Platform For AI:GBDT回帰

コンポーネントの設定

次のいずれかの方法を使用して、GBDT回帰コンポーネントを設定できます。

方法1: パイプラインページでコンポーネントを設定する

方法2: PAIコマンドを使用する

PAI -name gbdt
    -project algo_public
    -DfeatureSplitValueMaxSize="500"
    -DlossType="0"
    -DrandSeed="0"
    -DnewtonStep="0"
    -Dshrinkage="0.05"
    -DmaxLeafCount="32"
    -DlabelColName="campaign"
    -DinputTableName="bank_data_partition"
    -DminLeafSampleCount="500"
    -DsampleRatio="0.6"
    -DgroupIDColName="age"
    -DmaxDepth="11"
    -DmodelName="xlab_m_GBDT_83602"
    -DmetricType="2"
    -DfeatureRatio="0.6"
    -DinputTablePartitions="pt=20150501"
    -Dtau="0.6"
    -Dp="1"
    -DtestRatio="0.0"
    -DfeatureColNames="previous,cons_conf_idx,euribor3m"
    -DtreeCount="500"

例：

1. 次のSQL文を実行してテストデータを生成します。

   drop table if exists gbdt_ls_test_input;
   create table gbdt_ls_test_input
   as
   select
       *
   from
   (
       select
           cast(1 as double) as f0,
           cast(0 as double) as f1,
           cast(0 as double) as f2,
           cast(0 as double) as f3,
           cast(0 as bigint) as label
       union all
           select
               cast(0 as double) as f0,
               cast(1 as double) as f1,
               cast(0 as double) as f2,
               cast(0 as double) as f3,
               cast(0 as bigint) as label
       union all
           select
               cast(0 as double) as f0,
               cast(0 as double) as f1,
               cast(1 as double) as f2,
               cast(0 as double) as f3,
               cast(1 as bigint) as label
       union all
           select
               cast(0 as double) as f0,
               cast(0 as double) as f1,
               cast(0 as double) as f2,
               cast(1 as double) as f3,
               cast(1 as bigint) as label
       union all
           select
               cast(1 as double) as f0,
               cast(0 as double) as f1,
               cast(0 as double) as f2,
               cast(0 as double) as f3,
               cast(0 as bigint) as label
       union all
           select
               cast(0 as double) as f0,
               cast(1 as double) as f1,
               cast(0 as double) as f2,
               cast(0 as double) as f3,
               cast(0 as bigint) as label
   ) a;

   次のテストデータテーブルgbdt_ls_test_inputが生成されます。

   f0	f1	f2	f3	ラベル
   1.0	0.0	0.0	0.0	0
   0.0	0.0	1.0	0.0	1
   0.0	0.0	0.0	1.0	1
   0.0	1.0	0.0	0.0	0
   1.0	0.0	0.0	0.0	0
   0.0	1.0	0.0	0.0	0

2. 次のPAIコマンドを実行して、GBDT Regressionコンポーネントに設定されたトレーニングパラメータを送信します。

   gbdt_ls_test_modelが存在する場合は、オフラインモデルをドロップし

   drop offlinemodel if exists gbdt_ls_test_model;
   PAI -name gbdt
       -project algo_public
       -DfeatureSplitValueMaxSize="500"
       -DlossType="3"
       -DrandSeed="0"
       -DnewtonStep="1"
       -Dshrinkage="0.5"
       -DmaxLeafCount="32"
       -DlabelColName="label"
       -DinputTableName="gbdt_ls_test_input"
       -DminLeafSampleCount="1"
       -DsampleRatio="1"
       -DmaxDepth="10"
       -DmetricType="0"
       -DmodelName="gbdt_ls_test_model"
       -DfeatureRatio="1"
       -Dp="1"
       -Dtau="0.6"
       -DtestRatio="0"
       -DfeatureColNames="f0,f1,f2,f3"
       -DtreeCount="10"

3. 次のPAIコマンドを実行して、予測コンポーネントに設定されたパラメーターを送信します。

   drop table if exists gbdt_ls_test_prediction_result;
   PAI -name prediction
       -project algo_public
       -DdetailColName="prediction_detail"
       -DmodelName="gbdt_ls_test_model"
       -DitemDelimiter=","
       -DresultColName="prediction_result"
       -Dlifecycle="28"
       -DoutputTableName="gbdt_ls_test_prediction_result"
       -DscoreColName="prediction_score"
       -DkvDelimiter=":"
       -DinputTableName="gbdt_ls_test_input"
       -DenableSparse="false"
       -DappendColNames="label"

4. 予測結果テーブルgbdt_ls_test_prediction_resultを表示します。

   ラベル	prediction_result	prediction_score	予測_詳細
   0	NULL	0.0	{"label": 0}
   0	NULL	0.0	{"label": 0}
   1	NULL	0.9990234375	{"label": 0.9990234375}
   1	NULL	0.9990234375	{"label": 0.9990234375}
   0	NULL	0.0	{"label": 0}
   0	NULL	0.0	{"label": 0}