
# we read all files pertaining to the descriptive analysis together
# since we merge the outputs
files['df_descriptives'] = DataReader.read_from_file(descriptives_file, index_col=0)
# this df contains only the number of features. this is used later
# for another two tables to show the number of features
df_features_n_values = files['df_descriptives'][['N', 'min', 'max']]
files['df_descriptives'] = files['df_descriptives'][['N', 'mean', 'std. dev.',
'skewness', 'kurtosis']]
outliers_file = join(filedir, '{}_feature_outliers.{}'.format(experiment_id,
file_format))
df_outliers = DataReader.read_from_file(outliers_file, index_col=0)
df_outliers = df_outliers.rename(columns={'upper': 'Upper',
'lower': 'Lower',
'both': 'Both',
'upperperc': 'Upper %',
'lowerperc': 'Lower %',
'bothperc': 'Both %'})
df_outliers_columns = df_outliers.columns.tolist()
files['df_outliers'] = df_outliers
# join with df_features_n_values to get the value of N
files['df_outliers'] = pd.merge(files['df_outliers'], df_features_n_values,
left_index=True,
right_index=True)[['N'] + df_outliers_columns]
# join with df_features_n_values to get the value of N
percentiles_file = join(filedir, '{}_feature_descriptives'
'Extra.{}'.format(experiment_id,
file_format))
files['df_percentiles'] = DataReader.read_from_file(percentiles_file,
index_col=0)
files['df_percentiles'] = pd.merge(files['df_percentiles'],
df_features_n_values,
left_index=True,
right_index=True)
mild_outliers = (files['df_percentiles']["Mild outliers"] /
files['df_percentiles']["N"].astype(float) * 100)
files['df_percentiles']["Mild outliers (%)"] = mild_outliers
extreme_outliers = (files['df_percentiles']["Extreme outliers"] /
files['df_percentiles']["N"].astype(float) * 100)
files['df_percentiles']["Extreme outliers (%)"] = extreme_outliers
files['df_percentiles'] = files['df_percentiles'][['N', 'min', 'max',
'1%', '5%', '25%',
'50%', '75%', '95%',
'99%', 'IQR', 'Mild outliers',
'Mild outliers (%)',
'Extreme outliers',
'Extreme outliers (%)']]
confmatrix_file = join(filedir, '{}_confMatrix.{}'.format(experiment_id, file_format))
if exists(confmatrix_file):
conf_matrix = DataReader.read_from_file(confmatrix_file, index_col=0)
files['df_confmatrix'] = self.process_confusion_matrix(conf_matrix)
score_dist_file = join(filedir, '{}_score_dist.{}'.format(experiment_id, file_format))
if exists(score_dist_file):
df_score_dist = DataReader.read_from_file(score_dist_file, index_col=1)
df_score_dist.rename(columns={'sys_{}'.format(prefix): 'sys'}, inplace=True)
files['df_score_dist'] = df_score_dist[['human', 'sys', 'difference']]
# read in the feature boxplots by subgroup, if we were asked to
for group in groups_eval:
feature_boxplot_prefix = join(figdir,
'{}_feature_boxplot_by_{}'.format(experiment_id, group))
svg_file = join(feature_boxplot_prefix + '.svg')
png_file = join(feature_boxplot_prefix + '.png')
if exists(svg_file):
figs['feature_boxplots_by_{}_svg'.format(group)] = svg_file
elif exists(png_file):
figs['feature_boxplots_by_{}_png'.format(group)] = png_file
# read in the betas image if exists
betas_svg = join(figdir, '{}_betas.svg'.format(experiment_id))
if exists(betas_svg):
figs['betas'] = betas_svg
# read in the evaluation barplots by subgroup, if we were asked to
for group in groups_eval:
eval_barplot_svg_file = join(figdir, '{}_eval_by_{}.svg'.format(experiment_id, group))
if exists(eval_barplot_svg_file):
figs['eval_barplot_by_{}'.format(group)] = eval_barplot_svg_file
pca_svg_file = join(figdir, '{}_pca.svg'.format(experiment_id))
if exists(pca_svg_file):
figs['pca_scree_plot'] = pca_svg_file
return (files, figs, file_format)
def check_file_output(file1, file2, file_format='csv'):
"""
Check if two experiment files have values that are
the same to within three decimal places. Raises an
AssertionError if they are not.
Parameters
----------
file1 : str
Path to the first file.
file2 : str
Path to the second files.
file_format : str, optional
The format of the output files.
Defaults to 'csv'.
"""
# make sure that the main id columns are read as strings since
# this may affect merging in custom notebooks
string_columns = ['spkitemid', 'candidate']
converter_dict = {column: str for column in string_columns}
df1 = DataReader.read_from_file(file1, converters=converter_dict)
df2 = DataReader.read_from_file(file2, converters=converter_dict)
# if the first column is numeric, just force the index to string;
# however, if it is non-numeric, set it as the index and then
# force it to string. We do this to ensure string indices are
# preserved as such
for df in [df1, df2]:
if np.issubdtype(df[df.columns[0]].dtype, np.number):
df.index = df.index.map(str)
else:
df.index = df[df.columns[0]]
df.index = df.index.map(str)
# sort all the indices alphabetically
df1.sort_index(inplace=True)
df2.sort_index(inplace=True)
# convert any integer columns to floats in either data frame
for df in [df1, df2]:
for c in df.columns:
if df[c].dtype == np.int64:
df[c] = df[c].astype(np.float64)
# do the same for indices
for df in [df1, df2]:
if df.index.dtype == np.int64:
df.index = df.index.astype(np.float64)
# for pca and factor correlations convert all values to absolutes
# because the sign may not always be the same
if (file1.endswith('pca.{}'.format(file_format)) or
file1.endswith('factor_correlations.{}'.format(file_format))):
for df in [df1, df2]:
msk = df.dtypes == np.float64
df.loc[:, msk] = df.loc[:, msk].abs()
try:
assert_frame_equal(df1.sort_index(axis=1),
df2.sort_index(axis=1),
check_exact=False,
check_less_precise=True)
except AssertionError as e:
message = e.args[0]
new_message = 'File {} - {}'.format(basename(file1), message)
e.args = (new_message, )
raise
def load_rsmtool_output(self, filedir, figdir, experiment_id, prefix, groups_eval):
"""
Function to load all of the outputs of an rsmtool experiment.
For each type of output, we first check whether the file exists
to allow comparing experiments with different sets of outputs.
Parameters
----------
filedir : str
Path to the directory containing output files.
figdir : str
Path to the directory containing output figures.
experiment_id : str
Original ``experiment_id`` used to generate the output files.
prefix: str
Must be set to ``scale`` or ``raw``. Indicates whether the score
is scaled or not.
groups_eval: list
List of subgroup names used for subgroup evaluation.
Returns
-------
files : dict
A dictionary with outputs converted to pandas data
frames. If a particular type of output did not exist for the
experiment, its value will be an empty data frame.
figs: dict
A dictionary with experiment figures.
"""
file_format = get_output_directory_extension(filedir, experiment_id)
files = defaultdict(pd.DataFrame)
figs = {}
# feature distributions and the inter-feature correlations
feature_train_file = join(filedir, '{}_train_features.{}'.format(experiment_id,
file_format))
if exists(feature_train_file):
files['df_train_features'] = DataReader.read_from_file(feature_train_file)
feature_distplots_file = join(figdir, '{}_distrib.svg'.format(experiment_id))
if exists(feature_distplots_file):
figs['feature_distplots'] = feature_distplots_file
# with open(feature_distplots_file, 'rb') as f:
# figs['feature_distplots'] = base64.b64encode(f.read()).decode('utf-8')
feature_cors_file = join(filedir, '{}_cors_processed.{}'.format(experiment_id,
file_format))
if exists(feature_cors_file):
files['df_feature_cors'] = DataReader.read_from_file(feature_cors_file, index_col=0)
# df_scores
scores_file = join(filedir, '{}_pred_processed.{}'.format(experiment_id,
file_format))
if exists(scores_file):
df_scores = DataReader.read_from_file(scores_file, converters={'spkitemid': str})
files['df_scores'] = df_scores[['spkitemid', 'sc1', prefix]]
# model coefficients if present
betas_file = join(filedir, '{}_betas.{}'.format(experiment_id,
file_format))
if exists(betas_file):
files['df_coef'] = DataReader.read_from_file(betas_file, index_col=0)
files['df_coef'].index.name = None
# read in the model fit files if present
model_fit_file = join(filedir, '{}_model_fit.{}'.format(experiment_id,
file_format))
if exists(model_fit_file):
files['df_model_fit'] = DataReader.read_from_file(model_fit_file)
# human human agreement
consistency_file = join(filedir, '{}_consistency.{}'.format(experiment_id,
file_format))
# load if consistency file is present
if exists(consistency_file):
df_consistency = DataReader.read_from_file(consistency_file, index_col=0)
files['df_consistency'] = df_consistency
# degradation
degradation_file = join(filedir, "{}_degradation.{}".format(experiment_id,
file_format))
# load if degradation file is present
if exists(degradation_file):
df_degradation = DataReader.read_from_file(degradation_file, index_col=0)
files['df_degradation'] = df_degradation
# disattenuated correlations
dis_corr_file = join(filedir, "{}_disattenuated_correlations.{}".format(experiment_id,
file_format))
# load if disattenuated correlations is present
if exists(dis_corr_file):
df_dis_corr = DataReader.read_from_file(dis_corr_file, index_col=0)
# we only use the row for raw_trim or scale_trim score
files['df_disattenuated_correlations'] = df_dis_corr.loc[['{}_trim'.format(prefix)]]
# read in disattenuated correlations by group
for group in groups_eval:
group_dis_corr_file = join(filedir,
'{}_disattenuated_correlations_by_{}.{}'.format(experiment_id,
group,
file_format))
if exists(group_dis_corr_file):
df_dis_cor_group = DataReader.read_from_file(group_dis_corr_file, index_col=0)
files['df_disattenuated_correlations_by_{}'.format(group)] = df_dis_cor_group
files['df_disattenuated_correlations_by_{}_overview'.format(group)] = self.make_summary_stat_df(df_dis_cor_group)
# use the raw columns or the scale columns depending on the prefix
existing_eval_cols = (_df_eval_columns_existing_raw if prefix == 'raw'
else _df_eval_columns_existing_scale)
rename_dict = raw_rename_dict if prefix == 'raw' else scale_rename_dict
# read in the short version of the evaluation metrics for all data
short_metrics_list = ["N", "Adj. Agmt.(br)", "Agmt.(br)", "K(br)",
"Pearson(b)", "QWK(br)", "R2(b)", "RMSE(b)"]
eval_file_short = join(filedir, '{}_eval_short.{}'.format(experiment_id, file_format))
if exists(eval_file_short):
df_eval = DataReader.read_from_file(eval_file_short, index_col=0)
df_eval = df_eval[existing_eval_cols]
df_eval = df_eval.rename(columns=rename_dict)
files['df_eval'] = df_eval[short_metrics_list]
files['df_eval'].index.name = None
eval_file = join(filedir, '{}_eval.{}'.format(experiment_id, file_format))
if exists(eval_file):
files['df_eval_for_degradation'] = DataReader.read_from_file(eval_file, index_col=0)
# read in the evaluation metrics by subgroup, if we are asked to
for group in groups_eval:
group_eval_file = join(filedir, '{}_eval_by_{}.{}'.format(experiment_id,
group,
file_format))
if exists(group_eval_file):
df_eval = DataReader.read_from_file(group_eval_file, index_col=0)
df_eval = df_eval[existing_eval_cols]
df_eval = df_eval.rename(columns=rename_dict)
files['df_eval_by_{}'.format(group)] = df_eval[short_metrics_list]
files['df_eval_by_{}'.format(group)].index.name = None
series = files['df_eval_by_{}'.format(group)]
files['df_eval_by_{}_overview'.format(group)] = self.make_summary_stat_df(series)
# set the ordering of mean/SD/SMD statistics
files['df_eval_by_{}_m_sd'.format(group)] = df_eval[['N', 'H1 mean',
'H1 SD', 'score mean(br)',
'score SD(br)',
'score mean(b)',
'score SD(b)',
'SMD(br)', 'SMD(b)']]
files['df_eval_by_{}_m_sd'.format(group)].index.name = None
# read in the partial correlations vs. score for all data
pcor_score_file = join(filedir, '{}_pcor_score_all_data.{}'.format(experiment_id,
file_format))
if exists(pcor_score_file):
files['df_pcor_sc1'] = DataReader.read_from_file(pcor_score_file, index_col=0)
files['df_pcor_sc1_overview'] = self.make_summary_stat_df(files['df_pcor_sc1'])
# read in the partial correlations by subgroups, if we are asked to
for group in groups_eval:
group_pcor_file = join(filedir, '{}_pcor_score_by_{}.{}'.format(experiment_id,
group,
file_format))
if exists(group_pcor_file):
files['df_pcor_sc1_by_{}'
''.format(group)] = DataReader.read_from_file(group_pcor_file,
index_col=0)
series = files['df_pcor_sc1_by_{}'.format(group)]
files['df_pcor_sc1_{}_overview'.format(group)] = self.make_summary_stat_df(series)
# read in the marginal correlations vs. score for all data
mcor_score_file = join(filedir, '{}_margcor_score_all_data.{}'.format(experiment_id,
file_format))
if exists(mcor_score_file):
files['df_mcor_sc1'] = DataReader.read_from_file(mcor_score_file, index_col=0)
files['df_mcor_sc1_overview'] = self.make_summary_stat_df(files['df_mcor_sc1'])
# read in the partial correlations by subgroups, if we are asked to
for group in groups_eval:
group_mcor_file = join(filedir,
'{}_margcor_score_by_{}.{}'.format(experiment_id,
group,
file_format))
if exists(group_mcor_file):
files['df_mcor_sc1_by_{}'
''.format(group)] = DataReader.read_from_file(group_mcor_file,
index_col=0)
series = files['df_mcor_sc1_by_{}'.format(group)]
files['df_mcor_sc1_{}_overview'.format(group)] = self.make_summary_stat_df(series)
pca_file = join(filedir, '{}_pca.{}'.format(experiment_id, file_format))
if exists(pca_file):
files['df_pca'] = DataReader.read_from_file(pca_file, index_col=0)
files['df_pcavar'] = DataReader.read_from_file(join(filedir,
'{}_pcavar.{}'.format(experiment_id,
file_format)),
index_col=0)
descriptives_file = join(filedir, '{}_feature_descriptives.{}'.format(experiment_id,
file_format))
if exists(descriptives_file):
# we read all files pertaining to the descriptive analysis together
# since we merge the outputs
files['df_descriptives'] = DataReader.read_from_file(descriptives_file, index_col=0)
# this df contains only the number of features. this is used later
# for another two tables to show the number of features
df_features_n_values = files['df_descriptives'][['N', 'min', 'max']]
files['df_descriptives'] = files['df_descriptives'][['N', 'mean', 'std. dev.',
'skewness', 'kurtosis']]
outliers_file = join(filedir, '{}_feature_outliers.{}'.format(experiment_id,
file_format))
df_outliers = DataReader.read_from_file(outliers_file, index_col=0)
df_outliers = df_outliers.rename(columns={'upper': 'Upper',
'lower': 'Lower',
'both': 'Both',
'upperperc': 'Upper %',
'lowerperc': 'Lower %',
'bothperc': 'Both %'})
df_outliers_columns = df_outliers.columns.tolist()
files['df_outliers'] = df_outliers
# join with df_features_n_values to get the value of N
files['df_outliers'] = pd.merge(files['df_outliers'], df_features_n_values,
left_index=True,
right_index=True)[['N'] + df_outliers_columns]
# join with df_features_n_values to get the value of N
percentiles_file = join(filedir, '{}_feature_descriptives'
'Extra.{}'.format(experiment_id,
file_format))
files['df_percentiles'] = DataReader.read_from_file(percentiles_file,
index_col=0)
files['df_percentiles'] = pd.merge(files['df_percentiles'],
df_features_n_values,
left_index=True,
right_index=True)
mild_outliers = (files['df_percentiles']["Mild outliers"] /
files['df_percentiles']["N"].astype(float) * 100)
files['df_percentiles']["Mild outliers (%)"] = mild_outliers
extreme_outliers = (files['df_percentiles']["Extreme outliers"] /
files['df_percentiles']["N"].astype(float) * 100)
files['df_percentiles']["Extreme outliers (%)"] = extreme_outliers
files['df_percentiles'] = files['df_percentiles'][['N', 'min', 'max',
'1%', '5%', '25%',
'50%', '75%', '95%',
'99%', 'IQR', 'Mild outliers',
'Mild outliers (%)',
'Extreme outliers',
'Extreme outliers (%)']]
confmatrix_file = join(filedir, '{}_confMatrix.{}'.format(experiment_id, file_format))
if exists(confmatrix_file):
conf_matrix = DataReader.read_from_file(confmatrix_file, index_col=0)
files['df_confmatrix'] = self.process_confusion_matrix(conf_matrix)
score_dist_file = join(filedir, '{}_score_dist.{}'.format(experiment_id, file_format))
if exists(score_dist_file):
df_score_dist = DataReader.read_from_file(score_dist_file, index_col=1)
df_score_dist.rename(columns={'sys_{}'.format(prefix): 'sys'}, inplace=True)
files['df_score_dist'] = df_score_dist[['human', 'sys', 'difference']]
# read in the feature boxplots by subgroup, if we were asked to
for group in groups_eval:
feature_boxplot_prefix = join(figdir,
'{}_feature_boxplot_by_{}'.format(experiment_id, group))
svg_file = join(feature_boxplot_prefix + '.svg')
png_file = join(feature_boxplot_prefix + '.png')
if exists(svg_file):
figs['feature_boxplots_by_{}_svg'.format(group)] = svg_file
elif exists(png_file):
figs['feature_boxplots_by_{}_png'.format(group)] = png_file
# read in the betas image if exists
betas_svg = join(figdir, '{}_betas.svg'.format(experiment_id))
if exists(betas_svg):
figs['betas'] = betas_svg
# read in the evaluation barplots by subgroup, if we were asked to
for group in groups_eval:
eval_barplot_svg_file = join(figdir, '{}_eval_by_{}.svg'.format(experiment_id, group))
if exists(eval_barplot_svg_file):
figs['eval_barplot_by_{}'.format(group)] = eval_barplot_svg_file
pca_svg_file = join(figdir, '{}_pca.svg'.format(experiment_id))
if exists(pca_svg_file):
figs['pca_scree_plot'] = pca_svg_file
return (files, figs, file_format)