[BUG] Python random split sum to 1 error

[gramhagen]

Description

Python random split with multiple ratio options fails even when ratios sum to 1

In which platform does it happen?

Python all OSs

How do we replicate the issue?

from reco_utils.dataset.movielens import load_pandas_df
from reco_utils.dataset.python_splitters import python_random_split
df = load_pandas_df()
x, y, z = python_random_split(df, ratio=[.7, .2, .1])

yields

Traceback (most recent call last):
  File "C:\Users\scgraham\AppData\Local\Continuum\anaconda3\envs\reco_base\lib\site-packages\IPython\core\interactiveshell.py", line 3325, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-5-49ca4e7e6410>", line 1, in <module>
    x, y, z = python_random_split(df, ratio=[.7, .2, .1])
  File "C:\Users\scgraham\repos\Recommenders\reco_utils\dataset\python_splitters.py", line 38, in python_random_split
    splits = split_pandas_data_with_ratios(data, ratio, shuffle=True, seed=seed)
  File "C:\Users\scgraham\repos\Recommenders\reco_utils\dataset\split_utils.py", line 155, in split_pandas_data_with_ratios
    raise ValueError("The ratios have to sum to 1")
ValueError: The ratios have to sum to 1

this occurs because sum([.7, .2, .1]) == .999999.... and there is a check in split_utils.py:39 which fails and renormalizes, subsequently split_utils:154 has the same check which also fails which leads to this error state

Expected behavior (i.e. solution)

data is split into train, test, and validate sets with expected fraction.

Other Comments

[yueguoguo]

Good catch. Will fix it.

[yueguoguo]

Resolved in #874 and validated in #876