zhengruifeng opened a new pull request, #38148:
URL: https://github.com/apache/spark/pull/38148
### What changes were proposed in this pull request?
add a dedicated expression for `product`:
1. for integral inputs, directly use `LongType` to avoid the rounding error:
2. when ignoreNA is true, short circuit when meet a `zero`;
3. when ignoreNA is false, short circuit when meet a `zero` or `null`;
### Why are the changes needed?
1. existing computation logic is too complex in the PySpark side, with a
dedicated expression, we can simplify the PySpark side and apply it in more
cases.
2. existing computation of `product` is likely to introduce rounding error
for integral inputs, for example `55108 x 55108 x 55108 x 55108` in the
following case:
before:
```
In [14]: df = pd.DataFrame({"a": [55108, 55108, 55108, 55108], "b":
[55108.0, 55108.0, 55108.0, 55108.0], "c": [1, 2, 3, 4]})
In [15]: df.a.prod()
Out[15]: 9222710978872688896
In [16]: type(df.a.prod())
Out[16]: numpy.int64
In [17]: df.b.prod()
Out[17]: 9.222710978872689e+18
In [18]: type(df.b.prod())
Out[18]: numpy.float64
In [19]:
In [19]: psdf = ps.from_pandas(df)
In [20]: psdf.a.prod()
Out[20]: 9222710978872658944
In [21]: type(psdf.a.prod())
Out[21]: int
In [22]: psdf.b.prod()
Out[22]: 9.222710978872659e+18
In [23]: type(psdf.b.prod())
Out[23]: float
In [24]: df.a.prod() - psdf.a.prod()
Out[24]: 29952
```
after:
```
In [1]: import pyspark.pandas as ps
In [2]: import pandas as pd
In [3]: df = pd.DataFrame({"a": [55108, 55108, 55108, 55108], "b": [55108.0,
55108.0, 55108.0, 55108.0], "c": [1, 2, 3, 4]})
In [4]: df.a.prod()
Out[4]: 9222710978872688896
In [5]: psdf = ps.from_pandas(df)
In [6]: psdf.a.prod()
Out[6]: 9222710978872688896
In [7]: df.a.prod() - psdf.a.prod()
Out[7]: 0
```
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
existing UT
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
