mlSpark/scripts/train.py at main · AngeloM15/mlSpark · GitHub

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
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
"""Train Logistic Regression classifier with Apache SparkML"""
import logging
import os
import re
import shutil
import site
import sys

import wget
from pyspark import SparkConf, SparkContext, SQLContext
from pyspark2pmml import PMMLBuilder
from pyspark.ml import Pipeline
from pyspark.ml.classification import LogisticRegression
from pyspark.ml.evaluation import MulticlassClassificationEvaluator
from pyspark.ml.feature import MinMaxScaler, StringIndexer, VectorAssembler

from config import Config


def main():
    if sys.version[0:3] == "3.9":
        url = (
            "https://github.com/jpmml/jpmml-sparkml/releases/download/1.7.2/"
            "jpmml-sparkml-executable-1.7.2.jar"
        )
        wget.download(url)
        shutil.copy(
            "jpmml-sparkml-executable-1.7.2.jar",
            site.getsitepackages()[0] + "/pyspark/jars/",
        )
    elif sys.version[0:3] == "3.8":
        url = (
            "https://github.com/jpmml/jpmml-sparkml/releases/download/1.7.2/"
            "jpmml-sparkml-executable-1.7.2.jar"
        )
        wget.download(url)
        shutil.copy(
            "jpmml-sparkml-executable-1.7.2.jar",
            site.getsitepackages()[0] + "/pyspark/jars/",
        )
    elif sys.version[0:3] == "3.7":
        url = (
            "https://github.com/jpmml/jpmml-sparkml/releases/download/1.5.12/"
            "jpmml-sparkml-executable-1.5.12.jar"
        )
        wget.download(url)
    elif sys.version[0:3] == "3.6":
        url = (
            "https://github.com/jpmml/jpmml-sparkml/releases/download/1.5.12/"
            "jpmml-sparkml-executable-1.5.12.jar"
        )
        wget.download(url)
    else:
        raise Exception(
            "Currently only python 3.6 , 3.7, 3,8 and 3.9 is supported, in case "
            "you need a different version please open an issue at "
            "https://github.com/IBM/claimed/issues"
        )

    data_parquet = Config.data_parquet  # input file name (parquet)
    master = Config.master  # URL to Spark master
    model_target = Config.model  # model output file name

    input_columns = os.environ.get(
        "input_columns", '["x", "y", "z"]'
    )  # input columns to consider

    parameters = list(
        map(
            lambda s: re.sub("$", '"', s),
            map(
                lambda s: s.replace("=", '="'),
                filter(
                    lambda s: s.find("=") > -1
                    and bool(re.match(r"[A-Za-z0-9_]*=[.\/A-Za-z0-9]*", s)),
                    sys.argv,
                ),
            ),
        )
    )

    for parameter in parameters:
        logging.warning("Parameter: %s", parameter)
        exec(parameter)

    os.environ[
        "JAVA_HOME"
    ] = "/Library/Java/JavaVirtualMachines/adoptopenjdk-8.jdk/Contents/Home"

    conf = SparkConf().setMaster(master)
    # if sys.version[0:3] == '3.6' or sys.version[0:3] == '3.7':
    conf.set("spark.jars", "jpmml-sparkml-executable-1.5.12.jar")

    sc = SparkContext.getOrCreate(conf)
    sqlContext = SQLContext(sc)
    spark = sqlContext.sparkSession

    df = spark.read.parquet(data_parquet)

    # register a corresponding query table
    df.createOrReplaceTempView("df")

    from pyspark.sql.types import DoubleType

    df = df.withColumn("x", df.x.cast(DoubleType()))
    df = df.withColumn("y", df.y.cast(DoubleType()))
    df = df.withColumn("z", df.z.cast(DoubleType()))

    splits = df.randomSplit([0.8, 0.2])
    df_train = splits[0]
    df_test = splits[1]

    indexer = StringIndexer(inputCol="class", outputCol="label")

    vectorAssembler = VectorAssembler(
        inputCols=eval(input_columns), outputCol="features"
    )

    normalizer = MinMaxScaler(inputCol="features", outputCol="features_norm")

    lr = LogisticRegression(maxIter=1000, regParam=2.0, elasticNetParam=1.0)

    pipeline = Pipeline(stages=[indexer, vectorAssembler, normalizer, lr])

    model = pipeline.fit(df_train)

    prediction = model.transform(df_train)

    binEval = (
        MulticlassClassificationEvaluator()
        .setMetricName("accuracy")
        .setPredictionCol("prediction")
        .setLabelCol("label")
    )

    binEval.evaluate(prediction)

    pmmlBuilder = PMMLBuilder(sc, df_train, model)
    pmmlBuilder.buildFile(model_target)


if __name__ == "__main__":
    main()