import solara import numpy as np import pandas as pd from sentence_transformers import SentenceTransformer from huggingface_hub import snapshot_download from umap import UMAP from annoy import AnnoyIndex from cluestar import plot_text news = pd.read_csv('https://raw.githubusercontent.com/alonsosilvaallende/fake-and-real-news-titles/main/example.csv') texts = list(news["title"].values) texts = [str(text) for text in texts if str(text) != 'nan'] sentences = ["This is an example sentence", "Each sentence is converted"] model_path = snapshot_download( repo_id="TaylorAI/gte-tiny", allow_patterns=["*.json", "pytorch_model.bin"] ) embedder2 = SentenceTransformer(model_path) embeddings2 = [embedder2.encode(str(texts[i])) for i in range(500)] reducer = UMAP() X2 = reducer.fit_transform(embeddings2) f = len(embeddings2[0]) t = AnnoyIndex(f, 'angular') for i, embedded_text in enumerate(embeddings2): t.add_item(i, embedded_text) t.build(1000) query = solara.reactive("What did Nancy Pelosi said about Obamacare?") @solara.component def Page(): with solara.Column(margin=10): solara.Markdown("#Embeddings") solara.InputText("Enter some query:", query, continuous_update=True) if query.value != "": embedded_query = embedder2.encode(query.value) idx, distances = t.get_nns_by_vector(embedded_query, 10, include_distances=True) df_neighbors = pd.DataFrame() df_neighbors["neighbors"]=[texts[i] for i in idx] df_neighbors["distances"] = distances x = reducer.transform([embedded_query]) color_array = ["texts" if i not in idx else "neighbors" for i in range(len(texts[:500]))]+["query"] solara.AltairChart(plot_text(np.vstack((X2,x)), texts[:500]+[query.value], color_array=color_array).configure_range( category=['#0000ff', '#ff0000', '#a0aab4'] )) solara.DataFrame(df_neighbors, items_per_page=10) solara.Markdown("Dataset: 'Fake and real news' from [kaggle](https://www.kaggle.com/datasets/clmentbisaillon/fake-and-real-news-dataset)") else: color_array = ["texts" for _ in range(500)] solara.AltairChart(plot_text(X2, texts[:500], color_array=color_array))