FancySchmancyTestsplit 0.1.3

a more in-depth testsplit splitting intercategorical

fancy schmancy testsplit

it's like a testsplit, but fancy and also schmancy

---

for reference:

package	fancy	schmancy	testsplit
sklearn.model_selection	👎	👎	👍
fancy schmancy testsplit	👍	👍	👍

a testplit per label category, to ensure that every category is present

---

Examples

Assume the following DataFrame:

df = DataFrame(data= {"Column A":[10, 14, 12, 13, 9, 5, 13, 16, 18, 4, 12],
"Column B": ["Cat1", "Cat1", "Cat2", "Cat2", "Cat2", "Cat2", "Cat2", "Cat2", "Cat2", "Cat2", "Cat2"]})
print(df)

	Column A	Column B
0	10	Cat1
1	14	Cat1
2	12	Cat2
3	13	Cat2
4	9	Cat2
5	5	Cat2
6	13	Cat2
7	16	Cat2
8	18	Cat2
9	4	Cat2
10	12	Cat2

If we assume further that Column B contains the label categories, we'd run the risk of eliminating Cat1 by doing a train test split at 50%.

So, to preserve every existing category, the split will instead be made on every single subset of categories.

As an example for Cat1:

subset = df[df["Column B"] == "Cat1"]
X = subset.drop("Column B", axis= 1)
y = subset["Column B"]
if isinstance(y, Series): y = DataFrame(y)
X_tr, X_te, y_tr, y_te = \
    train_test_split(X, y, test_size = 0.5, random_state = 42)
print(y_tr)

	Column B
0	Cat1

This is done for every unique entry of the given label column, so that a random pick of train and test data is done for every category.