ENH: Add decimal parameter to to_numeric

[cancan101]

Create a class / method for number parsing. This would be a generalization of the code that remove the thousands separators from numbers (see: https://github.com/cancan101/pandas/blob/1703ef44cd6b98e17c785c9120e29bbeefdefd1c/pandas/io/parsers.py#L1502).

Other items include:

1. Alternative decimal point
2. Using parenthesis for negative numbers
3. Stripping currency character

I can submit PR for this. Do people have other suggestions for this?

[jreback]

there is a decimal parameter already, but I not sure in doc/source/io.rst (though in doc string)....maybe add that doc?

[cancan101]

The idea of factoring out this functionality is so that it can be used for parsing numbers in other locations without having to go through the TextParser.

[jreback]

why would you not simply use replace for that, which can do all of what you are talking about? (if its not parser related)

[jreback]

http://pandas.pydata.org/pandas-docs/dev/missing_data.html#string-regular-expression-replacement

[cancan101]

How would you use replace for the parenthesis? Bear in mind that the value then needs to be negated.

[cancan101]

I suppose I could replace the parenthesis with a leading negative sign and then convert that string to an int/ float.

[cancan101]

That being said I was thinking of places where I wanted to be able to take an arbitrary string which might not be in a DataFrame / Series and parse a number.

[jreback]

that would work; replace should infer the return types correctly

[cancan101]

@jreback That addresses using replace for a DataFrame or Series but not in the general sense where we have a string to parse.

[jreback]

out of scope for pandas

[cancan101]

Fair enough. I was just trying to factor out code that could be used both for pandas and for what is out of scope for pandas.

[jreback]

@cancan101 close this then? or do you want to try to do some of this in the actual parse itself? (which depending on the API/scope may of may not be useful). As its a small edge case and if you did have parens/dollar signs in your numbers you can just replace after you read it in, because there is a perf impact in even checking for it on more general parsing

[cancan101]

I was going to / could write this such that there is no performance impact. I would instantiate a parser at the following location: https://github.com/cancan101/pandas/blob/1703ef44cd6b98e17c785c9120e29bbeefdefd1c/pandas/io/parsers.py#L1505 based on parsing options. Theoretically are a few more branches per FILE (not per cell or line).

[jreback]

Well, so you want to parse (value) -> -value and $numeric -> numeric?

this is JUST for PythonParser, right? (if so I am -1 on this); the c-parser is most often the default/used

[cancan101]

What do you mean -1 on this? Like you don't like the change? Or you care less about the performance hit because this code path isn't the default?