def get_dendrogram(self, linkage='ward', dist='spearmanr'):
data = VBHelper.saveFullFloatXy(X_df=self.X_df,
y_df=self.y_df,
X_df_s=self.vbhelper.X_df_start_order,
y_df_s=self.vbhelper.y_df_start_order)
vbs = VBSummary()
vbs.setData(data)
return vbs.hierarchicalDendrogram(linkage=linkage, dist=dist)
def get_kerneldensity(self):
data = VBHelper.saveFullFloatXy(X_df=self.X_df,
y_df=self.y_df,
X_df_s=self.vbhelper.X_df_start_order,
y_df_s=self.vbhelper.y_df_start_order)
vbs = VBSummary()
vbs.setData(data)
return vbs.kernelDensityPie()
def get_missing_vals(self):
data = VBHelper.saveFullFloatXy(X_df=self.X_df,
y_df=self.y_df,
X_df_s=self.vbhelper.X_df_start_order,
y_df_s=self.vbhelper.y_df_start_order)
vbs = VBSummary()
vbs.setData(data)
return vbs.missingVals()
def __init__(self, dataset_id):
# TODO: replace the need for the project_id with providing the target variable
self.dataset_id = dataset_id
self.dataset = Dataset.objects.get(pk=dataset_id)
self.df = load_dataset(dataset_id, self.dataset)
self.dataset_metadata = Metadata(
parent=self.dataset).get_metadata("DatasetMetadata")
self.target_label = "target" if "target" not in self.dataset_metadata.keys(
) else self.dataset_metadata["target"]
self.features_label = None if "features" not in self.dataset_metadata.keys(
) else self.dataset_metadata["features"]
if self.features_label is None or self.features_label == "*":
self.features_label = list(self.df.columns)
self.features_label.remove(self.target_label)
else:
self.features_label = json.loads(self.features_label)
self.y_df = self.df[self.target_label].to_frame()
self.X_df = self.df[self.features_label]
self.vbhelper = VBHelper(pipeline_id=-1)
self.vbhelper.setData(X_df=self.X_df, y_df=self.y_df)
def get_components(self, num_cols, keep_cats=False):
try:
if "," in num_cols:
_num_cols = num_cols.split(",")
num_cols = []
for n in _num_cols:
num_cols.append(int(n))
else:
num_cols = [int(num_cols)]
except Exception:
num_cols = [1]
data = VBHelper.saveFullFloatXy(X_df=self.X_df,
y_df=self.y_df,
X_df_s=self.vbhelper.X_df_start_order,
y_df_s=self.vbhelper.y_df_start_order)
vbs = VBSummary()
vbs.setData(data)
return vbs.viewComponents(num_cols=num_cols, keep_cats=keep_cats)
def execute_task(project_id, dataset_id, pipeline_id):
# STAGE 1 - Data and parameter load from db
update_status(pipeline_id,
"Data and Model Setup: Retrieving dataset and pipeline",
"1/{}".format(pre_processing_steps),
log="Pipeline: {}, Type: {}, Setup: 1/{}".format(
pipeline_id, None, pre_processing_steps),
message="Cross validation")
project = Project.objects.get(id=int(project_id))
dataset = Dataset.objects.get(id=int(dataset_id))
pipeline = Pipeline.objects.get(id=int(pipeline_id))
project.dataset = int(dataset_id)
project.save()
df = load_dataset(dataset_id, dataset)
dataset_metadata = Metadata(
parent=dataset).get_metadata("DatasetMetadata")
pipeline_metadata = Metadata(
parent=pipeline).get_metadata("PipelineMetadata")
project_metadata = Metadata(
parent=project).get_metadata("ProjectMetadata")
target_label = None if "target" not in project_metadata.keys(
) else project_metadata["target"]
features_label = None if "features" not in project_metadata.keys(
) else project_metadata["features"]
target_label = "target" if (
"target" not in dataset_metadata.keys()
and target_label is None) else dataset_metadata["target"]
if "features" not in dataset_metadata.keys(
) and features_label is None:
features_label = None
else:
features_label = dataset_metadata["features"]
if features_label is None or features_label == "*":
features_label = list(df.columns)
features_label.remove(target_label)
else:
features_label = json.loads(features_label)
drop_vars = [] if "drop_features" not in project_metadata.keys(
) else json.loads(project_metadata["drop_features"].replace(
"\'", "\""))
for d in drop_vars:
features_label.remove(d)
# STAGE 2 - Data prep
update_status(pipeline_id,
"Data and Model Setup: Loading data",
"2/{}".format(pre_processing_steps),
log="Pipeline: {}, Type: {}, Setup: 2/{}".format(
pipeline_id, pipeline.name, pre_processing_steps),
message="Cross validation")
target = df[target_label].to_frame()
if features_label:
features = df[features_label]
else:
features = df.drop(target_label, axis=1)
# STAGE 3 - VBHelper execution
update_status(
pipeline_id,
"Data and Model Setup: Loading all parameters and settings",
"3/{}".format(pre_processing_steps),
log="Pipeline: {}, Type: {}, Setup: 3/{}".format(
pipeline_id, pipeline.name, pre_processing_steps),
message="Cross validation")
if pipeline_metadata:
vbhelper_parameters = None if "parameters" not in pipeline_metadata.keys(
) else json.loads(pipeline_metadata["parameters"].replace(
"'", "\""))
else:
vbhelper_parameters = {}
vbhelper_parameters["pipeline_id"] = pipeline_id
outer_cv = pipeline_metadata[
"outer_cv"] if "outer_cv" in pipeline_metadata.keys() else "True"
try:
vbhelper = VBHelper(**vbhelper_parameters)
if "estimators" in pipeline_metadata.keys():
est_str = pipeline_metadata["estimators"].replace("\'", "\"")
estimators = json.loads(est_str)
else:
update_status(
pipeline_id,
"Error: VB Helper requires an estimator.",
"-1/{}".format(pre_processing_steps),
log="Pipeline: {}, Type: {}, Setup: -1/{}".format(
pipeline_id, pipeline.name, pre_processing_steps),
message="Cross validation")
return
vbhelper.setData(X_df=features, y_df=target)
inner_cv_dict = {
'cv_reps': 1,
'cv_folds': 5,
'cv_strategy': ('quantile', 5)
}
inner_cv = vbhelper.getCV(cv_dict=inner_cv_dict)
# prep_dict = {'cat_approach': 'together', 'impute_strategy': 'IterativeImputer', 'cat_idx': vbhelper.cat_idx}
prep_dict = {
'cat_approach': 'together',
'impute_strategy': 'impute_middle',
'cat_idx': vbhelper.cat_idx
}
pipe_kwargs = dict(do_prep=not vbhelper.run_stacked,
prep_dict=prep_dict,
inner_cv=inner_cv,
cat_idx=vbhelper.cat_idx,
float_idx=vbhelper.float_idx,
bestT=False)
estimators_dict = {}
e_i = 0
for e in estimators:
name = e["name"] if "name" in e.keys(
) else e["type"] + "-{}".format(e_i)
n_i = 1
n_name = name
while n_name in estimators_dict.keys():
n_name = name + "-{}".format(n_i)
n_i += 1
name = n_name
estimator = DaskTasks.get_estimator(e["type"])
e_kwargs = copy.copy(pipe_kwargs)
for k, p in e["parameters"].items():
e_kwargs[k] = p
estimators_dict[name] = {
"pipe": estimator,
"pipe_kwargs": e_kwargs
}
e_i += 1
vbhelper.setPipeDict(estimators_dict)
vbhelper.setModelDict()
if outer_cv == "True":
vbhelper.runCrossValidate(verbose=True)
vbhelper.buildCVScoreDict()
else:
#TODO: check processing for non-outer-cv instance for data cleanup
vbhelper.fitEstimators()
try:
model = Model.objects.get(pipeline=pipeline)
model_id = model.id
except Model.DoesNotExist:
model_id = None
vbhelper.save(message="Completed.")
del model
except Exception as e:
update_status(pipeline_id,
"Error: Unknown error executing pipeline",
"-0/16",
log="Pipeline: {}, Type: {}, Error: {}".format(
pipeline_id, pipeline.name, e),
message="Cross validation")
del vbhelper
class DataExploration:
def __init__(self, dataset_id):
# TODO: replace the need for the project_id with providing the target variable
self.dataset_id = dataset_id
self.dataset = Dataset.objects.get(pk=dataset_id)
self.df = load_dataset(dataset_id, self.dataset)
self.dataset_metadata = Metadata(
parent=self.dataset).get_metadata("DatasetMetadata")
self.target_label = "target" if "target" not in self.dataset_metadata.keys(
) else self.dataset_metadata["target"]
self.features_label = None if "features" not in self.dataset_metadata.keys(
) else self.dataset_metadata["features"]
if self.features_label is None or self.features_label == "*":
self.features_label = list(self.df.columns)
self.features_label.remove(self.target_label)
else:
self.features_label = json.loads(self.features_label)
self.y_df = self.df[self.target_label].to_frame()
self.X_df = self.df[self.features_label]
self.vbhelper = VBHelper(pipeline_id=-1)
self.vbhelper.setData(X_df=self.X_df, y_df=self.y_df)
def get_missing_vals(self):
data = VBHelper.saveFullFloatXy(X_df=self.X_df,
y_df=self.y_df,
X_df_s=self.vbhelper.X_df_start_order,
y_df_s=self.vbhelper.y_df_start_order)
vbs = VBSummary()
vbs.setData(data)
return vbs.missingVals()
def get_components(self, num_cols, keep_cats=False):
try:
if "," in num_cols:
_num_cols = num_cols.split(",")
num_cols = []
for n in _num_cols:
num_cols.append(int(n))
else:
num_cols = [int(num_cols)]
except Exception:
num_cols = [1]
data = VBHelper.saveFullFloatXy(X_df=self.X_df,
y_df=self.y_df,
X_df_s=self.vbhelper.X_df_start_order,
y_df_s=self.vbhelper.y_df_start_order)
vbs = VBSummary()
vbs.setData(data)
return vbs.viewComponents(num_cols=num_cols, keep_cats=keep_cats)
def get_kerneldensity(self):
data = VBHelper.saveFullFloatXy(X_df=self.X_df,
y_df=self.y_df,
X_df_s=self.vbhelper.X_df_start_order,
y_df_s=self.vbhelper.y_df_start_order)
vbs = VBSummary()
vbs.setData(data)
return vbs.kernelDensityPie()
def get_dendrogram(self, linkage='ward', dist='spearmanr'):
data = VBHelper.saveFullFloatXy(X_df=self.X_df,
y_df=self.y_df,
X_df_s=self.vbhelper.X_df_start_order,
y_df_s=self.vbhelper.y_df_start_order)
vbs = VBSummary()
vbs.setData(data)
return vbs.hierarchicalDendrogram(linkage=linkage, dist=dist)
def execute_task(project_id, dataset_id, pipeline_id):
# STAGE 1 - Data and parameter load from db
update_status(
pipeline_id,
"Data and Model Setup: Retrieving dataset and pipeline", "1/{}".format(pre_processing_steps),
log="Pipeline: {}, Type: {}, Setup: 1/{}".format(pipeline_id, None, pre_processing_steps)
)
project = Project.objects.get(id=int(project_id))
dataset = Dataset.objects.get(id=int(dataset_id))
pipeline = Pipeline.objects.get(id=int(pipeline_id))
project.dataset = int(dataset_id)
project.save()
df = load_dataset(dataset_id, dataset)
pipeline_metadata = Metadata(parent=Pipeline.objects.get(id=pipeline_id)).get_metadata("PipelineMetadata")
project_metadata = Metadata(parent=Project.objects.get(id=project_id)).get_metadata("ProjectMetadata")
target_label = "response" if "target" not in project_metadata.keys() else project_metadata["target"]
features_label = None if "features" not in project_metadata.keys() else project_metadata["features"]
# STAGE 2 - Data prep
update_status(
pipeline_id,
"Data and Model Setup: Loading data", "2/{}".format(pre_processing_steps),
log="Pipeline: {}, Type: {}, Setup: 2/{}".format(pipeline_id, pipeline.name, pre_processing_steps)
)
target = df[target_label]
if features_label:
features_list = json.loads(features_label.replace("\'", "\""))
features = df[features_list]
else:
features = df.drop(target_label, axis=1)
# STAGE 3 - VBHelper execution
update_status(
pipeline_id,
"Data and Model Setup: Loading all parameters and settings", "3/{}".format(pre_processing_steps),
log="Pipeline: {}, Type: {}, Setup: 3/{}".format(pipeline_id, pipeline.name, pre_processing_steps)
)
if pipeline_metadata:
vbhelper_parameters = None if "parameters" not in pipeline_metadata.keys() else json.loads(pipeline_metadata["parameters"].replace("'", "\""))
else:
vbhelper_parameters = {}
vbhelper_parameters["pipeline_id"] = pipeline_id
outer_cv = pipeline_metadata["outer_cv"] if "outer_cv" in pipeline_metadata.keys() else "True"
try:
vbhelper = VBHelper(**vbhelper_parameters)
if "estimators" in pipeline_metadata.keys():
estimators = json.loads(pipeline_metadata["estimators"].replace("\'", "\""))
else:
update_status(pipeline_id, "Error: VB Helper requires an estimator.",
"-1/{}".format(pre_processing_steps),
log="Pipeline: {}, Type: {}, Setup: -1/{}".format(pipeline_id, pipeline.name,
pre_processing_steps)
)
return
vbhelper.setData(X_df=features, y_df=target)
inner_cv_dict = {'cv_reps': 1, 'cv_folds': 5, 'cv_strategy': ('quantile', 5)}
inner_cv = vbhelper.getCV(cv_dict=inner_cv_dict)
prep_dict = {'impute_strategy': 'impute_knn5', 'cat_idx': vbhelper.cat_idx}
pipe_kwargs = dict(do_prep=not vbhelper.run_stacked, prep_dict=prep_dict, inner_cv=inner_cv,
gridpoints=4, cat_idx=vbhelper.cat_idx, float_idx=vbhelper.float_idx,
bestT=False)
estimators_dict = {}
e_i = 0
for e in estimators:
name = e["name"] if "name" in e.keys() else e["type"] + "-{}".format(e_i)
n_i = 1
n_name = name
while n_name in estimators_dict.keys():
n_name = name + "-{}".format(n_i)
n_i += 1
name = n_name
estimator = DaskTasks.get_estimator(e["type"])
e_kwargs = copy.copy(pipe_kwargs)
for k, p in e["parameters"].items():
e_kwargs[k] = p
estimators_dict[name] = {"pipe": estimator, "pipe_kwargs": e_kwargs}
e_i += 1
vbhelper.setPipeDict(estimators_dict)
vbhelper.setModelDict()
if outer_cv == "True":
vbhelper.runCrossValidate()
vbhelper.buildCVScoreDict()
else:
vbhelper.fitEstimators()
vbhelper.save()
except Exception as e:
update_status(pipeline_id, "Error: Unknown error executing pipeline",
"-0/16",
log="Pipeline: {}, Type: {}, Error: {}".format(pipeline_id, pipeline.name, e)
)