def set_model_name(job: Job) -> None:
if job.create_models:
if job.predictive_model.model_path != '':
# job.predictive_model = duplicate_orm_row(PredictiveModel.objects.filter(pk=job.predictive_model.pk)[0]) #todo: replace with simple CREATE
job.predictive_model = PredictiveModel.init(
job.predictive_model.get_full_dict(
) #todo: doublecheck me, are you sure get_full_dict is returning everything needed?
) #todo: futurebug if object changes
job.predictive_model.save()
job.save()
if job.clustering.clustering_method != ClusteringMethods.NO_CLUSTER.value:
job.clustering.model_path = 'cache/model_cache/job_{}-split_{}-clusterer-{}-v0.sav'.format(
job.id, job.split.id, job.type)
job.clustering.save()
if job.type == JobTypes.UPDATE.value:
job.type = JobTypes.PREDICTION.value #TODO: Y am I doing this?
predictive_model_filename = 'cache/model_cache/job_{}-split_{}-predictive_model-{}-v{}.sav'.format(
job.id, job.split.id, job.type, str(time.time()))
else:
predictive_model_filename = 'cache/model_cache/job_{}-split_{}-predictive_model-{}-v0.sav'.format(
job.id, job.split.id, job.type)
job.predictive_model.model_path = predictive_model_filename
job.predictive_model.save()
job.save()
def create_test_predictive_model(
predictive_model: str = PredictiveModels.CLASSIFICATION.value,
prediction_method: str = ClassificationMethods.RANDOM_FOREST.value,
configuration: dict = {}) -> PredictiveModel:
pred_model = PredictiveModel.init(
get_prediction_method_config(predictive_model, prediction_method,
configuration))
return pred_model
def _calculate_and_evaluate(args) -> dict:
global trial_number
if trial_number % 20 == 0:
logger.info("Trial {}".format(trial_number))
trial_number += 1
local_job = global_job
predictive_model = local_job.predictive_model.predictive_model
prediction_method = local_job.predictive_model.prediction_method
model_config = {
'predictive_model': predictive_model,
'prediction_method': prediction_method,
**args
}
new_predictive_model = PredictiveModel.init(model_config)
local_job.predictive_model = duplicate_orm_row(new_predictive_model)
local_job.predictive_model.save()
local_job.save()
# local_job = duplicate_orm_row(local_job) #TODO not sure it is ok to have this here.
performance_metric = local_job.hyperparameter_optimizer.__getattribute__(
local_job.hyperparameter_optimizer.optimization_method.lower(
)).performance_metric
multiplier = _get_metric_multiplier(performance_metric)
try:
results, model_split = run_by_type(training_df.copy(), test_df.copy(),
local_job)
return {
'loss': -results[performance_metric] * multiplier,
'status': STATUS_OK,
'results': results,
'predictive_model_id': local_job.predictive_model.pk,
'model_split': model_split,
'config': model_config
}
except:
return {
'loss': 100,
'status': STATUS_FAIL,
'results': {},
'predictive_model_id': {},
'model_split': {},
'config': {}
}
def update(split, payload, generation_type=PredictiveModels.CLASSIFICATION.value): # TODO adapt to allow selecting the predictive_model to update
jobs = []
config = payload['config']
labelling_config = config['labelling'] if 'labelling' in config else {}
for method in payload['config']['methods']:
for clustering in payload['config']['clusterings']:
for incremental_base_model in payload['config']['incremental_train']:
for encMethod in payload['config']['encodings']:
encoding = payload['config']['encoding']
if encoding['generation_type'] == UP_TO:
for i in range(1, encoding['prefix_length'] + 1):
job, _ = Job.objects.get_or_create(
status=JobStatuses.CREATED.value,
type=JobTypes.UPDATE.value,
split=split,
encoding=Encoding.objects.get_or_create( # TODO fixme
data_encoding=DataEncodings.LABEL_ENCODER.value,
value_encoding=encMethod,
add_elapsed_time=labelling_config.get('add_elapsed_time', False),
add_remaining_time=labelling_config.get('add_remaining_time', False),
add_executed_events=labelling_config.get('add_executed_events', False),
add_resources_used=labelling_config.get('add_resources_used', False),
add_new_traces=labelling_config.get('add_new_traces', False),
prefix_length=i,
# TODO static check?
padding=True if config['encoding']['padding'] == 'zero_padding' else False,
task_generation_type=config['encoding'].get('generation_type', 'only_this'),
features=config['encoding'].get('features', [])
)[0],
labelling=Labelling.objects.get_or_create(
type=labelling_config.get('type', None),
# TODO static check?
attribute_name=labelling_config.get('attribute_name', None),
threshold_type=labelling_config.get('threshold_type', None),
threshold=labelling_config.get('threshold', None)
)[0] if labelling_config != {} else None,
clustering=Clustering.init(clustering, configuration=config.get(clustering, {})),
predictive_model=PredictiveModel.init(
get_prediction_method_config(generation_type, method, payload)
),
hyperparameter_optimizer=HyperparameterOptimization.init(
config.get('hyperparameter_optimizer', None)),
create_models=config.get('create_models', False),
incremental_train=Job.objects.filter(
pk=incremental_base_model
)[0]
)
check_predictive_model_not_overwrite(job)
set_model_name(job)
jobs.append(job)
else:
job, _ = Job.objects.get_or_create(
status=JobStatuses.CREATED.value,
type=JobTypes.UPDATE.value,
split=split,
encoding=Encoding.objects.get_or_create( # TODO fixme
data_encoding=DataEncodings.LABEL_ENCODER.value,
value_encoding=encMethod,
add_elapsed_time=labelling_config.get('add_elapsed_time', False),
add_remaining_time=labelling_config.get('add_remaining_time', False),
add_executed_events=labelling_config.get('add_executed_events', False),
add_resources_used=labelling_config.get('add_resources_used', False),
add_new_traces=labelling_config.get('add_new_traces', False),
prefix_length=config['encoding']['prefix_length'],
# TODO static check?
padding=True if config['encoding']['padding'] == 'zero_padding' else False,
task_generation_type=config['encoding'].get('generation_type', 'only_this'),
features=config['encoding'].get('features', [])
)[0],
labelling=Labelling.objects.get_or_create(
type=labelling_config.get('type', None),
# TODO static check?
attribute_name=labelling_config.get('attribute_name', None),
threshold_type=labelling_config.get('threshold_type', None),
threshold=labelling_config.get('threshold', None)
)[0] if labelling_config != {} else None,
clustering=Clustering.init(clustering, configuration=config.get(clustering, {})),
predictive_model=PredictiveModel.init(
get_prediction_method_config(generation_type, method, payload)
),
hyperparameter_optimizer=HyperparameterOptimization.init(
config.get('hyperparameter_optimizer', None)),
create_models=config.get('create_models', False),
incremental_train=Job.objects.filter(
pk=incremental_base_model
)[0]
)
check_predictive_model_not_overwrite(job)
set_model_name(job)
jobs.append(job)
return jobs
def generate(split, payload):
jobs = []
config = payload['config']
labelling_config = config['labelling'] if 'labelling' in config else {}
job_type = JobTypes.PREDICTION.value
prediction_type = payload['type']
for method in config['methods']:
for clustering in config['clusterings']:
for encMethod in config['encodings']:
encoding = config['encoding']
if encoding['generation_type'] == UP_TO:
for i in range(1, encoding['prefix_length'] + 1):
encoding = Encoding.objects.get_or_create(
data_encoding=DataEncodings.LABEL_ENCODER.value,
value_encoding=encMethod,
add_elapsed_time=labelling_config.get('add_elapsed_time', False),
add_remaining_time=labelling_config.get('add_remaining_time', False),
add_executed_events=labelling_config.get('add_executed_events', False),
add_resources_used=labelling_config.get('add_resources_used', False),
add_new_traces=labelling_config.get('add_new_traces', False),
prefix_length=i,
# TODO static check?
padding=True if config['encoding']['padding'] == 'zero_padding' else False,
task_generation_type=config['encoding'].get('generation_type', 'only_this'),
features=config['encoding'].get('features', [])
)[0]
predictive_model = PredictiveModel.init(
get_prediction_method_config(prediction_type, method, config))
job = Job.objects.get_or_create(
status=JobStatuses.CREATED.value,
type=job_type,
split=split,
encoding=encoding,
labelling=Labelling.objects.get_or_create(
type=labelling_config.get('type', None),
# TODO static check?
attribute_name=labelling_config.get('attribute_name', None),
threshold_type=labelling_config.get('threshold_type', None),
threshold=labelling_config.get('threshold', None)
)[0] if labelling_config != {} else None,
clustering=Clustering.init(clustering, configuration=config.get(clustering, {}))
if predictive_model.predictive_model != PredictiveModels.TIME_SERIES_PREDICTION.value
else Clustering.init(ClusteringMethods.NO_CLUSTER.value, configuration={}),
# TODO TEMPORARY workaround,
hyperparameter_optimizer=HyperparameterOptimization.init(
config.get('hyperparameter_optimizer', {
'type': None}) if predictive_model.predictive_model != PredictiveModels.TIME_SERIES_PREDICTION.value else {
'type': None}),
# TODO TEMPORARY workaround
predictive_model=predictive_model,
create_models=config.get('create_models', False)
)[0]
check_predictive_model_not_overwrite(job)
set_model_name(job)
jobs.append(job)
else:
predictive_model = PredictiveModel.init(
get_prediction_method_config(prediction_type, method, config))
job = Job.objects.get_or_create(
status=JobStatuses.CREATED.value,
type=job_type,
split=split,
encoding=Encoding.objects.get_or_create(
data_encoding=DataEncodings.LABEL_ENCODER.value,
value_encoding=encMethod,
add_elapsed_time=labelling_config.get('add_elapsed_time', False),
add_remaining_time=labelling_config.get('add_remaining_time', False),
add_executed_events=labelling_config.get('add_executed_events', False),
add_resources_used=labelling_config.get('add_resources_used', False),
add_new_traces=labelling_config.get('add_new_traces', False),
prefix_length=config['encoding']['prefix_length'],
# TODO static check?
padding=True if config['encoding']['padding'] == 'zero_padding' else False,
task_generation_type=config['encoding'].get('generation_type', 'only_this'),
features=config['encoding'].get('features', [])
)[0],
labelling=Labelling.objects.get_or_create(
type=labelling_config.get('type', None),
# TODO static check?
attribute_name=labelling_config.get('attribute_name', None),
threshold_type=labelling_config.get('threshold_type', None),
threshold=labelling_config.get('threshold', None)
)[0] if labelling_config != {} else None,
clustering=Clustering.init(clustering, configuration=config.get(clustering, {}))
if predictive_model.predictive_model != PredictiveModels.TIME_SERIES_PREDICTION.value
else Clustering.init(ClusteringMethods.NO_CLUSTER.value, configuration={}),
hyperparameter_optimizer=HyperparameterOptimization.init(
config.get('hyperparameter_optimizer', {
'type': 'none'}) if predictive_model.predictive_model != PredictiveModels.TIME_SERIES_PREDICTION.value else {
'type': 'none'}),
# TODO TEMPORARY workaround
predictive_model=predictive_model,
create_models=config.get('create_models', False)
)[0]
check_predictive_model_not_overwrite(job)
set_model_name(job)
jobs.append(job)
return jobs
def generate(split, payload):
jobs = []
config = payload['config']
labelling_config = config['labelling'] if 'labelling' in config else {}
job_type = JobTypes.PREDICTION.value
prediction_type = payload['type']
for method in config['methods']:
for clustering in config['clusterings']:
for encMethod in config['encodings']:
encoding = config['encoding']
if encoding['generation_type'] == UP_TO:
for i in range(1, encoding['prefix_length'] + 1):
encoding = Encoding.objects.get_or_create(
data_encoding='label_encoder',
value_encoding=encMethod,
add_elapsed_time=labelling_config.get(
'add_elapsed_time', False),
add_remaining_time=labelling_config.get(
'add_remaining_time', False),
add_executed_events=labelling_config.get(
'add_executed_events', False),
add_resources_used=labelling_config.get(
'add_resources_used', False),
add_new_traces=labelling_config.get(
'add_new_traces', False),
prefix_length=i,
# TODO static check?
padding=True if config['encoding']['padding']
== 'zero_padding' else False,
task_generation_type=config['encoding'].get(
'generation_type', 'only_this'))[0]
predictive_model = PredictiveModel.init(
get_prediction_method_config(
prediction_type, method, config))
job = Job.objects.get_or_create(
status=JobStatuses.CREATED.value,
type=job_type,
split=split,
encoding=encoding,
labelling=Labelling.objects.get_or_create(
type=labelling_config.get('type', None),
# TODO static check?
attribute_name=labelling_config.get(
'attribute_name', None),
threshold_type=labelling_config.get(
'threshold_type', None),
threshold=labelling_config.get(
'threshold', None))[0]
if labelling_config != {} else None,
clustering=Clustering.init(
clustering,
configuration=config.get(clustering, {})),
predictive_model=predictive_model)[0]
jobs.append(job)
else:
predictive_model = PredictiveModel.init(
get_prediction_method_config(prediction_type, method,
config))
job = Job.objects.get_or_create(
status=JobStatuses.CREATED.value,
type=job_type,
split=split,
encoding=Encoding.objects.get_or_create(
data_encoding='label_encoder',
value_encoding=encMethod,
add_elapsed_time=labelling_config.get(
'add_elapsed_time', False),
add_remaining_time=labelling_config.get(
'add_remaining_time', False),
add_executed_events=labelling_config.get(
'add_executed_events', False),
add_resources_used=labelling_config.get(
'add_resources_used', False),
add_new_traces=labelling_config.get(
'add_new_traces', False),
prefix_length=config['encoding']['prefix_length'],
# TODO static check?
padding=True if config['encoding']['padding']
== 'zero_padding' else False,
task_generation_type=config['encoding'].get(
'generation_type', 'only_this'))[0],
labelling=Labelling.objects.get_or_create(
type=labelling_config.get('type', None),
# TODO static check?
attribute_name=labelling_config.get(
'attribute_name', None),
threshold_type=labelling_config.get(
'threshold_type', None),
threshold=labelling_config.get('threshold',
None))[0]
if labelling_config != {} else None,
clustering=Clustering.init(clustering,
configuration=config.get(
clustering, {})),
predictive_model=predictive_model)[0]
jobs.append(job)
return jobs
def progetto_padova():
JOB = Job.objects.get_or_create(
status=JobStatuses.CREATED.value,
type=JobTypes.PREDICTION.value,
# split=Split.objects.get_or_create( # this creates the split of the log
# type=SplitTypes.SPLIT_SINGLE.value,
# original_log=create_log( # this imports the log
# import_log(BASE_DIR + RELATIVE_TRAIN_PATH),
# RELATIVE_TRAIN_PATH,
# BASE_DIR,
# import_in_cache=False
# ),
# splitting_method=SplitOrderingMethods.SPLIT_TEMPORAL.value,
# test_size=0.2
# )[0],
split=Split.objects.get_or_create( # this creates the split of the log
type=SplitTypes.SPLIT_DOUBLE.value,
train_log=create_log( # this imports the log
import_log(BASE_DIR + RELATIVE_TRAIN_PATH),
RELATIVE_TRAIN_PATH,
BASE_DIR,
import_in_cache=False
),
test_log=create_log( # this imports the log
import_log(BASE_DIR + RELATIVE_VALIDATION_PATH),
RELATIVE_VALIDATION_PATH,
BASE_DIR,
import_in_cache=False
)
)[0],
encoding=Encoding.objects.get_or_create( # this defines the encoding method
data_encoding=DataEncodings.LABEL_ENCODER.value,
value_encoding=ValueEncodings.SIMPLE_INDEX.value,
add_elapsed_time=False,
add_remaining_time=False,
add_executed_events=False,
add_resources_used=False,
add_new_traces=False,
prefix_length=5,
padding=True,
task_generation_type=TaskGenerationTypes.ALL_IN_ONE.value,
features=[]
)[0],
labelling=Labelling.objects.get_or_create( # this defines the label
type=LabelTypes.ATTRIBUTE_STRING.value,
attribute_name='label',
threshold_type=None,
threshold=None
)[0],
clustering=Clustering.init(ClusteringMethods.NO_CLUSTER.value, configuration={}),
predictive_model=PredictiveModel.init( # this defines the predictive model
get_prediction_method_config(
PredictiveModels.CLASSIFICATION.value,
ClassificationMethods.DECISION_TREE.value,
payload={
'max_depth': 2,
'min_samples_split': 2,
'min_samples_leaf': 2
}
)
),
hyperparameter_optimizer=HyperparameterOptimization.init({ # this defines the hyperparameter optimisation procedure
'type': HyperparameterOptimizationMethods.HYPEROPT.value,
'max_evaluations': 10,
'performance_metric': HyperOptAlgorithms.TPE.value,
'algorithm_type': HyperOptLosses.AUC.value
}),
create_models=True
)[0]
# load log
train_log, test_log, additional_columns = get_train_test_log(JOB.split)
# encode
train_df, test_df = encode_label_logs(train_log, test_log, JOB)
# train + evaluate
results, model_split = MODEL[JOB.predictive_model.predictive_model][ModelActions.BUILD_MODEL_AND_TEST.value](
train_df,
test_df,
_init_clusterer(JOB.clustering, train_df),
JOB
)
if JOB.create_models:
check_predictive_model_not_overwrite(JOB)
set_model_name(JOB)
save_models(model_split, JOB)
# predict
data_df = pd.concat([train_df, test_df])
results = MODEL[JOB.predictive_model.predictive_model][ModelActions.PREDICT.value](JOB, data_df)
results_with_probs = MODEL[JOB.predictive_model.predictive_model][ModelActions.PREDICT_PROBA.value](JOB, data_df)
# lime
exp = Explanation.objects.get_or_create(
type=ExplanationTypes.LIME.value,
split=JOB.split, # this defines the analysed log, you can use a different one from the training one
predictive_model=JOB.predictive_model,
job=JOB
)[0]
error, result = explanation(exp.id, int(EXPLANATION_TARGET))