有没有办法在BigQuery中重新采样时间序列数据?

时间:2017-02-17 14:17:22

标签: python pandas google-bigquery

pandas DataFrame具有如下所示的resample方法,我想通过查询BigQuery来实现等效方法。

pandas中的示例方法
现在我有这样的数据。假设相同的数据存储在bigquery中。

In [2]: df.head()
Out[2]: 
                        Open     High      Low    Close  Volume
Gmt time                                                       
2016-01-03 22:00:00  1.08730  1.08730  1.08702  1.08714    8.62
2016-01-03 22:01:00  1.08718  1.08718  1.08713  1.08713    3.75
2016-01-03 22:02:00  1.08714  1.08721  1.08714  1.08720    4.60
2016-01-03 22:03:00  1.08717  1.08721  1.08714  1.08721    7.57
2016-01-03 22:04:00  1.08718  1.08718  1.08711  1.08711    5.52

然后使用DataFrame以5分钟的频率重新采样数据。

In [3]: ohlcv = {
      :         'Open':'first',
      :         'High':'max',
      :         'Low':'min',
      :         'Close':'last',
      :         'Volume':'sum'
      :         }
      : df = df.resample('5T').apply(ohlcv)  # 5 minutes frequency
      : df = df[['Open', 'High', 'Low', 'Close', 'Volume']]  # reorder columns
      : df.head()
      : 
      : 
Out[3]: 
                        Open     High      Low    Close  Volume
Gmt time                                                       
2016-01-03 22:00:00  1.08730  1.08730  1.08702  1.08711   30.06
2016-01-03 22:05:00  1.08711  1.08727  1.08709  1.08709  190.63
2016-01-03 22:10:00  1.08708  1.08709  1.08662  1.08666  168.79
2016-01-03 22:15:00  1.08666  1.08674  1.08666  1.08667  223.83
2016-01-03 22:20:00  1.08667  1.08713  1.08666  1.08667  170.17

这可以在从bigquery获取1分钟频率数据后完成 但是有没有办法QUERY重新采样bigquery?

修改

pandas DataFrame重新采样的详细说明。

                        Open     High      Low    Close  Volume
Gmt time                                                       
# 1 minute frequency data stored in bigquery
2016-01-03 22:00:00  1.08730  1.08730  1.08702  1.08714    8.62
2016-01-03 22:01:00  1.08718  1.08718  1.08713  1.08713    3.75
2016-01-03 22:02:00  1.08714  1.08721  1.08714  1.08720    4.60
2016-01-03 22:03:00  1.08717  1.08721  1.08714  1.08721    7.57
2016-01-03 22:04:00  1.08718  1.08718  1.08711  1.08711    5.52

2016-01-03 22:05:00  1.08711  1.08714  1.08711  1.08711   27.47
2016-01-03 22:06:00  1.08717  1.08720  1.08711  1.08711   21.58
2016-01-03 22:07:00  1.08713  1.08718  1.08712  1.08715   28.12
2016-01-03 22:08:00  1.08714  1.08723  1.08712  1.08718   49.74
2016-01-03 22:09:00  1.08722  1.08727  1.08709  1.08709   63.72

# expected query result
# above will be resampled into below..
2016-01-03 22:00:00  1.08730  1.08730  1.08702  1.08711   30.06
2016-01-03 22:05:00  1.08711  1.08727  1.08709  1.08709  190.63
# method to resample 'first'  'max'    'min'    'last'    'sum'

1分钟频率的前5行(22:00至22:04)重新采样为1行(22:00),
接下来的5行(22:05到22:09)进入(22:05) 重采样方法分别为firstmaxminlastsum

first计算组的第一个值(这里表示5行)
max计算最大值,
min计算最小值,
last计算最后一个值,
sum计算组

中列的总和

有关详细信息,请参阅pandas Document

1 个答案:

答案 0 :(得分:4)

尝试以下

  
#standardSQL
SELECT * EXCEPT(step) 
FROM (
  SELECT *, TIMESTAMP_DIFF(TIMESTAMP(ts), 
              TIMESTAMP(MIN(ts) OVER(ORDER BY ts)), MINUTE) AS step
  FROM yourTable
)
WHERE MOD(step, 5) = 0
-- ORDER BY ts   

可以通过更改5中的MOD(step, 5)TIMESTAMP_DIFF中的MINUTE

来控制采样间隔

您可以使用以下虚拟数据

来玩这个
WITH yourTable AS (
  SELECT '2016-01-03 22:00:00' AS ts, 1.08730 AS Open, 1.08730 AS High, 1.08702 AS Low, 1.08714 AS Close, 8.62 AS Volume UNION ALL
  SELECT '2016-01-03 22:01:00', 1.08718, 1.08718, 1.08713, 1.08713, 3.75 UNION ALL
  SELECT '2016-01-03 22:02:00', 1.08714, 1.08721, 1.08714, 1.08720, 4.60 UNION ALL
  SELECT '2016-01-03 22:03:00', 1.08717, 1.08721, 1.08714, 1.08721, 7.57 UNION ALL
  SELECT '2016-01-03 22:04:00', 1.08718, 1.08718, 1.08711, 1.08711, 5.52 UNION ALL
  SELECT '2016-01-03 22:05:00', 1.08718, 1.08718, 1.08713, 1.08713, 3.75 UNION ALL
  SELECT '2016-01-03 22:06:00', 1.08714, 1.08721, 1.08714, 1.08720, 4.60 UNION ALL
  SELECT '2016-01-03 22:07:00', 1.08717, 1.08721, 1.08714, 1.08721, 7.57 UNION ALL
  SELECT '2016-01-03 22:08:00', 1.08718, 1.08718, 1.08711, 1.08711, 5.52 UNION ALL
  SELECT '2016-01-03 22:09:00', 1.08718, 1.08718, 1.08713, 1.08713, 3.75 UNION ALL
  SELECT '2016-01-03 22:10:00', 1.08714, 1.08721, 1.08714, 1.08720, 4.60 UNION ALL
  SELECT '2016-01-03 22:11:00', 1.08717, 1.08721, 1.08714, 1.08721, 7.57 UNION ALL
  SELECT '2016-01-03 22:12:00', 1.08718, 1.08718, 1.08711, 1.08711, 5.52 
)
  

以下版本实施" panda' resample" (根据更新的问题中的逻辑)

#standardSQL
SELECT 
  MIN(ts) AS ts,
  ARRAY_AGG(Open ORDER BY ts)[OFFSET (0)] AS Open,
  MAX(High) AS High,
  MIN(Low) AS Low,
  ARRAY_AGG(Close ORDER BY ts DESC)[OFFSET (0)] AS Close,
  SUM(Volume) AS Volume
FROM (
  SELECT *, DIV(TIMESTAMP_DIFF(TIMESTAMP(ts), 
              TIMESTAMP(MIN(ts) OVER(ORDER BY ts)), MINUTE), 5) AS grp
  FROM yourTable
)
GROUP BY grp
-- ORDER BY ts

或者进一步简化版本,只有一个GROUP BY和窗口函数。同时假设您的数据晚于2000-01-01 00:00:00' - 否则你可以相应调整

#standardSQL
SELECT 
  MIN(ts) AS ts,
  ARRAY_AGG(Open ORDER BY ts)[OFFSET (0)] AS Open,
  MAX(High) AS High,
  MIN(Low) AS Low,
  ARRAY_AGG(Close ORDER BY ts DESC)[OFFSET (0)] AS Close,
  SUM(Volume) AS Volume
FROM yourTable
GROUP BY DIV(TIMESTAMP_DIFF(TIMESTAMP(ts), 
             TIMESTAMP('2000-01-01 00:00:00'), MINUTE), 5)
-- ORDER BY ts