Dev tool: python script for text clustering based on local embeddings (#58691)

python script for text clustering based on local embeddings

.tool-versions

@@ -13,3 +13,4 @@ python 3.11.3 system
 rust 1.73.0
 ruby 3.1.3
 pnpm 8.9.2
+python 3.11.3

dev/clustering/README.md

@@ -0,0 +1,47 @@
+# Text Clustering
+
+This directory contains Python code to cluster text data using sentence embeddings and KMeans clustering.
+
+## Overview
+
+The `cluster.py` script takes in a TSV file with a text field, generates sentence embeddings using the SentenceTransformers library, clusters the embeddings with KMeans, and outputs a TSV file with cluster assignments.
+
+The goal is to group similar text snippets together into a predefined number of clusters.
+
+## Usage
+
+Ensure the required packges are installed:
+
+```
+pip install -r requirements.txt
+```
+
+The script accepts the following arguments:
+
+| Argument       | Description                                      | Default    |
+| -------------- | ------------------------------------------------ | ---------- |
+| `--input`      | Path to input TSV file                           | _Required_ |
+| `--text_field` | Name of text field in the tsv file to operate on | "text"     |
+| `--clusters`   | Number of clusters to generate                   | 4          |
+| `--output`     | Path for output TSV file with clusters           | _Optional_ |
+| `--model`      | Sentence transformer model to use                | _Optional_ |
+| `--silent`     | Whether to hide plots                            | False      |
+
+Example
+
+```
+python cluster.py --input data.tsv --text_field chat_message --clusters 5 --output out.tsv
+```
+
+## Output
+
+The output TSV file contains the original data plus a new "cluster" column with the assigned cluster IDs per row.
+
+## Code Overview
+
+**Libraries Used**
+
+- pandas - for loading and manipulating data
+- SentenceTransformers - generating embeddings
+- sklearn - KMeans clustering
+- matplotlib - visualization

dev/clustering/cluster.py

@@ -0,0 +1,58 @@
+import pandas as pd
+from sentence_transformers import SentenceTransformer
+from sklearn.decomposition import PCA
+
+
+import sys
+import argparse
+import os
+os.environ["TOKENIZERS_PARALLELISM"] = "false" #avoids parallelism warnings
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--clusters", default=4, type=int, help="Number of clusters to generate")
+parser.add_argument("--input", type=str, help="Path to input TSV file containing text")
+parser.add_argument("--text_field", default="text", type=str, help="Name of column in TSV containing the text to create embeddings and cluster")
+parser.add_argument("--output", type=str, help="Path to output file")
+parser.add_argument("--model", default="all-MiniLM-L6-v2", type=str, help="Sentence transformer model name")
+parser.add_argument("--quiet",action="store_true", help="hides progress bars and skip displaying plots")
+
+args = parser.parse_args()
+
+# Initialize an embeddings model
+embedding_model = SentenceTransformer(args.model)
+
+# Read the input tsv file into a dataframe
+df = pd.read_csv(args.input, sep='\t')
+
+# Generate embeddings
+embeddings = embedding_model.encode(df[args.text_field], show_progress_bar=not args.quiet,normalize_embeddings=True)
+
+from sklearn.cluster import KMeans
+import matplotlib.pyplot as plt
+
+
+# Perform KMeans clustering
+cluster_model = KMeans(n_clusters=args.clusters, random_state=42, n_init='auto')
+clusters = cluster_model.fit_predict(embeddings)
+
+# update the dataframe with the cluster assignments
+df['cluster'] = cluster_model.labels_
+
+
+if not args.quiet:
+    # plot the clusters
+    # Initialize PCA and reduce dimensionality to 2 components
+    pca = PCA(n_components=2)
+    reduced_embeddings = pca.fit_transform(embeddings)
+    plt.figure(figsize=(10, 6))
+    plt.scatter(reduced_embeddings[:, 0], reduced_embeddings[:, 1], c=clusters, cmap='rainbow')
+    plt.xlabel('PCA Component 1')
+    plt.ylabel('PCA Component 2')
+    plt.title('Clusters of English statements')
+    plt.grid(True)
+    plt.show()
+
+
+# if an output was specified write the dataframe to a tsv file
+if args.output is not None:
+    df.sort_values(by=["cluster"]).to_csv(args.output, sep='\t', index=False)

dev/clustering/requirements.txt

@@ -0,0 +1,4 @@
+pandas~=2.1.3
+scikit-learn~=1.3.2
+matplotlib~=3.8.2
+sentence-transformers~=2.2.2