def persist(cls, training_configuration, metrics, data_statistics,
algorithm, sampler):
identifier = str(uuid.uuid4())
base_path = f"{BASE_PATH}/evaluation"
os.makedirs(base_path, exist_ok=True)
evaluation_path = f"{base_path}/{identifier}.evaluation"
dill.dump(metrics[["y_true", "y_pred", "y_probs"]],
open(evaluation_path, "wb"))
obj = cls(
training_configuration_id=training_configuration.id,
algorithm=algorithm.__name__,
sampler=sampler.__name__,
auc_roc_mean=metrics.auc_roc.mean(),
auc_roc_std=metrics.auc_roc.std(),
auc_prc_mean=metrics.auc_prc.mean(),
auc_prc_std=metrics.auc_prc.std(),
f1_mean=metrics.f1.mean(),
f1_std=metrics.f1.std(),
accuracy_mean=metrics.accuracy.mean(),
accuracy_std=metrics.accuracy.std(),
precision_mean=metrics.precision.mean(),
precision_std=metrics.precision.std(),
recall_mean=metrics.recall.mean(),
recall_std=metrics.recall.std(),
average_precision_mean=metrics.average_precision.mean(),
average_precision_std=metrics.average_precision.std(),
num_features_mean=data_statistics.num_features.mean(),
num_features_std=data_statistics.num_features.std(),
num_test_rows_mean=data_statistics.num_test_rows.mean(),
num_test_rows_std=data_statistics.num_test_rows.std(),
num_train_rows_mean=data_statistics.num_train_rows.mean(),
num_train_rows_std=data_statistics.num_train_rows.std(),
num_sampled_train_rows_mean=data_statistics.num_sampled_train_rows.
mean(),
num_sampled_train_rows_std=data_statistics.num_sampled_train_rows.
std(),
incidence_rate_train_mean=data_statistics.incidence_rate_train.
mean(),
incidence_rate_train_std=data_statistics.incidence_rate_train.std(
),
incidence_rate_test_mean=data_statistics.incidence_rate_test.mean(
),
incidence_rate_test_std=data_statistics.incidence_rate_test.std(),
incidence_rate_sampled_train_mean=data_statistics.
incidence_rate_sampled_train.mean(),
incidence_rate_sampled_train_std=data_statistics.
incidence_rate_sampled_train.std(),
evaluation_path=evaluation_path,
)
session.add(obj)
session.commit()
return obj
def persist(cls, comment, version, cohort):
obj = cls(
comment=comment,
version=version,
start_time=time.time(),
end_time=math.inf,
cohort=cohort,
)
session.add(obj)
session.commit()
return obj
def persist(cls, training_configuration, data):
identifier = str(uuid.uuid4())
base_path = f"{BASE_PATH}/data"
os.makedirs(base_path, exist_ok=True)
path = f"{base_path}/{identifier}.data.gz"
data.to_csv(path, index=False)
obj = cls(training_configuration_id=training_configuration.id,
path=path)
session.add(obj)
session.commit()
return obj
def persist(cls, df, comment, version, cohort):
# persist df on disk
identifier = str(uuid.uuid4())
base_path = f"{BASE_PATH}/onset-df"
os.makedirs(base_path, exist_ok=True)
path = f"{base_path}/{identifier}.df.gz"
df.to_csv(path, index=False)
obj = cls(comment=comment,
version=version,
created_at=time.time(),
path=path,
cohort=cohort)
session.add(obj)
session.commit()
return obj
def persist(cls, cohort, **kwargs):
# persist cohort on disk
identifier = str(uuid.uuid4())
base_path = f"{BASE_PATH}/cohorts"
os.makedirs(base_path, exist_ok=True)
path = f"{base_path}/{identifier}.cohort"
cohort_dict = cohort.to_json(path, **kwargs)
obj = cls(comment=cohort_dict["comment"],
version=cohort_dict["version"],
created_at=cohort_dict["createdAt"],
fiber_version=cohort_dict["fiberVersion"],
path=path)
session.add(obj)
session.commit()
return obj
def persist(cls, training_pipeline, threshold_occurring,
window_start_occurring, window_end_occurring,
feature_type_occurring, threshold_numeric,
window_start_numeric, window_end_numeric, feature_type_numeric,
target):
obj = cls(training_pipeline_id=training_pipeline.id,
threshold_occurring=threshold_occurring,
window_start_occurring=window_start_occurring,
window_end_occurring=window_end_occurring,
feature_type_occurring=feature_type_occurring,
threshold_numeric=threshold_numeric,
window_start_numeric=window_start_numeric,
window_end_numeric=window_end_numeric,
feature_type_numeric=feature_type_numeric,
target=target)
session.add(obj)
session.commit()
return obj
def create(cls, comment, version, cohort, onset_dataframe, data_loader,
feature_pipeline):
identifier = str(uuid.uuid4())
base_path = f"{BASE_PATH}/data-loader"
os.makedirs(base_path, exist_ok=True)
path = f"{base_path}/{identifier}.data-loader.gz"
data_loader.dump(path)
obj = cls(comment=comment,
version=version,
start_time=time.time(),
end_time=math.inf,
cohort=cohort,
onset_dataframe=onset_dataframe,
data_loader_path=path,
feature_pipeline=feature_pipeline)
session.add(obj)
session.commit()
return obj
def persist(cls, df, window, min_threshold, feature_type, condition,
feature_pipeline, **kwargs):
# persist df on disk
identifier = str(uuid.uuid4())
base_path = f"{BASE_PATH}/features"
os.makedirs(base_path, exist_ok=True)
path = f"{base_path}/{identifier}.df.gz"
df.to_csv(path, index=False)
obj = cls(
feature_pipeline=feature_pipeline,
feature_type=feature_type,
path=path,
condition=condition.__class__.__name__,
window_start=window[0],
window_end=window[1],
min_threshold=min_threshold,
)
session.add(obj)
session.commit()
return obj
def persist(cls, df, clf, training_configuration, comment, version, explanations=None):
# persist df on disk
identifier = str(uuid.uuid4())
base_path = f"{BASE_PATH}/prediction-data"
os.makedirs(base_path, exist_ok=True)
data_path = f"{base_path}/{identifier}.df.gz"
df.to_csv(data_path, index=False)
# persist clf on disk
base_path = f"{BASE_PATH}/predictors"
os.makedirs(base_path, exist_ok=True)
predictor_path = f"{base_path}/{identifier}.clf"
dill.dump(clf, open(predictor_path, 'wb'))
if explanations:
ax = plt.gca()
plot_explanation_heatmap(explanations, top_features=20, ax=ax)
base_path = f"{BASE_PATH}/interpretation"
os.makedirs(base_path, exist_ok=True)
interpretation_path = f"{base_path}/{identifier}.png"
ax.figure.savefig(interpretation_path, dpi=300, bbox_inches="tight")
else:
interpretation_path = ''
obj = cls(
comment=comment,
version=version,
predictor_path=predictor_path,
data_path=data_path,
created_at=time.time(),
training_configuration_id=training_configuration.id,
interpretation_path=interpretation_path
)
session.add(obj)
session.commit()
return obj