Decimals: string to decimal fails if more decimal places than "scale"

[Julian-J-S]

Checks

• [X] I have checked that this issue has not already been reported.
• [X] I have confirmed this bug exists on the latest version of Polars.

Reproducible example

pl.DataFrame({
    'x': ['1', '1.23', '1.2345'],
}).with_columns(
    d=pl.col("x").cast(pl.Decimal(scale=2)),
)

Log output

ComputeError: conversion from `str` to `decimal[*,2]` failed in column 'x' for 1 out of 3 values: ["1.2345"]

Issue description

Converting from str to decimal fails if the str values contain more decimal places than the specified "scale" of the Decimal.

Expected behavior

This works in other libraries (e.g. pyspark) and should also work in polars. Also polars supports casting from f64 "1.2345" to Decimal "1.23"

pyspark:

---

Note: I am opening a lot of Decimal Issues (Decimals are required for many financial calculations where I currently use pyspark). I heard a bigger Decimal update in coming soon? Not sure abount the roadmap but happy to help.

Installed versions

``` 0.20.5 ```

[petrosbar]

Not all libraries handle it the same. For example, in pyarrow this would raise:

import pyarrow as pa
import pyarrow.compute as pc

pc.cast(pa.array(["1.2345"]), pa.decimal128(5, 2))
# ...
# ArrowInvalid: Rescaling Decimal128 value would cause data loss

In any case, it seems that this has been fixed since polars==0.20.6.

import polars as pl

pl.__version__
# '0.20.6'

pl.DataFrame({
    'x': ['1', '1.23', '1.2345'],
}).with_columns(
    d=pl.col("x").cast(pl.Decimal(scale=2)),
)

Output:

  shape: (3, 2)
┌────────┬──────────────┐
│ x      ┆ d            │
│ ---    ┆ ---          │
│ str    ┆ decimal[*,2] │
╞════════╪══════════════╡
│ 1      ┆ 1            │
│ 1.23   ┆ 1.23         │
│ 1.2345 ┆ 1.23         │
└────────┴──────────────┘

[Julian-J-S]

@petrosbar thanks for the info! You are right, it works now.

Maybe we should add a dedicated to_decimal with a allow_data_loss / allow_decimal_truncation parameter that allows to explicitly opt in/out of this functionality 🤔

[matquant14]

Any update on this? I try to cast a string value to a decimal, but it returns NULL values.

I'm on polars 1.81

[cmdlineluser]

@matquant14 It's a bit easier to help if you post an example as runnable code.

It seems commas need to be stripped from strings in order for .cast / .str.to_decimal to work properly.

>>> pl.select(pl.lit("1,000.12").str.to_decimal())
shape: (1, 1)
┌──────────────┐
│ literal      │
│ ---          │
│ decimal[*,2] │
╞══════════════╡
│ null         │
└──────────────┘

>>> pl.select(pl.lit("1,000.12").str.replace_all(r",", "").str.to_decimal())
shape: (1, 1)
┌──────────────┐
│ literal      │
│ ---          │
│ decimal[*,2] │
╞══════════════╡
│ 1000.12      │
└──────────────┘

(It may be useful for the docs to explain this?)

The original example in this issue also seems to work now.

pl.DataFrame({
    'x': ['1', '1.23', '1.2345'],
}).with_columns(
    d=pl.col("x").cast(pl.Decimal(scale=2)),
)

# shape: (3, 2)
# ┌────────┬──────────────┐
# │ x      ┆ d            │
# │ ---    ┆ ---          │
# │ str    ┆ decimal[*,2] │
# ╞════════╪══════════════╡
# │ 1      ┆ 1.00         │
# │ 1.23   ┆ 1.23         │
# │ 1.2345 ┆ 1.23         │
# └────────┴──────────────┘

[matquant14]

ahh, thanks @cmdlineluser. I thought that was being done under the hood, but my oversight. I got it to work now. appreciate the feedback.