aws--amazon-sagemaker-examples
wehub-resource-sync
82 行
2.2 KiB
Python
82 行
2.2 KiB
Python
import os
|
|
os.system('python3 -m pip install -U sagemaker xgboost==1.0.1 scikit-learn')
|
|
|
|
import glob
|
|
import boto3
|
|
import json
|
|
import pathlib
|
|
import pickle
|
|
import tarfile
|
|
from datetime import datetime
|
|
import joblib
|
|
import numpy as np
|
|
import pandas as pd
|
|
import xgboost
|
|
import argparse
|
|
|
|
from sklearn.metrics import mean_squared_error
|
|
|
|
from sagemaker.session import Session
|
|
from sagemaker.experiments.run import Run, load_run
|
|
|
|
session = Session(boto3.session.Session(region_name="us-east-1"))
|
|
|
|
|
|
def read_parameters():
|
|
"""
|
|
Read job parameters
|
|
Returns:
|
|
(Namespace): read parameters
|
|
"""
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument('--model_path', type=str, default="/opt/ml/processing/model")
|
|
parser.add_argument('--test_csv', type=str, default="/opt/ml/processing/output/test")
|
|
parser.add_argument('--output_path', type=str, default="/opt/ml/processing/evaluation")
|
|
params, _ = parser.parse_known_args()
|
|
return params
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
# reading job parameters
|
|
args = read_parameters()
|
|
|
|
available_model = glob.glob(os.path.join(args.model_path, "*.tar.gz"))
|
|
assert len(available_model) == 1, f"multiple tar gz models found in models dir: {available_model}"
|
|
print(f"Model found: {available_model}")
|
|
model_path = available_model[0]
|
|
|
|
with tarfile.open(model_path) as tar:
|
|
tar.extractall(path=".")
|
|
|
|
model = pickle.load(open("xgboost-model", "rb"))
|
|
|
|
test_path = os.path.join(args.test_csv, "test.csv")
|
|
df = pd.read_csv(test_path, header=None)
|
|
|
|
y_test = df.iloc[:, 0].to_numpy()
|
|
df.drop(df.columns[0], axis=1, inplace=True)
|
|
|
|
X_test = xgboost.DMatrix(df.values)
|
|
|
|
predictions = model.predict(X_test)
|
|
|
|
mse = mean_squared_error(y_test, predictions)
|
|
std = np.std(y_test - predictions)
|
|
report_dict = {
|
|
"regression_metrics": {
|
|
"mse": {"value": mse, "standard_deviation": std},
|
|
}
|
|
}
|
|
|
|
print(report_dict)
|
|
|
|
output_dir = args.output_path
|
|
pathlib.Path(output_dir).mkdir(parents=True, exist_ok=True)
|
|
|
|
evaluation_path = f"{output_dir}/evaluation.json"
|
|
with open(evaluation_path, "w") as f:
|
|
f.write(json.dumps(report_dict))
|
|
|
|
print("Done")
|