SemHash

SemHash is a lightweight and flexible tool for deduplicating datasets, filtering outliers, and finding representative samples using semantic similarity. It combines fast embedding generation from Model2Vec with efficient ANN-based similarity search through Vicinity.

SemHash supports both single-dataset deduplication & filtering (e.g., cleaning up a train set by removing duplicates and outliers) and multi-dataset deduplication & filtering (e.g., ensuring no overlap between a test set and a train set). It works with simple datasets, such as text lists, and more complex ones, like multi-column QA datasets. Additionally, it includes functions to inspect deduplication results, making it easier to understand and refine your data cleaning process.

​

Quick Start

Install SemHash with the following command:

pip install semhash

Then, you can use it like this to deduplicate, filter outliers, and select representative samples (this example assumes you have the datasets library installed):

from datasets import load_dataset
from semhash import SemHash

# Load a dataset to deduplicate
texts = load_dataset("ag_news", split="train")["text"]

# Initialize a SemHash instance
semhash = SemHash.from_records(records=texts)

# Deduplicate the texts
deduplicated_texts = semhash.self_deduplicate().selected

# Filter outliers
filtered_texts = semhash.self_filter_outliers().selected

# Find representative texts
representative_texts = semhash.self_find_representative().selected

For advanced usage, check out the other documentation pages.