Error with dataframes containing non-string column names

[MarinManuel]

I've noticed DABEST does not let me work with dataframes where column names are not strings.

example:

import dabest
import pandas as pd
import numpy as np
from itertools import combinations
from pprint import pprint

print(dabest.__version__)

columns = [1, 2.0]
# create a test database
N = 100
df = pd.DataFrame(np.vstack([np.random.normal(loc=i, size=(N,)) for i in range(len(columns))]).T, columns=columns)

db = dabest.load(data=df, idx=columns)
print(db.mean_diff)

results in an Exception:

0.3.0
Traceback (most recent call last):
 File "<ipython-input-10-19e1ff62e09f>", line 14, in <module>
   db = dabest.load(data=df, idx=columns)
 File "/.../anaconda3/lib/python3.7/site-packages/dabest/_api.py", line 65, in load
  return Dabest(data, idx, x, y, paired, id_col, ci, resamples, random_seed)
 File "/.../anaconda3/lib/python3.7/site-packages/dabest/_classes.py", line 67, in __init__
   raise ValueError(err)
ValueError: There seems to be a problem with the idx you

As far as I understand, any hashable can be used as a column name (an Index really). I'm not sure what use case there is for exotic types, but I often have columns that are named with int or float.

I'm going to propose a fix, but I'm not entirely sure the solution is robust. I'm would test for Hashable instead of str, but some iterables are hashables (frozensets, tuples, range) and I don't quite know how you'd deal with that.

PLUS there is an independent but related bug in pandas that prevents the fix from working. See here. The bug was fixed in v.1.0.0 and DABEST would need to require that version to work properly.

[MarinManuel]

disregard the part about change the req for pandas. I don't know why my pandas version was stuck at 0.25.x in my test environment. I just realized that DABEST requires pandas v.1.1 now.

[MarinManuel]

Further tests showed that my initial fix was not working. I suggest testing for str, int and float idx. I'm sure there might be cases where that's not enough, but that probably covers most use cases?

[Jacobluke-]

The bug is solved.

[Jacobluke-]

Solved in PR #174.