feat: add Expr|Series log and log10 methods
#1048
Conversation
narwhals/_pandas_like/series.py
Outdated
| def log(self: Self, base: float) -> Self: | ||
| import numpy as np # ignore-banned-import() | ||
|
|
||
| return self._from_native_series(np.log(self._native_series) / np.log(base)) | ||
|
|
||
| def log10(self: Self) -> Self: | ||
| import numpy as np # ignore-banned-import() | ||
|
|
||
| return self._from_native_series(np.log10(self._native_series)) |
There was a problem hiding this comment.
numpy maintains the underlying dtype backend (numpy, nullable numpy or pyarrow)
Edit: I wish it did 🥲
There was a problem hiding this comment.
hmm not sure if this is true when starting from Int64[pyarrow]
In [29]: s
Out[29]:
0 1
1 <NA>
2 3
dtype: int64[pyarrow]
In [30]: np.log(s)
Out[30]:
0 0.000000
1 NaN
2 1.098612
dtype: float64There was a problem hiding this comment.
Polars seems to always go for Float64. This should be doable. I am pushing a change for that.
Let me know if that's too much manipulation for the user
There was a problem hiding this comment.
Polars seems to always go for Float64. This should be doable. I am pushing a change for that.
Let me know if that's too much hidden manipulation for the user.
Edit: example with output
import narwhals as nw
import pandas as pd
import polars as pl
data = [1, None, 3]
@nw.narwhalify
def func(series):
return series.log()
func(pl.Series(data))
Out[10]:
# shape: (3,)
# Series: '' [f64]
# [
# 0.0
# null
# 1.098612
# ]
func(pd.Series(data))
# 0 0.000000
# 1 NaN
# 2 1.098612
# dtype: float64
func(pd.Series(data).astype("int32[pyarrow]"))
# 0 0.0
# 1 <NA>
# 2 1.098612
# dtype: double[pyarrow]There was a problem hiding this comment.
polars preserves float32:
In [14]: pl.Series([1., 2.], dtype=pl.Float32).log10()
Out[14]:
shape: (2,)
Series: '' [f32]
[
0.0
0.30103
]There was a problem hiding this comment.
polars preserves float32:
In [14]: pl.Series([1., 2.], dtype=pl.Float32).log10()
Out[14]:
shape: (2,)
Series: '' [f32]
[
0.0
0.30103
]There was a problem hiding this comment.
it looks like pandas preserves the backend for float - maybe we can do the cast to the appropriate type before calling numpy log?
| values_log_10: [[0,0.3010299956639812,0.6020599913279624]] | ||
|
|
||
| """ | ||
| return self.__class__(lambda plx: self._call(plx).log10()) |
There was a problem hiding this comment.
Although polars calls log(..., base=10) I would imagine that numpy and/or pyarrow dedicated functions may behave differently
pyproject.toml
Outdated
| @@ -115,7 +115,8 @@ filterwarnings = [ | |||
| 'ignore:.*You are using pyarrow version', | |||
| 'ignore:.*but when imported by', | |||
| 'ignore:Distributing .*This may take some time', | |||
| 'ignore:.*The default coalesce behavior' | |||
| 'ignore:.*The default coalesce behavior', | |||
| 'ignore::RuntimeWarning' | |||
There was a problem hiding this comment.
this seems to broad to turn off globally, can we just catch it in the code?
narwhals/_pandas_like/series.py
Outdated
| def log(self: Self, base: float) -> Self: | ||
| import numpy as np # ignore-banned-import() | ||
|
|
||
| return self._from_native_series(np.log(self._native_series) / np.log(base)) | ||
|
|
||
| def log10(self: Self) -> Self: | ||
| import numpy as np # ignore-banned-import() | ||
|
|
||
| return self._from_native_series(np.log10(self._native_series)) |
There was a problem hiding this comment.
hmm not sure if this is true when starting from Int64[pyarrow]
In [29]: s
Out[29]:
0 1
1 <NA>
2 3
dtype: int64[pyarrow]
In [30]: np.log(s)
Out[30]:
0 0.000000
1 NaN
2 1.098612
dtype: float64
FBruzzesi
changed the title
Expr & Series log & log10 methodsExpr|Series log and log10 methods
What type of PR is this? (check all applicable)
Related issues
Used in plotly
Checklist
If you have comments or can explain your changes, please do so below.