You could use field_transform to accomplish most of this, by passing String.replace:

String.replace("-123,45", ~r/^(-?)(\d+),(\d+)$/, "\\1\\2.\\3")

(adjust further if you’re expecting scientific notation too)

Can you give an example of a few CSV rows? F.ex. do they look like this?

ABC,123,4,58,DEF

And you expect ["ABC", "123", "4.58", "DEF"] but get ["ABC", "123", "4", "58", "DEF"] instead? Is that the problem? Sounds like you basically have an invalid CSV when we get down to it. Don’t think that even field_transform can help you in this case.

You can use the xsv tool to extract the “defective” columns and reformat the numbers to be dot-separated and then you can re-merge (or replace) the data back – I’ve done something very similar in the past, successfully. From then on you can just use any normal CSV parser.

I did some digging. Polars actually has a relevant option to polars.read_csv() called decimal_comma that Explorer does not expose:

decimal_comma
Parse floats using a comma as the decimal separator instead of a period.

We could certainly expose it. However even if it were exposed, you can’t use it for your example because your CSV also has a , as the separator:

polars.exceptions.InvalidOperationError: 'decimal_comma' argument cannot be combined with ',' separator

So it seems Polars, and therefore Explorer, requires a certain subset of CSV to parse directly into a float like you’re hoping to do.

However @dimitarvp is right to call out mutate. General rule with Explorer: it’s usually fastest to get Rust to do the work if you can. So I suggest the following:

require Explorer.DataFrame, as: DataFrame

csv_path
|> DataFrame.from_csv!()
|> DataFrame.mutate(for col <- across() do
  {col.name, col |> replace(",", ".") |> cast(:f64)}
end)

With this approach, you load what happens to be in the CSV into Rust, then let Rust do the string manipulation.