Java 类名:com.alibaba.alink.operator.stream.timeseries.ProphetStreamOp
Python 类名:ProphetStreamOp

功能介绍

对每一行的MTable数据, 进行Prophet时间序列预测,给出下一时间段的预测结果。

使用方式

参考文档 https://www.yuque.com/pinshu/alink_guide/xbp5ky

算法原理

Prophet是facebook开源的一个时间序列预测算法, github地址:https://github.com/facebook/prophet.
Prophet适用于具有明显的内在规律的数据, 例如:

  • 有一定的历史数据,有至少几个月的每小时、每天或每周观察的历史数据
  • 有较强的季节性趋势:每周的一些天,每年的一些时间
  • 有已知的以不定期的间隔发生的重要节假日(比如国庆节)
  • 缺失的历史数据或较大的异常数据的数量在合理范围内
  • 对于数据中蕴含的非线性增长的趋势都有一个自然极限或饱和状态

    使用方式

    参考文档 https://www.yuque.com/pinshu/alink_guide/xbp5ky

    参数说明

    | 名称 | 中文名称 | 描述 | 类型 | 是否必须? | 取值范围 | 默认值 | | —- | —- | —- | —- | —- | —- | —- |

| predictionCol | 预测结果列名 | 预测结果列名 | String | ✓ | | |

| valueCol | value列,类型为MTable | value列,类型为MTable | String | ✓ | 所选列类型为 [M_TABLE] | |

| predictNum | 预测条数 | 预测条数 | Integer | | | 1 |

| predictionDetailCol | 预测详细信息列名 | 预测详细信息列名 | String | | | |

| pythonEnv | Python 环境路径 | Python 环境路径,一般情况下不需要填写。如果是压缩文件,需要解压后得到一个目录,且目录名与压缩文件主文件名一致,可以使用 http://, https://, oss://, hdfs:// 等路径;如果是目录,那么只能使用本地路径,即 file://。 | String | | | “” |

| reservedCols | 算法保留列名 | 算法保留列 | String[] | | | null |

| stanInit | 初始值 | 初始值 | String | | | null |

| uncertaintySamples | 用来计算指标的采样数目 | 用来计算指标的采样数目,设置成0,不计算指标。 | Integer | | | 1000 |

| numThreads | 组件多线程线程个数 | 组件多线程线程个数 | Integer | | | 1 |

代码示例

Python 代码

  1. from pyalink.alink import *
  2. import pandas as pd
  3. useLocalEnv(1)
  4. import time, datetime
  5. import numpy as np
  6. import pandas as pd
  7. downloader = AlinkGlobalConfiguration.getPluginDownloader()
  8. downloader.downloadPlugin('tf115_python_env_linux')
  9. data = pd.DataFrame([
  10. [1, datetime.datetime.fromtimestamp(1), 10.0],
  11. [1, datetime.datetime.fromtimestamp(2), 11.0],
  12. [1, datetime.datetime.fromtimestamp(3), 12.0],
  13. [1, datetime.datetime.fromtimestamp(4), 13.0],
  14. [1, datetime.datetime.fromtimestamp(5), 14.0],
  15. [1, datetime.datetime.fromtimestamp(6), 15.0],
  16. [1, datetime.datetime.fromtimestamp(7), 16.0],
  17. [1, datetime.datetime.fromtimestamp(8), 17.0],
  18. [1, datetime.datetime.fromtimestamp(9), 18.0],
  19. [1, datetime.datetime.fromtimestamp(10), 19.0]
  20. ])
  21. streamSource = dataframeToOperator(data, schemaStr='id int, ds1 timestamp, y1 double', op_type='stream')
  22. over = OverCountWindowStreamOp()\
  23. .setTimeCol("ds1")\
  24. .setPrecedingRows(4)\
  25. .setClause("mtable_agg_preceding(ds1,y1) as tensor")
  26. streamPred = ProphetStreamOp()\
  27. .setValueCol("tensor")\
  28. .setPredictNum(1)\
  29. .setPredictionCol("pred")\
  30. .setPredictionDetailCol("pred_detail")
  31. valueOp = LookupVectorInTimeSeriesStreamOp()\
  32. .setTimeSeriesCol("pred")\
  33. .setTimeCol("ds1")\
  34. .setReservedCols(["ds1", "tensor", "pred"])\
  35. .setOutputCol("y_hat")
  36. streamSource\
  37. .link(over)\
  38. .link(streamPred)\
  39. .link(valueOp)\
  40. .print()
  41. StreamOperator.execute()

Java 代码

  1. package com.alibaba.alink.operator.stream.timeseries;
  2. import org.apache.flink.types.Row;
  3. import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
  4. import com.alibaba.alink.operator.batch.timeseries.ProphetTrainBatchOp;
  5. import com.alibaba.alink.operator.stream.StreamOperator;
  6. import com.alibaba.alink.operator.stream.feature.OverCountWindowStreamOp;
  7. import com.alibaba.alink.operator.stream.source.MemSourceStreamOp;
  8. import org.junit.Test;
  9. import java.sql.Timestamp;
  10. import java.util.Arrays;
  11. public class ProphetPredictStreamOpTest {
  12. @Test
  13. public void test() throws Exception {
  14. Row[] rowsData =
  15. new Row[] {
  16. Row.of("1", new Timestamp(117, 11, 1, 0, 0, 0, 0), "9.59076113897809 9.59076113897809"),
  17. Row.of("1", new Timestamp(117, 11, 2, 0, 0, 0, 0), "8.51959031601596 8.51959031601596"),
  18. Row.of("2", new Timestamp(117, 11, 3, 0, 0, 0, 0), "9.59076113897809 8.51959031601596"),
  19. Row.of("1", new Timestamp(117, 11, 4, 0, 0, 0, 0), "8.18367658262066 8.51959031601596"),
  20. Row.of("2", new Timestamp(117, 11, 5, 0, 0, 0, 0), "8.51959031601596 8.51959031601596"),
  21. Row.of("1", new Timestamp(117, 11, 6, 0, 0, 0, 0), "8.07246736935477 8.51959031601596"),
  22. Row.of("2", new Timestamp(117, 11, 7, 0, 0, 0, 0), "8.18367658262066 8.51959031601596"),
  23. Row.of("2", new Timestamp(117, 11, 8, 0, 0, 0, 0), "8.18367658262066 8.51959031601596"),
  24. Row.of("2", new Timestamp(117, 11, 9, 0, 0, 0, 0), "8.18367658262066 8.51959031601596"),
  25. Row.of("2", new Timestamp(117, 11, 10, 0, 0, 0, 0), "8.18367658262066 8.51959031601596"),
  26. Row.of("2", new Timestamp(117, 11, 11, 0, 0, 0, 0), "8.18367658262066 8.51959031601596"),
  27. Row.of("2", new Timestamp(117, 11, 12, 0, 0, 0, 0), "8.18367658262066 8.51959031601596"),
  28. Row.of("2", new Timestamp(117, 11, 13, 0, 0, 0, 0), "8.18367658262066 8.51959031601596"),
  29. };
  30. String[] colNames = new String[] {"id", "ds1", "y1"};
  31. MemSourceStreamOp streamSource = new MemSourceStreamOp(Arrays.asList(rowsData), colNames);
  32. OverCountWindowStreamOp over = new OverCountWindowStreamOp()
  33. .setTimeCol("ds1")
  34. .setPrecedingRows(4)
  35. .setClause("mtable_agg_preceding(ds1,y1) as tensor");
  36. ProphetStreamOp streamPred = new ProphetStreamOp()
  37. .setValueCol("tensor")
  38. .setPredictNum(1)
  39. .setPredictionCol("pred")
  40. .setPredictionDetailCol("pred_detail");
  41. LookupVectorInTimeSeriesStreamOp valueOp = new LookupVectorInTimeSeriesStreamOp()
  42. .setTimeSeriesCol("pred")
  43. .setTimeCol("ds1")
  44. .setReservedCols("ds1", "tensor", "pred")
  45. .setOutputCol("y_hat");
  46. streamSource
  47. .link(over)
  48. .link(streamPred)
  49. .link(valueOp)
  50. .print();
  51. StreamOperator.execute();
  52. }
  53. }

运行结果

| id | data | predict | | —- | —- | —- |

| 1 | {“data”:{“ts”:[“1970-01-01 08:00:00.001”,”1970-01-01 08:00:00.002”,”1970-01-01 08:00:00.003”,”1970-01-01 08:00:00.004”,”1970-01-01 08:00:00.005”,”1970-01-01 08:00:00.006”,”1970-01-01 08:00:00.007”,”1970-01-01 08:00:00.008”,”1970-01-01 08:00:00.009”,”1970-01-01 08:00:00.01”],”val”:[10.0,11.0,12.0,13.0,14.0,15.0,16.0,17.0,18.0,19.0]},”schema”:”ts TIMESTAMP,val DOUBLE”} | {“data”:{“ts”:[“1970-01-01 08:00:00.011”,”1970-01-01 08:00:00.012”,”1970-01-01 08:00:00.013”,”1970-01-01 08:00:00.014”,”1970-01-01 08:00:00.015”,”1970-01-01 08:00:00.016”,”1970-01-01 08:00:00.017”,”1970-01-01 08:00:00.018”,”1970-01-01 08:00:00.019”,”1970-01-01 08:00:00.02”,”1970-01-01 08:00:00.021”,”1970-01-01 08:00:00.022”],”val”:[20.0,21.0,22.0,23.0,24.0,25.0,26.0,27.0,28.0,29.0,30.0,31.0]},”schema”:”ts TIMESTAMP,val DOUBLE”} |

t