analyze.ShowClassSplits returns empty data frame

[tarbaig]

Description:

Using Analyze.ShowClassSplits on an object returned by the importer returns an empty dataframe.

Expected:

The returned frame should contain the proportion of the samples labeled train, test and validation by the splitter.

How to reproduce:

Assume "https://github.com/pylabelalpha/notebook/blob/main/BCCD_coco.zip?raw=true" is in the subdirectory 'data'

from pylabel import importer
from pathlib import Path
data_path = Path('data')
annotation_path = data_path / 'bccd' / 'BCCD_Dataset.json'
image_path = ""
dataset = importer.ImportCoco(path=str(annotation_path), path_to_images=str(image_path),name="BCCD_COCO")
dataset.splitter.GroupShuffleSplit(train_pct=.6,test_pct=.2,val_pct=.2)
assert not dataset.analyze.ShowClassSplits().empty

Suggested Fix

Looking into the code of the Analyze class it seems that the way the dataframes are created there does not work properly and the intermediat dataframes are already empty.

This can be fixed by creating those frames differntly e.g.

pylabel/pylabel/analyze.py

Lines 137 to 143 in 6e815ff

	df_value_counts = pd.DataFrame(
	ds.df["cat_name"].value_counts(normalize=normalize), columns=["cat_name"]
	)
	df_value_counts.index.name = "cat_name"
	df_value_counts.columns = ["all"]

should be

df_value_counts = ds.df["cat_name"].value_counts(normalize=normalize).to_frame(name='all')

I can prepare a pull request if you like