Examples with DeepARPredictStreamOp com.alibaba.alink.operator.stream.timeseries.DeepARPredictStreamOp used on opensource projects

Example 1 with DeepARPredictStreamOp

use of com.alibaba.alink.operator.stream.timeseries.DeepARPredictStreamOp in project Alink by alibaba.

the class DeepARPredictStreamOpTest method testDeepARTrainBatchOp.

@Test
public void testDeepARTrainBatchOp() throws Exception {
    BatchOperator.setParallelism(1);
    List<Row> data = Arrays.asList(Row.of(0, Timestamp.valueOf("2021-11-01 00:00:00"), 100.0), Row.of(0, Timestamp.valueOf("2021-11-02 00:00:00"), 100.0), Row.of(0, Timestamp.valueOf("2021-11-03 00:00:00"), 100.0), Row.of(0, Timestamp.valueOf("2021-11-04 00:00:00"), 100.0), Row.of(0, Timestamp.valueOf("2021-11-05 00:00:00"), 100.0));
    MemSourceBatchOp memSourceBatchOp = new MemSourceBatchOp(data, "id int, ts timestamp, series double");
    MemSourceStreamOp memSourceStreamOp = new MemSourceStreamOp(data, "id int, ts timestamp, series double");
    DeepARTrainBatchOp deepARTrainBatchOp = new DeepARTrainBatchOp().setTimeCol("ts").setSelectedCol("series").setNumEpochs(10).setWindow(2).setStride(1).linkFrom(memSourceBatchOp);
    OverCountWindowStreamOp overCountWindowStreamOp = new OverCountWindowStreamOp().setClause("MTABLE_AGG_PRECEDING(ts, series) as mtable_agg_series").setTimeCol("ts").setPrecedingRows(2);
    DeepARPredictStreamOp deepARPredictStreamOp = new DeepARPredictStreamOp(deepARTrainBatchOp).setPredictNum(2).setPredictionCol("pred").setValueCol("mtable_agg_series");
    deepARPredictStreamOp.linkFrom(overCountWindowStreamOp.linkFrom(memSourceStreamOp).filter("ts = TO_TIMESTAMP('2021-11-05 00:00:00')")).print();
    StreamOperator.execute();
}

Example 2 with DeepARPredictStreamOp

use of com.alibaba.alink.operator.stream.timeseries.DeepARPredictStreamOp in project Alink by alibaba.

the class DeepARTrainBatchOpTest method testSingleVar.

@Test
public void testSingleVar() throws Exception {
    BatchOperator.setParallelism(1);
    final String timeColName = "ts";
    BatchOperator<?> source = new RandomTableSourceBatchOp().setNumRows(1000L).setNumCols(1);
    String colName = source.getColNames()[0];
    AppendIdBatchOp appendIdBatchOp = new AppendIdBatchOp().setIdCol(timeColName).linkFrom(source);
    BatchOperator<?> timeBatchOp = new SelectBatchOp().setClause(String.format("%s, FLOOR(TO_TIMESTAMP(%s * 3600000) TO HOUR) as %s", colName, timeColName, timeColName)).linkFrom(appendIdBatchOp);
    StringBuilder groupByPredicate = new StringBuilder();
    String selectClause = timeColName + String.format(", SUM(%s) as %s", colName, colName);
    groupByPredicate.append(timeColName);
    BatchOperator<?> groupedTimeBatchOp = new GroupByBatchOp().setSelectClause(selectClause).setGroupByPredicate(groupByPredicate.toString()).linkFrom(timeBatchOp);
    BatchOperator<?> deepArTrainBatchOp = new DeepARTrainBatchOp().setSelectedCol(colName).setTimeCol(timeColName).setWindow(24 * 7).setStride(24).setNumEpochs(1).linkFrom(groupedTimeBatchOp);
    StreamOperator<?> sourceStreamOp = new RandomTableSourceStreamOp().setNumCols(1).setMaxRows(1000L);
    AppendIdStreamOp appendIdStreamOp = new AppendIdStreamOp().setIdCol(timeColName).linkFrom(sourceStreamOp);
    StreamOperator<?> timeStreamOp = new SelectStreamOp().setClause(String.format("%s, FLOOR(TO_TIMESTAMP(%s * 3600000) TO HOUR) as %s", colName, timeColName, timeColName)).linkFrom(appendIdStreamOp);
    String selectClausePred = String.format("TUMBLE_START() as %s", timeColName) + String.format(", SUM(%s) as %s", colName, colName);
    TumbleTimeWindowStreamOp timeWindowStreamOp = new TumbleTimeWindowStreamOp().setWindowTime(3600).setTimeCol(timeColName).setClause(selectClausePred).linkFrom(timeStreamOp);
    HopTimeWindowStreamOp hopTimeWindowStreamOp = new HopTimeWindowStreamOp().setTimeCol(timeColName).setClause(String.format("MTABLE_AGG(%s, %s) as %s", timeColName, colName, "mt")).setHopTime(24 * 3600).setWindowTime((24 * 7 - 24) * 3600).linkFrom(timeWindowStreamOp);
    DeepARPredictStreamOp deepARPredictStreamOp = new DeepARPredictStreamOp(deepArTrainBatchOp).setValueCol("mt").setPredictionCol("pred").setPredictNum(24).linkFrom(hopTimeWindowStreamOp);
    FilePath tmpAkFile = new FilePath(new Path(folder.getRoot().getPath(), "deepar_test_stream_single_var_result.ak"));
    deepARPredictStreamOp.link(new AkSinkStreamOp().setOverwriteSink(true).setFilePath(tmpAkFile));
    StreamOperator.execute();
}

Example 3 with DeepARPredictStreamOp

use of com.alibaba.alink.operator.stream.timeseries.DeepARPredictStreamOp in project Alink by alibaba.

the class DeepARTrainBatchOpTest method testMultiVar.

@Test
public void testMultiVar() throws Exception {
    BatchOperator.setParallelism(1);
    final String timeColName = "ts";
    final int numCols = 10;
    final String vecColName = "vec";
    BatchOperator<?> source = new RandomTableSourceBatchOp().setNumRows(1000L).setNumCols(numCols);
    String[] colNames = source.getColNames();
    AppendIdBatchOp appendIdBatchOp = new AppendIdBatchOp().setIdCol(timeColName).linkFrom(source);
    BatchOperator<?> timeBatchOp = new SelectBatchOp().setClause(String.format("%s, FLOOR(TO_TIMESTAMP(%s * 3600000) TO HOUR) as %s", Joiner.on(",").join(colNames), timeColName, timeColName)).linkFrom(appendIdBatchOp);
    StringBuilder selectClause = new StringBuilder();
    StringBuilder groupByPredicate = new StringBuilder();
    selectClause.append(timeColName);
    for (int i = 0; i < numCols; ++i) {
        selectClause.append(", ");
        selectClause.append(String.format("SUM(%s) as %s", colNames[i], colNames[i]));
    }
    groupByPredicate.append(timeColName);
    BatchOperator<?> groupedTimeBatchOp = new GroupByBatchOp().setSelectClause(selectClause.toString()).setGroupByPredicate(groupByPredicate.toString()).linkFrom(timeBatchOp);
    ColumnsToVectorBatchOp columnsToVectorBatchOp = new ColumnsToVectorBatchOp().setSelectedCols(colNames).setVectorCol(vecColName).linkFrom(groupedTimeBatchOp);
    BatchOperator<?> deepArTrainBatchOp = new DeepARTrainBatchOp().setVectorCol(vecColName).setTimeCol(timeColName).setWindow(24 * 7).setStride(24).setNumEpochs(1).linkFrom(columnsToVectorBatchOp);
    StreamOperator<?> sourceStreamOp = new RandomTableSourceStreamOp().setNumCols(numCols).setMaxRows(1000L);
    AppendIdStreamOp appendIdStreamOp = new AppendIdStreamOp().setIdCol(timeColName).linkFrom(sourceStreamOp);
    StreamOperator<?> timeStreamOp = new SelectStreamOp().setClause(String.format("%s, FLOOR(TO_TIMESTAMP(%s * 3600000) TO HOUR) as %s", Joiner.on(",").join(colNames), timeColName, timeColName)).linkFrom(appendIdStreamOp);
    StringBuilder selectClausePred = new StringBuilder();
    selectClausePred.append(String.format("TUMBLE_START() as %s", timeColName));
    for (int i = 0; i < numCols; ++i) {
        selectClausePred.append(", ");
        selectClausePred.append(String.format("SUM(%s) as %s", colNames[i], colNames[i]));
    }
    TumbleTimeWindowStreamOp timeWindowStreamOp = new TumbleTimeWindowStreamOp().setWindowTime(3600).setTimeCol(timeColName).setClause(selectClausePred.toString()).linkFrom(timeStreamOp);
    ColumnsToVectorStreamOp columnsToVectorStreamOp = new ColumnsToVectorStreamOp().setSelectedCols(colNames).setVectorCol(vecColName).linkFrom(timeWindowStreamOp);
    HopTimeWindowStreamOp hopTimeWindowStreamOp = new HopTimeWindowStreamOp().setTimeCol(timeColName).setClause(String.format("MTABLE_AGG(%s, %s) as %s", timeColName, vecColName, "mt")).setHopTime(24 * 3600).setWindowTime((24 * 7 - 24) * 3600).linkFrom(columnsToVectorStreamOp);
    DeepARPredictStreamOp deepARPredictStreamOp = new DeepARPredictStreamOp(deepArTrainBatchOp).setValueCol("mt").setPredictionCol("pred").linkFrom(hopTimeWindowStreamOp);
    FilePath tmpAkFile = new FilePath(new Path(folder.getRoot().getPath(), "deepar_test_stream_multi_var_result.ak"));
    deepARPredictStreamOp.link(new AkSinkStreamOp().setOverwriteSink(true).setFilePath(tmpAkFile));
    StreamOperator.execute();
}