Skip to main content

🎉 We released Spotlight 1.6.0 check it out

Version: Next

Find data slices with Sliceguard

We use Sliceguard to identify data segments where our machine learning model performs anomalously (data slices). We interactively explore these data slices to find model failure modes and problematic data segments.

Use Chrome to run Spotlight in Colab. Due to Colab restrictions (e.g. no websocket support), the performance is limited. Run the notebook locally for the full Spotlight experience.

Open In Colab
  • categories contains contain the names of the features and metadata columns that are to be analyzed
  • df['label'] contains the label for each data sample
  • df['prediction'] contains the prediction for each data sample
  • df['embedding'] contains the embeddings for each data sample (optional)


Imports and play as copy-n-paste functions

# Install dependencies
#@title Install required packages with PIP

!pip install renumics-spotlight sliceguard datasets cleanvision
Play as copy-n-paste functions
#@title Play as copy-n-paste snippet

from sklearn.metrics import accuracy_score
import pandas as pd
import datasets
from renumics.spotlight import Image
from sliceguard import SliceGuard
from cleanvision.imagelab import Imagelab

def find_data_slices(df, categories, category_types={}, spotlight_dtype={}, embedding_name='embedding', label_name='label', prediction_name='prediction'):
sg = SliceGuard()
df_slices = sg.find_issues(
precomputed_embeddings = {'embedding': df[embedding_name].to_numpy()},

return df_slices

Step-by-step example on CIFAR-100

Load CIFAR-100 from Huggingface hub and convert it to Pandas dataframe

dataset = datasets.load_dataset("renumics/cifar100-enriched", split="test")
df = dataset.to_pandas()

Enrich dataset with metadata using Cleanvision

def cv_issues_cleanvision(df, image_name='image'):

image_paths = df['image'].to_list()
imagelab = Imagelab(filepaths=image_paths)


return df_cv

df_cv = cv_issues_cleanvision(df)
df = pd.concat([df, df_cv], axis=1)

Identify and explore data slices with Sliceguard

categories=['dark_score', 'low_information_score', 'light_score', 'blurry_score', 'fine_label']
prediction = 'fine_label_prediction'
label = 'fine_label'
category_types={'fine_label': 'nominal'}
spotlight_dtype={"image": Image}

find_data_slices(df, categories, category_types=category_types, spotlight_dtype=spotlight_dtype, embedding_name='embedding', label_name=label, prediction_name=prediction)