Error in null: Ambiguous truth value of a Series about cleanlab HOT 4 CLOSED

To be clear, this is not an error that results in a crashing program, but an unsupported input type causes the null issue check to fail.

from cleanlab.

Thank you for the report @MengChenC

You're trying to demonstrate this kind of workflow (your pred_probs will not be passed to the null issue-check method, so we'll omit that in the example, we can also skip the labels for demonstration purposes).

### CREATE DATASET

import numpy as np
import pandas as pd

dataset = {
    "features": np.array([[1, 2, np.nan], [4, np.nan, 6], [np.nan, 8, 9]]),
}

feature_columns = [f"x{i+1}" for i in range(len(dataset["features"][0]))]
df = pd.DataFrame(dataset["features"], columns=feature_columns),


print(df)
#      x1   x2   x3
#  0  1.0  2.0  NaN
#  1  4.0  NaN  6.0
#  2  NaN  8.0  9.0


### RUN DATALAB

from cleanlab import Datalab
lab = Datalab(data=df)

lab.find_issues(features=df, issue_types={"null": {}})
# Finding null issues ...
# Error in null: The truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all().
# Failed to check for these issue types: [NullIssueManager]
# 
# Audit complete. 0 issues found in the dataset.

You're passing in features as a DataFrame instead of an array, which is the recommended approach and primary expected input format to Datalab.find_issues. ( See relevant API reference)

The best approach is to run

lab.find_issues(features=df.to_numpy(), issue_types={"null": {}})
# Finding null issues ...
# 
# Audit complete. 0 issues found in the dataset.

How this will be addressed.

We can apply the changes you suggest about the null_tracker to explicitly look at every element in the null_tracker.
Thank you for the suggestion! 👍

But we'll turn internally turn features: pd.DataFrame into a numpy array while performing the null issue check. We'll still make the np.any check more explicit.
See the ongoing PR #1045.

from cleanlab.

Hey @MengChenC

The changes you suggested have made its way to a new patch release.

Try running this on cleanlab v2.6.2.

# Upgrade to 
pip install -U cleanlab

from cleanlab.

Hi @elisno

Thank you for making the update nimbly.

I can confirm this is now working for the same dataset 🎉

from cleanlab.