Python Quick Start¶
Project¶
Create workdir.
Add conf.py with configuration dictionary CNFG.
Pass CNFG to
pycnfg.run().
Code¶
conf.py:
"""Configuration example.
Create pipeline (sgd) and optimize hp_grid:
* target transformer on/off.
* polynomial degree 1/2.
"""
import lightgbm
import mlshell
import pycnfg
import sklearn
target_transformer = sklearn.preprocessing.PowerTransformer(
method='yeo-johnson', standardize=True, copy=True)
# Optimization hp ranges.
hp_grid = {
# 'process_parallel__pipeline_numeric__transform_normal__skip': [False],
# 'process_parallel__pipeline_numeric__scale_column_wise__quantile_range': [(1, 99)],
'process_parallel__pipeline_numeric__add_polynomial__degree': [1, 2],
'estimate__transformer': [None, target_transformer],
# sgd
# 'estimate__regressor__alpha': np.logspace(-2, -1, 10),
# 'estimate__regressor__l1_ratio': np.linspace(0.1, 1, 10),
}
CNFG = {
# Path section - specify project directory.
'path': {
'default': {
'priority': 1,
'init': pycnfg.find_path,
'producer': pycnfg.Producer,
}
},
# Logger section - create logger.
'logger': {
'default': {
'priority': 2,
'init': 'default',
'producer': mlshell.LoggerProducer,
'steps': [
('make',),
],
}
},
# Pipeline section - specify pipelines creation/loading.
'pipeline': {
'sgd': {
'priority': 3,
'init': mlshell.Pipeline,
'producer': mlshell.PipelineProducer,
'steps': [
('make', {
'estimator_type': 'regressor',
'estimator': sklearn.linear_model.SGDRegressor(
penalty='elasticnet', l1_ratio=1, shuffle=False,
max_iter=1000, alpha=0.02, random_state=42),
}),
],
},
'lgbm': {
'priority': 3,
'init': mlshell.Pipeline,
'producer': mlshell.PipelineProducer,
'steps': [
('make', {
'estimator_type': 'regressor',
'estimator': lightgbm.LGBMRegressor(
num_leaves=2, min_data_in_leaf=60,
n_estimators=200, max_depth=-1, random_state=42),
}),
],
}
},
# Metric section - specify metric creation/loading.
'metric': {
'r2': {
'priority': 4,
'init': mlshell.Metric,
'producer': mlshell.MetricProducer,
'steps': [
('make', {
'score_func': sklearn.metrics.r2_score,
'greater_is_better': True,
}),
],
},
'mse': {
'priority': 4,
'init': mlshell.Metric,
'producer': mlshell.MetricProducer,
'steps': [
('make', {
'score_func': sklearn.metrics.mean_squared_error,
'greater_is_better': False,
'squared': False
}),
],
},
},
# Dataset section - specify dataset loading/preprocessing/splitting.
'dataset': {
'train': {
'priority': 5,
'init': mlshell.Dataset,
'producer': mlshell.DatasetProducer,
'steps': [
('load', {'filepath': './data/train.csv'}),
('info',),
('preprocess', {'targets_names': ['wage'],
'categor_names': ['union', 'goodhlth',
'black', 'female',
'married', 'service']}),
('split', {'train_size': 0.75, 'shuffle': False, }),
],
},
'test': {
'priority': 5,
'init': mlshell.Dataset,
'producer': mlshell.DatasetProducer,
'steps': [
('load', {'filepath': 'data/test.csv'}),
('info',),
('preprocess', {'categor_names': ['union', 'goodhlth',
'black', 'female',
'married', 'service'],
'targets_names': ['wage']}),
],
},
},
# Workflow section - fit/predict pipelines on datasets, optimize/validate
# metrics.
'workflow': {
'conf': {
'priority': 6,
'init': {},
'producer': mlshell.Workflow,
'steps': [
# Train 'sgd' pipeline on 'train' subset of 'train' dataset
# with zero position hp from 'hp_grid'.
('fit', {
'pipeline_id': 'pipeline__sgd',
'dataset_id': 'dataset__train',
'subset_id': 'train',
'hp': hp_grid,
}),
# Validate 'sgd' pipeline on 'train' and 'test' subsets of
# 'train' dataset with 'r2' scorer (after fit).
('validate', {
'pipeline_id': 'pipeline__sgd',
'dataset_id': 'dataset__train',
'subset_id': ['train', 'test'],
'metric_id': ['metric__r2', 'metric__mse'],
}),
# Optimize 'sgd' pipeline on 'train' subset of 'train' dataset
# on hp combinations from 'hp_grid'. Score and refit on 'r2'
# scorer.
('optimize', {
'pipeline_id': 'pipeline__sgd',
'dataset_id': 'dataset__train',
'subset_id': 'train',
'metric_id': ['metric__r2', 'metric__mse'],
'hp_grid': hp_grid,
'gs_params': {
'n_iter': None,
'n_jobs': 1,
'refit': 'metric__r2',
'cv': sklearn.model_selection.KFold(n_splits=3,
shuffle=True,
random_state=42),
'verbose': 1,
'pre_dispatch': 'n_jobs',
'return_train_score': True,
},
}),
# Validate 'sgd' pipeline on 'train' and 'test' subsets of
# 'train' dataset with 'r2' scorer (after optimization).
('validate', {
'pipeline_id': 'pipeline__sgd',
'dataset_id': 'dataset__train',
'subset_id': ['train', 'test'],
'metric_id': ['metric__r2', 'metric__mse'],
}),
# Predict with 'sgd' pipeline on whole 'test' dataset.
('predict', {
'pipeline_id': 'pipeline__sgd',
'dataset_id': 'dataset__test',
'subset_id': '',
}),
# Dump 'sgd' pipeline on disk.
('dump', {'pipeline_id': 'pipeline__sgd',
'dirpath': None}),
],
},
},
}
if __name__ == '__main__':
objects = pycnfg.run(CNFG, dcnfg={})
Shortened version using pycnfg features :
CNFG = {
# In ``pycnfg.run`` set default configuration :data:`mlshell.CNFG`, that
# has pre-defined path, logger sections and main sub-keys (see below).
'pipeline': {
'sgd': {
'estimator': sklearn.linear_model.SGDRegressor(
penalty='elasticnet', l1_ratio=1, shuffle=False,
max_iter=1000, alpha=0.02, random_state=42),
},
'lgbm': {
'estimator': lightgbm.LGBMRegressor(
num_leaves=2, min_data_in_leaf=60, n_estimators=200,
max_depth=-1, random_state=42),
}
},
'metric': {
'r2': {
'score_func': sklearn.metrics.r2_score,
'kwargs': {'greater_is_better': True},
},
'mse': {
'score_func': sklearn.metrics.mean_squared_error,
'kwargs': {
'greater_is_better': False,
'squared': False
},
},
},
'dataset': {
# Section level 'global' to specify common kwargs for test and train.
'global': {'targets_names': ['wage'],
'categor_names': ['union', 'goodhlth', 'black', 'female',
'married', 'service'],
},
'train': {
'filepath': './data/train.csv',
'split__kwargs': {'train_size': 0.75, 'shuffle': False},
},
'test': {
'filepath': 'data/test.csv',
'split__kwargs': {'train_size': 1},
},
},
'workflow': {
'conf': {
# Global values will replace kwargs in corresponding default steps
# => easy switch between pipeline for example (pycnfg move unknown
# keys to 'global' by default).
'pipeline_id': 'pipeline__sgd',
'dataset_id': 'dataset__train',
'predict__dataset_id': 'dataset__test',
'hp': hp_grid,
'hp_grid': hp_grid,
'gs_params': 'gs_params__conf',
'metric_id': ['metric__r2', 'metric__mse'],
'steps': [
('fit',),
('validate',),
('optimize',),
('validate',),
('predict',),
('dump',),
],
},
},
# Separate section for 'gs_params' kwarg.
'gs_params': {
'conf': {
'priority': 3,
'init': {
'n_iter': None,
'n_jobs': 1,
'refit': 'metric__r2',
'cv': sklearn.model_selection.KFold(n_splits=3,
shuffle=True,
random_state=42),
'verbose': 1,
'pre_dispatch': 'n_jobs',
'return_train_score': True,
},
},
},
}
if __name__ == '__main__':
objects = pycnfg.run(CNFG, dcnfg=mlshell.CNFG)