HTML styling set_precision should take subset

[mattilyra]

There's a problem in the way .set_precision currently works as it is being applied to all table cells indiscriminately. For instance a dataframe with means and standard errors cross validated across some parameter values the set_precision will easily hide a lot of detail in the results.

The second parameter (2nd index) is not all zeros, they're just values in the range [1e-4, 1e-3]. Also while for the mean columns a precision of 2 is fine, for the standard error 4 might be more appropriate. So having a subset= keyword argument also for .set_precision would be great.

---

code for making the df

import itertools
import pandas as pd
import numpy as np

jobs = itertools.product(['a', 'b', 'c', 'd'], np.arange(1e-4, 1e-3, .0003), range(10))
rows = []
for v, v2, itr in jobs:
    rows.append({'param_1': v, 'param_2': v2, 'iter': itr,
                 'score_1': np.random.randint(0, 100, size=(1,))[0],
                 'score_2': np.random.rand(1, )[0]})
df_multi = pd.DataFrame(rows)
agg = df_multi.groupby(by=['param_1', 'param_2'])[['score_1', 'score_2']].agg(['mean', 'sem'])

[jreback]

yep seems reasonable to me..

pull-requests are welcome!

[TomAugspurger]

This is a good addition, but might be a bit of work. Right now we

                    {% if c.value is number %}
                        {{c.value|round(precision)}}
                    {% else %}
                        {{c.value}}
                    {% endif %}

c is a dict with {'type', 'value', 'class', 'props', 'selector'} that's built inStyler._translate. The best way is probably to add aprecision` field to that.

[mattilyra]

I can take this, although I'll probably require some help as I've never looked at the pandas codebase. The above should get me started at least.

[TomAugspurger]

That'd be great if you want to take it. We're releasing 0.17.1 today sometime, so change probably won't make it. All of this stuff is in pandas/core/style.py.

Right now styler.precision is just an int. That will need to be converted to a dict of {column: int}s (but we should still accept a single int meaning "use this precision for everything". Then in the ._translate step you'll need to add that "precision" field to row_es. That part is the messiest / trickiest code, so let me know if you get stuck.

[mattilyra]

Ok, yeah this certainly won't make it to 0.17.1 in that case. Any thoughts on how the precision would be applied to the index, or should we just leave the index as is?

[TomAugspurger]

I'd say leave index alone, at least to start with. That is all going to get an overhaul before the next release when we add support for styling Indexes.