python - 有没有办法在 BigQuery 中对时间序列数据重新采样？

Question

pandas DataFrame 有如下所示的 resample 方法，我想实现的是通过在 BigQuery 中查询的等效方法。

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 分钟的频率对数据重新采样。

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。
但是有没有办法在 bigquery 中进行 QUERY 重采样？

编辑

pandas DataFrame resample详解

                        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'

前 5 行(22:00 到 22:04)以 1 分钟的频率重新采样为 1 行(22:00)，
接下来的 5 行(22:05 到 22:09)变成了 (22:05)。
重采样方法分别为first、max、min、last和sum。

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

有关更多详细信息，请参阅 pandas Document

Answer 1

试试下面

#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MOD(step, 5) 中的5 和TIMESTAMP_DIFF 中的MINUTE 来控制采样间隔。

你可以使用下面的虚拟数据来玩这个

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 
)

Below version implements "panda's resample" (per logic in updated question)

#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

或者进一步简化的版本，只有一个 GROUP BY 和窗口函数。还假设您的数据晚于 '2000-01-01 00:00:00' - 否则您可以相应地调整

#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