Help Center/ ModelArts/ Model Inference/ Inference Specifications/ Examples of Custom Scripts/ PySpark
Updated on 2024-06-12 GMT+08:00
View PDF

PySpark

Training and Saving a Model

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
 
from pyspark.ml import Pipeline, PipelineModel
from pyspark.ml.linalg import Vectors
from pyspark.ml.classification import LogisticRegression

# Prepare training data using tuples.
# Prepare training data from a list of (label, features) tuples.
training = spark.createDataFrame([
    (1.0, Vectors.dense([0.0, 1.1, 0.1])),
    (0.0, Vectors.dense([2.0, 1.0, -1.0])),
    (0.0, Vectors.dense([2.0, 1.3, 1.0])),
    (1.0, Vectors.dense([0.0, 1.2, -0.5]))], ["label", "features"])

# Create a training instance. The logistic regression algorithm is used for training.
# Create a LogisticRegression instance. This instance is an Estimator.
lr = LogisticRegression(maxIter=10, regParam=0.01)

# Train the logistic regression model.
# Learn a LogisticRegression model. This uses the parameters stored in lr.
model = lr.fit(training)

# Save the model to a local directory.
# Save model to local path.
model.save("/tmp/spark_model")

After the model is saved, it must be uploaded to the OBS directory before being published. The config.json configuration and the customize_service.py inference code must be included during the publishing. For details about how to compile config.json, see Specifications for Editing a Model Configuration File. For details about inference code, see Inference Code.

Inference Code

In the model inference code file customize_service.py, add a child model class. This child model class inherits properties from its parent model class. For details about the import statements of different types of parent model classes, see Table 1.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
 
# coding:utf-8
import collections
import json
import traceback

import model_service.log as log
from model_service.spark_model_service import SparkServingBaseService
from pyspark.ml.classification import LogisticRegression

logger = log.getLogger(__name__)


class UserService(SparkServingBaseService):
    # Pre-process data.
    def _preprocess(self, data):
        logger.info("Begin to handle data from user data...")
        # Read data.
        req_json = json.loads(data, object_pairs_hook=collections.OrderedDict)
        try:
            # Convert data to the spark dataframe format.
            predict_spdf = self.spark.createDataFrame(pd.DataFrame(req_json["data"]["req_data"]))
        except Exception as e:
            logger.error("check your request data does meet the requirements ?")
            logger.error(traceback.format_exc())
            raise Exception("check your request data does meet the requirements ?")
        return predict_spdf

    # Perform model inference.
    def _inference(self, data):
        try:
             # Load a model file.
            predict_model = LogisticRegression.load(self.model_path)
            # Perform data inference.
            prediction_result = predict_model.transform(data)
 except Exception as e:
            logger.error(traceback.format_exc())
            raise Exception("Unable to load model and do dataframe transformation.")
        return prediction_result

    # Post-process data.
    def _postprocess(self, pre_data):
        logger.info("Get new data to respond...")
        predict_str = pre_data.toPandas().to_json(orient='records')
        predict_result = json.loads(predict_str)
        return predict_result