atm package

Subpackages

• atm.api package
  • Submodules
    • atm.api.utils module
  • Module contents

Submodules

• atm.classifier module
• atm.cli module
• atm.compat module
• atm.config module
• atm.constants module
• atm.core module
• atm.data module
• atm.database module
• atm.encoder module
• atm.method module
• atm.metrics module
• atm.utilities module
• atm.worker module

Module contents

Auto Tune Models A multi-user, multi-data AutoML framework.

Classes

ATM([dialect, database, username, password, …])
Model(method, params, judgment_metric, …)	This class contains everything needed to run an end-to-end ATM classifier pipeline.

class atm.ATM(dialect='sqlite', database='atm.db', username=None, password=None, host=None, port=None, query=None, access_key=None, secret_key=None, s3_bucket=None, s3_folder=None, models_dir='models', metrics_dir='metrics', verbose_metrics=False)

Bases: object

Methods

add_datarun(dataset_id[, budget, …])	Register one or more Dataruns to the Database.
add_dataset(train_path[, test_path, name, …])	Add a new dataset to the Database.
load_model(classifier_id)	Load a Model from the Database.
run(train_path[, test_path, name, …])	Create a Dataset and a Datarun and then work on it.
work([datarun_ids, save_files, …])	Get unfinished Dataruns from the database and work on them.

add_datarun(dataset_id, budget=100, budget_type='classifier', gridding=0, k_window=3, metric='f1', methods=['logreg', 'dt', 'knn'], r_minimum=2, run_per_partition=False, score_target='cv', priority=1, selector='uniform', tuner='uniform', deadline=None)

Register one or more Dataruns to the Database.

The methods hyperparameters will be analyzed and Hyperpartitions generated from them. If run_per_partition is True, one Datarun will be created for each Hyperpartition. Otherwise, a single one will be created for all of them.

Parameters

• dataset_id (int) – Id of the Dataset which this Datarun will belong to.

• budget (int) – Budget amount. Optional. Defaults to 100.

• budget_type (str) – Budget Type. Can be ‘classifier’ or ‘walltime’. Optional. Defaults to 'classifier'.

• gridding (int) – gridding setting for the Tuner. Optional. Defaults to 0.

• k_window (int) – k setting for the Selector. Optional. Defaults to 3.

• metric (str) – Metric to use for the tuning and selection. Optional. Defaults to 'f1'.

• methods (list) – List of methods to try. Optional. Defaults to ['logreg', 'dt', 'knn'].

• r_minimum (int) – r_minimum setting for the Tuner. Optional. Defaults to 2.

• run_per_partition (bool) – whether to create a separated Datarun for each Hyperpartition or not. Optional. Defaults to False.

• score_target (str) – Which score to use for the tuning and selection process. It can be 'cv' or 'test'. Optional. Defaults to 'cv'.

• priority (int) – Priority of this Datarun. The higher the better. Optional. Defaults to 1.

• selector (str) – Type of selector to use. Optional. Defaults to 'uniform'.

• tuner (str) – Type of tuner to use. Optional. Defaults to 'uniform'.

• deadline (str) – Time deadline. It must be a string representing a datetime in the format '%Y-%m-%d %H:%M'. If given, budget_type will be set to 'walltime'.

Returns

The created Datarun or list of Dataruns.

Return type

Datarun

add_dataset(train_path, test_path=None, name=None, description=None, class_column=None)

Add a new dataset to the Database.

Parameters

• train_path (str) – Path to the training CSV file. It can be a local filesystem path, absolute or relative, or an HTTP or HTTPS URL, or an S3 path in the format s3://{bucket_name}/{key}. Required.

• test_path (str) – Path to the testing CSV file. It can be a local filesystem path, absolute or relative, or an HTTP or HTTPS URL, or an S3 path in the format s3://{bucket_name}/{key}. Optional. If not given, the training CSV will be split in two parts, train and test.

• name (str) – Name given to this dataset. Optional. If not given, a hash will be generated from the training_path and used as the Dataset name.

• description (str) – Human friendly description of the Dataset. Optional.

• class_column (str) – Name of the column that will be used as the target variable. Optional. Defaults to 'class'.

Returns

The created dataset.

Return type

Dataset

load_model(classifier_id)

Load a Model from the Database.

Parameters

classifier_id (int) – Id of the Model to load.

Returns

The loaded model instance.

Return type

Model

run(train_path, test_path=None, name=None, description=None, class_column='class', budget=100, budget_type='classifier', gridding=0, k_window=3, metric='f1', methods=['logreg', 'dt', 'knn'], r_minimum=2, run_per_partition=False, score_target='cv', selector='uniform', tuner='uniform', deadline=None, priority=1, save_files=True, choose_randomly=True, cloud_mode=False, total_time=None, verbose=True)

Create a Dataset and a Datarun and then work on it.

Parameters

• train_path (str) – Path to the training CSV file. It can be a local filesystem path, absolute or relative, or an HTTP or HTTPS URL, or an S3 path in the format s3://{bucket_name}/{key}. Required.

• test_path (str) – Path to the testing CSV file. It can be a local filesystem path, absolute or relative, or an HTTP or HTTPS URL, or an S3 path in the format s3://{bucket_name}/{key}. Optional. If not given, the training CSV will be split in two parts, train and test.

• name (str) – Name given to this dataset. Optional. If not given, a hash will be generated from the training_path and used as the Dataset name.

• description (str) – Human friendly description of the Dataset. Optional.

• class_column (str) – Name of the column that will be used as the target variable. Optional. Defaults to 'class'.

• budget (int) – Budget amount. Optional. Defaults to 100.

• budget_type (str) – Budget Type. Can be ‘classifier’ or ‘walltime’. Optional. Defaults to 'classifier'.

• gridding (int) – gridding setting for the Tuner. Optional. Defaults to 0.

• k_window (int) – k setting for the Selector. Optional. Defaults to 3.

• metric (str) – Metric to use for the tuning and selection. Optional. Defaults to 'f1'.

• methods (list) – List of methods to try. Optional. Defaults to ['logreg', 'dt', 'knn'].

• r_minimum (int) – r_minimum setting for the Tuner. Optional. Defaults to 2.

• run_per_partition (bool) – whether to create a separated Datarun for each Hyperpartition or not. Optional. Defaults to False.

• score_target (str) – Which score to use for the tuning and selection process. It can be 'cv' or 'test'. Optional. Defaults to 'cv'.

• priority (int) – Priority of this Datarun. The higher the better. Optional. Defaults to 1.

• selector (str) – Type of selector to use. Optional. Defaults to 'uniform'.

• tuner (str) – Type of tuner to use. Optional. Defaults to 'uniform'.

• deadline (str) – Time deadline. It must be a string representing a datetime in the format '%Y-%m-%d %H:%M'. If given, budget_type will be set to 'walltime'.

• verbose (bool) – Whether to be verbose about the process. Optional. Defaults to True.

Returns

The created Datarun or list of Dataruns.

Return type

Datarun

work(datarun_ids=None, save_files=True, choose_randomly=True, cloud_mode=False, total_time=None, wait=True, verbose=False)

Get unfinished Dataruns from the database and work on them.

Check the ModelHub Database for unfinished Dataruns, and work on them as they are added. This process will continue to run until it exceeds total_time or there are no more Dataruns to process or it is killed.

Parameters

• datarun_ids (list) – list of IDs of Dataruns to work on. If None, this will work on any unfinished Dataruns found in the database. Optional. Defaults to None.

• save_files (bool) – Whether to save the fitted classifiers and their metrics or not. Optional. Defaults to True.

• choose_randomly (bool) – If True, work on all the highest-priority dataruns in random order. Otherwise, work on them in sequential order (by ID). Optional. Defaults to True.

• cloud_mode (bool) – Save the models and metrics in AWS S3 instead of locally. This option works only if S3 configuration has been provided on initialization. Optional. Defaults to False.

• total_time (int) – Total time to run the work process, in seconds. If None, continue to run until interrupted or there are no more Dataruns to process. Optional. Defaults to None.

• wait (bool) – If True, wait for more Dataruns to be inserted into the Database once all have been processed. Otherwise, exit the worker loop when they run out. Optional. Defaults to False.

• verbose (bool) – Whether to be verbose about the process. Optional. Defaults to True.

class atm.Model(method, params, judgment_metric, class_column, testing_ratio=0.3, verbose_metrics=False)

Bases: object

This class contains everything needed to run an end-to-end ATM classifier pipeline. It is initialized with a set of parameters and trained like a normal sklearn model. This class can be pickled and saved to disk, then unpickled outside of ATM and used to classify new datasets.

Attributes

ATM_KEYS	list() -> new empty list
MINMAX	str(object=’‘) -> str
N_FOLDS	int(x=0) -> integer
PCA	str(object=’‘) -> str
PCA_DIMS	str(object=’‘) -> str
SCALE	str(object=’‘) -> str
WHITEN	str(object=’‘) -> str

Methods

load(path)	Loads a saved Model instance from a path.
predict(data)	Generate predictions from new data.
save(path[, force])	Save this Model using pickle.
train_test(dataset)	Train and test this model using Cross Validation and Holdout.

ATM_KEYS = ['_scale', '_whiten', '_scale_minmax', '_pca', '_pca_dimensions']

MINMAX = '_scale_minmax'

N_FOLDS = 5

PCA = '_pca'

PCA_DIMS = '_pca_dimensions'

SCALE = '_scale'

WHITEN = '_whiten'

classmethod load(path)

Loads a saved Model instance from a path.

Parameters

path (str) – path where the model is saved.

Returns

New model instance.

Return type

Model

predict(data)

Generate predictions from new data.

Parameters

data (pandas.DataFrame) – Data for which to predict classes

Returns

Vector of predictions

Return type

pandas.Series

save(path, force=False)

Save this Model using pickle.

Parameters

• path (str) – Path where the model should be saved.

• force (bool) – If True, overwrite the model if it already exists.

train_test(dataset)

Train and test this model using Cross Validation and Holdout.

Parameters

dataset (Dataset) – Dataset object from database.

Returns

Dictionary containing:

• cv (list): The cross validation scores array

• test (dict): The test scores dictionary

Return type

dict