PUBG_Mobile：pyspark.ml，随机森林

背景与数据清洗可视化，请查看

https://blog.csdn.net/weixin_43502706/article/details/130777766?csdn_share_tail=%7B%22type%22%3A%22blog%22%2C%22rType%22%3A%22article%22%2C%22rId%22%3A%22130777766%22%2C%22source%22%3A%22weixin_43502706%22%7D
直接开始，读取数据

   spark = SparkSession.builder.\
            master("local[*]").\
            config('spark.sql.shuffle.partitions','4').\
            enableHiveSupport().\
            getOrCreate()
    data = spark.read.format('csv').\
        option('sep', ',').\
        option('header', True).\
        option('encoding', 'utf-8'). \
        option("inferSchema", True).\
        load('/tmp/pycharm_project_0508/data/train_V2.csv')

字段情况

    print(data.printSchema())  #字段情况
    print("原始数据条数：{}".format(data.count()))

剔除可能开挂的数据

    data_m = data.na.drop().\
        filter(data.kills<=20).\
        filter(data.roadKills<=3).\
        filter("not(walkDistance=0 and kills>0)").\
        filter("not(kills>3 and headshotKills!=kills)").\
        drop('Id', 'groupId', 'matchId', 'matchType'). \
        withColumnRenamed('winPlacePerc', 'label')
    print(data_m.count(),"删除数据{}".format(data.count()-data_m.count()))

向量化数据

    assembler = VectorAssembler(inputCols=data_m.columns[:-1], outputCol="features")
    df = assembler.transform(data_m)
    model_df = df.select(['features','label'])
    train, test = model_df.randomSplit([0.8, 0.2], seed=10)

创建随机森林模型

    # 创建随机森林模型
    rf = RandomForestRegressor(featuresCol='features', labelCol='label')
    model = rf.fit(train)
    predictions = model.transform(test)

评估

   evaluator = RegressionEvaluator(predictionCol="prediction", labelCol="label")
    mse = evaluator.evaluate(predictions, {evaluator.metricName: 'mse'})
    r2 = evaluator.evaluate(predictions, {evaluator.metricName: 'r2'})
    mae = evaluator.evaluate(predictions, {evaluator.metricName: 'mae'})
    print("MSE为:", mse)
    print("R2得分：", r2)
    print("MAE得分：", mae)