# Define comfigs
# allowed arguments are: randomforest, sklearn, deeplearning
# randomforest will perform 1 run of randomforest fit
# sklearnmodels will fit 15 models from sklearn
# deeplearning will fit a neural network with pytorch
models = 'randomforest'
data_local = False
# if data_local is true, subset is alwats true
subset = False
# hyperdrive only works with deeplearning
hyperdrive = False
# If deep learning define hyperparameters
# Set parameters for search
param_sampling = BayesianParameterSampling({
"learning_rate": uniform(0.05, 0.1),
"num_epochs": choice(5, 10, 15),
"batch_size": choice(150, 200),
"hidden_size": choice(50, 100)
})
# load Azure ML workspace
workspace = Workspace.from_config(auth=AzureCliAuthentication())
if subset is True:
# define data set names
input_name_train = 'newsgroups_subset_train'
input_name_test = 'newsgroups_subset_test'
filepath = "environments/sklearn_subset/RunConfig/runconfig_subset.yml"
else:
input_name_train = 'newsgroups_train'
# In[62]:
from azureml.widgets import RunDetails
from azureml.train.sklearn import SKLearn
from azureml.train.hyperdrive.run import PrimaryMetricGoal
from azureml.train.hyperdrive.policy import BanditPolicy
from azureml.train.hyperdrive.sampling import RandomParameterSampling
from azureml.train.hyperdrive.runconfig import HyperDriveConfig
from azureml.train.hyperdrive.parameter_expressions import uniform
import os
# Specify parameter sampler
from azureml.train.hyperdrive.parameter_expressions import choice
ps = RandomParameterSampling( {
"--C": uniform(0.1,1),
"--max_iter": choice(50,100,150,200)
})
# Specify a Policy
policy = BanditPolicy(slack_factor = 0.1, evaluation_interval=1, delay_evaluation=5)
if "training" not in os.listdir():
os.mkdir("./training")
# Create a SKLearn estimator for use with train.py
est = SKLearn(source_directory = '.', entry_script = 'train.py', compute_target = cpu_cluster)
# Create a HyperDriveConfig using the estimator, hyperparameter sampler, and policy.
hyperdrive_config = HyperDriveConfig(hyperparameter_sampling = ps,
primary_metric_name = 'Accuracy',
datadrift_name = 'data_drift_report'
datadriftreport = PipelineData(name='data_drift_report',
datastore=datastore,
pipeline_output_name=datadrift_name)
datadrift_subset_name = 'data_drift_report_subset'
datadriftreportsubset = PipelineData(
name='data_drift_report_subset',
datastore=datastore,
pipeline_output_name=datadrift_subset_name)
# Set parameters for search
param_sampling = BayesianParameterSampling({
"learning_rate":
uniform(10e-6, 1e0),
"num_epochs":
choice(10, 20),
"batch_size":
choice(10, 20, 50, 100, 200, 300, 500, 1000),
"hidden_size":
choice(300, 400)
})
# LOAD ALL SCRIPT PARAMETERS FOR EVERY STEP IN PIPELINE
script_params_data_validation = [
'--data_folder_train',
dataset_train.as_named_input('train').as_mount(), '--data_folder_test',
dataset_test.as_named_input('test').as_mount(), '--local', 'no',
'--output_train', train_validated, '--output_test', test_validated,
'--data_drift_report', datadriftreport