def train_model(experiment_name, groups=1, time_window=pio.TimeWindow(-90, -30, 1, 15)):
path = os.path.join(DATA_PATH, 'ts.csv')
data, group_list = get_data(path, groups)
fname = '{}_{}.csv'.format(experiment_name, '-'.join(group_list))
data.to_csv(fname, index=False)
project = pio.Project.from_id(PROJECT_ID)
dataset = project.create_dataset(name=experiment_name,
dataframe=data)
experiment_version_config = pio.TrainingConfig(
advanced_models=[pio.AdvancedModel.LinReg],
normal_models=[pio.NormalModel.LinReg],
features=[pio.Feature.Counts],
profile=pio.Profile.Quick,
)
col_config = pio.ColumnConfig(target_column='target',
time_column='time',
# group_columns=group_list
)
experiment_version = project.fit_timeseries_regression(
experiment_name,
dataset,
time_window=time_window,
training_config=experiment_version_config,
column_config=col_config,
)
return experiment_version
def setup_module(module):
# Create project
global project
project = pio.Project.new(name=PROJECT_NAME,
description="description test sdk")
# Create dataset
global dataset
dataset = project.create_dataset('test_exporter',
file_name='data_exporter/titanic.csv')
# Train one model
training_config = pio.TrainingConfig(
advanced_models=[],
normal_models=[],
simple_models=[pio.SimpleModel.DecisionTree],
features=[],
profile=pio.Profile.Quick)
column_config = pio.ColumnConfig(target_column='Survived',
id_column='PassengerId')
experiment_version = project.fit_classification(
'test_exporter_classif',
dataset,
column_config,
metric=pio.metrics.Classification.AUC,
training_config=training_config,
)
# Create validation_prediction
experiment_version.wait_until(lambda experimentv: (len(
experimentv.models) > 0) or (experimentv._status['state'] == 'failed'))
if experiment_version._status['state'] == 'failed':
raise RuntimeError('Could not train experiment')
global validation_prediction
validation_prediction = experiment_version.predict_from_dataset(dataset)
# Create experiment deployment
experiment_version_best_model = experiment_version.best_model
experiment_deployment = project.create_experiment_deployment(
'test_sdk_' + TESTING_ID, experiment_version_best_model)
# Create deployment_prediction
experiment_deployment.wait_until(
lambda experimentd: experimentd.run_state == 'done')
global deployment_prediction
deployment_prediction = experiment_deployment.predict_from_dataset(dataset)
def train_model(uc_name, groups=1, time_window=pio.TimeWindow(-90, -30, 1, 15)):
path = os.path.join(DATA_PATH, 'ts.csv')
data, group_list = get_data(path, groups)
fname = '{}_{}.csv'.format(uc_name, '-'.join(group_list))
data.to_csv(fname, index=False)
dataset = pio.Dataset.new(name=uc_name,
dataframe=data)
uc_config = pio.TrainingConfig(normal_models=[pio.Model.LinReg],
lite_models=[pio.Model.LinReg],
features=[pio.Feature.Counts],
profile=pio.Profile.Quick)
col_config = pio.ColumnConfig(target_column='target',
time_column='time',
# group_columns=group_list
)
uc = pio.TimeSeries.fit(uc_name,
dataset,
time_window=time_window,
training_config=uc_config,
column_config=col_config)
return uc
import os
import pandas as pd
import previsionio as pio
from .utils import get_testing_id
TESTING_ID = get_testing_id()
pio.config.default_timeout = 120
col_config = pio.ColumnConfig(target_column='class', filename_column='filename')
uc_config = pio.TrainingConfig(normal_models=[pio.Model.LinReg],
lite_models=[],
simple_models=[],
features=[pio.Feature.Counts],
profile=pio.Profile.Quick)
test_datasets = {}
dataset_name = 'cats_and_dogs_train'
dataset_test_name = TESTING_ID + '-' + dataset_name
def upload_datasets():
datapath = os.path.join(os.path.dirname(os.path.realpath(__file__)), 'data_img/{}'.format(dataset_name))
# upload CSV reference file
dataset_csv = pio.Dataset.new(
name=dataset_test_name,
dataframe=pd.read_csv(os.path.join(datapath, '{}.csv'.format(dataset_name)))
)
# upload ZIP images folder
dataset_zip = pio.DatasetImages.new(
name=dataset_test_name,
import os
import pandas as pd
import pytest
import previsionio as pio
from .datasets import make_supervised_datasets, remove_datasets
from . import DATA_PATH
from .utils import train_model, get_testing_id, DROP_COLS
TESTING_ID = get_testing_id()
pio.config.zip_files = False
pio.config.default_timeout = 80
uc_config = pio.TrainingConfig(models=[pio.Model.LinReg],
simple_models=[pio.SimpleModel.DecisionTree],
features=[pio.Feature.Counts],
profile=pio.Profile.Quick)
test_datasets = {}
type_problem_2_pio_class = {
'regression': pio.Regression,
'classification': pio.Classification,
'multiclassification': pio.MultiClassification,
}
type_problems = type_problem_2_pio_class.keys()
def make_pio_datasets(paths):
for problem_type, p in paths.items():
dataset = pio.Dataset.new(p.split('/')[-1].replace(
'.csv',
# CLIENT INITIALIZATION -----------------------------------------
url = """https://<your instance>.prevision.io"""
token = """<your token>"""
pio.client.init_client(url, token)
# DATA LOADING --------------------------------------------------
# load data from a CSV
dataframe = pd.read_csv('helloworld_train.csv')
# upload it to the platform
dataset = pio.Dataset.new(name='helloworld_train', dataframe=dataframe)
# USECASE TRAINING ----------------------------------------------
# setup usecase
uc_config = pio.TrainingConfig(models=[pio.Model.XGBoost],
features=pio.Feature.Full,
profile=pio.Profile.Quick)
# run training
uc = pio.Classification.fit('helloworld_classif',
dataset,
metric=pio.metrics.Classification.AUC,
training_config=uc_config)
# (block until there is at least 1 model trained)
uc.wait_until(lambda usecase: len(usecase) > 0)
# check out the usecase status and other info
uc.print_info()
print('Current number of models:', len(uc))
print('Current (best model) score:', uc.score)
import pandas as pd
import previsionio as pio
from . import DATA_PATH
from .datasets import make_supervised_datasets, remove_datasets
from .utils import train_model, get_testing_id
TESTING_ID = get_testing_id()
PROJECT_NAME = "sdk_test_experiment_deployment" + str(TESTING_ID)
PROJECT_ID = ""
experiment_version_config = pio.TrainingConfig(
advanced_models=[pio.AdvancedModel.LinReg],
normal_models=[pio.NormalModel.LinReg],
simple_models=[pio.SimpleModel.DecisionTree],
features=[pio.Feature.Counts],
profile=pio.Profile.Quick,
)
training_type_2_pio_class = {
'regression': "fit_regression",
'classification': "fit_classification",
'multiclassification': "fit_multiclassification",
}
training_types = training_type_2_pio_class.keys()
test_datasets = {}
def make_pio_datasets(paths):
# get train & test dataset stocked on the datastore
train = pio.dataset.Dataset.get_by_name(dataset_name='regression_house_80')
test = pio.dataset.Dataset.get_by_name(dataset_name='regression_house_20')
#transform a var
train._data['bathrooms'] = train._data['bathrooms'].astype('int').apply(
lambda x: round(x))
test._data['bathrooms'] = test._data['bathrooms'].astype('int').apply(
lambda x: round(x))
#register new datasets
train_fe = pio.Dataset.new('regression_house_80_fe', dataframe=train._data)
test_fe = pio.Dataset.new('regression_house_20_fe', dataframe=test._data)
## auto ml use case starting
uc_config = pio.TrainingConfig(
models=[pio.Model.XGBoost, pio.Model.RandomForest],
features=pio.Feature.Full,
profile=pio.Profile.Quick,
with_blend=False)
col_config = pio.ColumnConfig(target_column='TARGET', id_column='ID')
uc = pio.Regression.fit('housing_from_sdk',
dataset=train_fe,
holdout_dataset=test_fe,
column_config=col_config,
training_config=uc_config)
uc.wait_until(lambda u: len(u) > 1)
## Get some Use case derived informations:
# correlation matrix
import os
import pandas as pd
import previsionio as pio
from .utils import get_testing_id
TESTING_ID = get_testing_id()
pio.config.default_timeout = 120
col_config = pio.ColumnConfig(target_column='class',
filename_column='filename')
uc_config = pio.TrainingConfig(models=[pio.Model.LinReg],
features=[pio.Feature.Counts],
profile=pio.Profile.Quick)
test_datasets = {}
dataset_name = 'cats_and_dogs_train'
dataset_test_name = TESTING_ID + '-' + dataset_name
def upload_datasets():
datapath = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'data_img/{}'.format(dataset_name))
# upload CSV reference file
dataset_csv = pio.Dataset.new(name=dataset_test_name,
dataframe=pd.read_csv(
os.path.join(
datapath,
'{}.csv'.format(dataset_name))))
# upload ZIP images folder
dataset_zip = pio.DatasetImages.new(name=dataset_test_name,