分布式机器学习

Prepare data

from pyspark.conf import SparkConf
from pyspark.sql import SparkSession
from pyspark.ml import Pipeline
from pyspark.ml.feature import VectorAssembler
from pyspark.ml.evaluation import BinaryClassificationEvaluator
from pyspark.sql import Row, functions as fn
import pyspark.ml.feature as ft

import pandas as pd
import numpy as np
import time
import warnings

# Setting configuration.
warnings.filterwarnings('ignore')
SEED = 42

# Use 0.11.4-spark3.3 version for Spark3.3 and 1.0.2 version for Spark3.4
spark = SparkSession.builder.appName("MyApp") \
            .config("spark.driver.memory", "5g") \
            .config("spark.driver.cores", "2") \
            .config("spark.executor.memory", "5g") \
            .config("spark.executor.cores", "2") \
            .enableHiveSupport() \
            .getOrCreate()

sc = spark.sparkContext
sc.setLogLevel('ERROR')

24/04/26 21:44:36 WARN Utils: Your hostname, MacBook-Air resolves to a loopback address: 127.0.0.1; using 192.168.1.4 instead (on interface en0)
24/04/26 21:44:36 WARN Utils: Set SPARK_LOCAL_IP if you need to bind to another address
Setting default log level to "WARN".
To adjust logging level use sc.setLogLevel(newLevel). For SparkR, use setLogLevel(newLevel).
24/04/26 21:44:37 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable

# Load dataset
path = '/Users/***/Documents/Project/datasets/Home-Credit-Default-Risk/prepared_data.csv'
data = spark.read.format("csv").option("header", True).load(f"file://{path}")

# print dataset size
print("records read: " + str(data.count()))

records read: 307511

# Check if the data is unbalanced
data.groupBy("TARGET").count().show()

+------+------+
|TARGET| count|
+------+------+
|     0|282686|
|     1| 24825|
+------+------+

for colname in data.columns:
    data = data.withColumn(colname, data[colname].cast('float'))

# Add featurizer to convert features to vector
feature_name = data.columns[1:-1]
assembler = VectorAssembler(
    inputCols=feature_name,
    outputCol="features"
)
data = assembler.transform(data)['features', 'TARGET']

# Split the data into train and test.
train, test = data.randomSplit([0.75, 0.25], seed=SEED)

Spark MLlib

from pyspark.ml.classification import RandomForestClassifier

gbt = RandomForestClassifier(
    labelCol="TARGET", 
    featuresCol="features", 
    maxDepth=8,
    numTrees=500,
    subsamplingRate=1.0,
    featureSubsetStrategy='auto',
    seed=SEED
)

# Train a GBT model.
model = gbt.fit(train)

# Select (prediction, true label) and compute areaUnderROC
evaluator = BinaryClassificationEvaluator(
    labelCol="TARGET", 
    metricName='areaUnderROC'
)
train_auc = evaluator.evaluate(model.transform(train))
test_auc = evaluator.evaluate(model.transform(test))

print(f"Train auc: {train_auc:.4f}")
print(f"Test auc: {test_auc:.4f}")

Java HotSpot(TM) 64-Bit Server VM warning: CodeCache is full. Compiler has been disabled.
Java HotSpot(TM) 64-Bit Server VM warning: Try increasing the code cache size using -XX:ReservedCodeCacheSize=


CodeCache: size=131072Kb used=38814Kb max_used=39023Kb free=92257Kb
 bounds [0x000000010464c000, 0x0000000106c9c000, 0x000000010c64c000]
 total_blobs=13345 nmethods=12309 adapters=949
 compilation: disabled (not enough contiguous free space left)        

Train auc: 0.7526
Test auc: 0.7235

feature_imp = pd.Series(
    model.featureImportances.toArray(),
    index=assembler.getInputCols()
).sort_values(ascending=False)

print(feature_imp.head(20))

EXT_SOURCE_2                   0.183568
EXT_SOURCE_3                   0.175979
EXT_SOURCE_1                   0.094980
DAYS_EMPLOYED                  0.050050
OCCUPATION_TYPE                0.032153
DAYS_BIRTH                     0.032032
NAME_EDUCATION_TYPE            0.025601
DAYS_LAST_PHONE_CHANGE         0.022394
AMT_GOODS_PRICE                0.019779
REGION_RATING_CLIENT_W_CITY    0.014936
CODE_GENDER_M                  0.014736
REGION_RATING_CLIENT           0.012078
ORGANIZATION_TYPE              0.011209
AMT_CREDIT                     0.010922
NAME_INCOME_TYPE_Working       0.010745
DAYS_ID_PUBLISH                0.010505
FLAG_DOCUMENT_3                0.009315
OWN_CAR_AGE                    0.009004
AMT_ANNUITY                    0.007916
TOTALAREA_MODE                 0.007510
dtype: float64

XGBoost with spark

from xgboost.spark import SparkXGBClassifier
import xgboost as xgb

train = train.withColumn('isVal', fn.rand() < 0.2)

xgb_clf = SparkXGBClassifier(
    features_col='features', 
    label_col='TARGET',
    # validation_indicator_col='isVal', 
    eval_metric='auc',
    scale_pos_weight=11,
    learning_rate=0.015,
    max_depth=8,
    subsample=1.0,
    colsample_bytree=0.35,
    reg_alpha=65,
    reg_lambda=15,
    # early_stopping_rounds=20, 
    n_estimators=1200,
    verbosity=0
)
xgb_model = xgb_clf.fit(train)

train_auc = evaluator.evaluate(xgb_model.transform(train))
test_auc = evaluator.evaluate(xgb_model.transform(test))

print(f"Train auc: {train_auc:.4f}")
print(f"Test auc: {test_auc:.4f}")

2024-04-26 21:48:31,541 INFO XGBoost-PySpark: _fit Running xgboost-2.0.3 on 1 workers with
	booster params: {'objective': 'binary:logistic', 'colsample_bytree': 0.35, 'device': 'cpu', 'learning_rate': 0.015, 'max_depth': 8, 'reg_alpha': 65, 'reg_lambda': 15, 'scale_pos_weight': 11, 'subsample': 1.0, 'verbosity': 0, 'eval_metric': 'auc', 'nthread': 1}
	train_call_kwargs_params: {'verbose_eval': True, 'num_boost_round': 1200}
	dmatrix_kwargs: {'nthread': 1, 'missing': nan}
[21:48:37] task 0 got new rank 0                                    (0 + 1) / 1]
2024-04-26 21:50:12,299 INFO XGBoost-PySpark: _fit Finished xgboost training!   
INFO:XGBoost-PySpark:Do the inference on the CPUs                   (0 + 8) / 8]
2024-04-26 21:52:34,899 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:52:35,749 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:52:35,830 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:52:35,838 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:52:35,882 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:52:35,895 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:52:35,920 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:57:30,388 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:57:30,806 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:57:30,938 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:57:30,941 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
INFO:XGBoost-PySpark:Do the inference on the CPUs
2024-04-26 21:57:30,982 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:57:31,003 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs
2024-04-26 21:57:31,007 INFO XGBoost-PySpark: predict_udf Do the inference on the CPUs


Train auc: 0.8524
Test auc: 0.7626

feature_imp = xgb_model.get_feature_importances()
indices = [int(name[1:]) for name in feature_imp.keys()]

feature_imp = pd.Series(
    feature_imp.values(),
    index=np.array(feature_name)[indices]
).sort_values(ascending=False)

print(feature_imp.head(20))

DAYS_BIRTH                    5801.0
AMT_ANNUITY                   5501.0
DAYS_REGISTRATION             5303.0
EXT_SOURCE_2                  5275.0
DAYS_ID_PUBLISH               5231.0
EXT_SOURCE_3                  4985.0
AMT_CREDIT                    4893.0
DAYS_LAST_PHONE_CHANGE        4798.0
DAYS_EMPLOYED                 4634.0
EXT_SOURCE_1                  4610.0
anomaly_score                 4529.0
AMT_GOODS_PRICE               4387.0
AMT_INCOME_TOTAL              3886.0
REGION_POPULATION_RELATIVE    3813.0
ORGANIZATION_TYPE             3130.0
OWN_CAR_AGE                   3017.0
HOUR_APPR_PROCESS_START       2845.0
OCCUPATION_TYPE               2504.0
AMT_REQ_CREDIT_BUREAU_YEAR    1975.0
TOTALAREA_MODE                1655.0
dtype: float64

LightGBM with spark

from synapse.ml.lightgbm import LightGBMClassifier
import lightgbm as lgb

lgb_clf = LightGBMClassifier(
    featuresCol="features", 
    labelCol="TARGET",
    boostingType='gbdt',
    objective='binary',
    metric='auc',
    isUnbalance=True,
    learningRate=0.015,
    numIterations=1200,
    maxDepth=8,
    featureFraction=0.35,
    baggingFraction=1.0,
    lambdaL1=65,
    lambdaL2=15,
    # subsampleFreq=5,
    earlyStoppingRound=20,
    dataRandomSeed=SEED,
    verbosity=-1
)

lgb_model = lgb_clf.fit(train)

train_auc = evaluator.evaluate(lgb_model.transform(train))
test_auc = evaluator.evaluate(lgb_model.transform(test))

print(f"Train auc: {train_auc:.4f}")
print(f"Test auc: {test_auc:.4f}")

LightGBM的参数比SynapseML公开的要多得多，若要添加额外的参数，请使用passThroughArgs字符串参数配置。您可以混合passThroughArgs和显式args，SynapseML合并它们以创建一个要发送到LightGBM的参数字符串。如果您在两个地方都设置参数，则以passThroughArgs为优先。

feature_imp = pd.Series(
    lgb_model.getFeatureImportances(),
    index=assembler.get_inputCols()
).sort_values(ascending=False)

print(feature_imp.head(20))

spark.stop()