Question / Possibly a Feature request: comparison machinery for unit tests

[rs2]

When it comes to writing unit tests that return a DataFrame or a TimeSeries there's a clear need to compare the output to the expected result. The comparison can be strict, or relaxed in terms of the order of columns and/or order of rows, and could also be forgiving for floats.

I'm thinking of an approach implemented in https://pythonhosted.org/toytable/toytable.table.html#table-literals, i.e. defining a table as

>>> pokedex = table_literal("""
...     | Pokemon (str) | Level (int) | Owner (str) | Type (str) |
...     | Pikchu        | 12          | Ash Ketchum | Electric   |
...     | Bulbasaur     | 16          | Ash Ketchum | Grass      |
...     | Charmander    | 19          | Ash Ketchum | Fire       |
...     | Raichu        | 23          | Lt. Surge   | Electric   |
...     """)

and subsequently invoking self.assertTableEqual or self.assertTableContentsEqual.

Does anything similar exist in pandas (I could not find), and can the above approach be upvoted please?

[jreback]

well we have an entire module for this: https://github.com/pydata/pandas/blob/master/pandas/util/testing.py

not to mention that the entire test suite uses things like this.

try

from pandas .util import testing as tm
tm.assert_frame_equal

[rs2]

How would you build & visualize the expected DataFrame? The table_literal approach seems to be appropriate.

[jorisvandenbossche]

@rs2 Just the same as you otherwise build a small dataframe, eg pd.DataFrame(..) with eg a dictionary.
If you want to ahve something visual like the above, you can always pass such a string to read_csv or read_fwf

See also #9895 for proposal to make the testing functions more 'officially' public