mirror of
https://github.com/Azure/MachineLearningNotebooks.git
synced 2025-12-19 17:17:04 -05:00
76 lines
2.4 KiB
Python
76 lines
2.4 KiB
Python
import argparse
|
|
import os
|
|
import azureml.core
|
|
from datetime import datetime
|
|
import pandas as pd
|
|
import pytz
|
|
from azureml.core import Dataset, Model
|
|
from azureml.core.run import Run, _OfflineRun
|
|
from azureml.core import Workspace
|
|
|
|
run = Run.get_context()
|
|
ws = None
|
|
if type(run) == _OfflineRun:
|
|
ws = Workspace.from_config()
|
|
else:
|
|
ws = run.experiment.workspace
|
|
|
|
|
|
def write_output(df, path):
|
|
os.makedirs(path, exist_ok=True)
|
|
print("%s created" % path)
|
|
df.to_csv(path + "/part-00000", index=False)
|
|
|
|
|
|
print("Check for new data and prepare the data")
|
|
|
|
parser = argparse.ArgumentParser("split")
|
|
parser.add_argument("--target_column", type=str, help="input split features")
|
|
parser.add_argument("--ds_name", help="input dataset name")
|
|
parser.add_argument("--model_name", help="name of the deployed model")
|
|
parser.add_argument("--output_x", type=str,
|
|
help="output features")
|
|
parser.add_argument("--output_y", type=str,
|
|
help="output labels")
|
|
|
|
|
|
args = parser.parse_args()
|
|
|
|
print("Argument 1(ds_name): %s" % args.ds_name)
|
|
print("Argument 2(target_column): %s" % args.target_column)
|
|
print("Argument 3(model_name): %s" % args.model_name)
|
|
print("Argument 4(output_x): %s" % args.output_x)
|
|
print("Argument 5(output_y): %s" % args.output_y)
|
|
|
|
# Get the latest registered model
|
|
try:
|
|
model = Model(ws, args.model_name)
|
|
last_train_time = model.created_time
|
|
print("Model was last trained on {0}.".format(last_train_time))
|
|
except Exception as e:
|
|
print("Could not get last model train time.")
|
|
last_train_time = datetime.min.replace(tzinfo=pytz.UTC)
|
|
|
|
train_ds = Dataset.get_by_name(ws, args.ds_name)
|
|
dataset_changed_time = train_ds.data_changed_time
|
|
|
|
if dataset_changed_time > last_train_time:
|
|
# New data is available since the model was last trained
|
|
print("Dataset was last updated on {0}. Retraining...".format(dataset_changed_time))
|
|
train_ds = train_ds.drop_columns(["partition_date"])
|
|
X_train = train_ds.drop_columns(
|
|
columns=[args.target_column]).to_pandas_dataframe()
|
|
y_train = train_ds.keep_columns(
|
|
columns=[args.target_column]).to_pandas_dataframe()
|
|
|
|
non_null = y_train[args.target_column].notnull()
|
|
y = y_train[non_null]
|
|
X = X_train[non_null]
|
|
|
|
if not (args.output_x is None and args.output_y is None):
|
|
write_output(X, args.output_x)
|
|
write_output(y, args.output_y)
|
|
else:
|
|
print("Cancelling run since there is no new data.")
|
|
run.parent.cancel()
|