NotSoBad/HFSpaceStreamLitHeatMap

1import streamlit as st2import nltk3from transformers import pipeline4from sentence_transformers import SentenceTransformer5from scipy.spatial.distance import cosine6import numpy as np7import seaborn as sns8import matplotlib.pyplot as plt9from sklearn.cluster import KMeans10import tensorflow as tf11import tensorflow_hub as hub12 13 14def cluster_examples(messages, embed, nc=3):15    km = KMeans(16    n_clusters=nc, init='random',17    n_init=10, max_iter=300, 18    tol=1e-04, random_state=019    )20    km = km.fit_predict(embed)21    for n in range(nc):22        idxs = [i for i in range(len(km)) if km[i] == n]23        ms = [messages[i] for i in idxs]24        st.markdown ("CLUSTER : %d"%n)25        for m in ms:26            st.markdown (m)27 28 29def plot_heatmap(labels, heatmap, rotation=90):30  sns.set(font_scale=1.2)31  fig, ax = plt.subplots()32  g = sns.heatmap(33      heatmap,34      xticklabels=labels,35      yticklabels=labels,36      vmin=-1,37      vmax=1,38      cmap="coolwarm")39  g.set_xticklabels(labels, rotation=rotation)40  g.set_title("Textual Similarity")41  st.pyplot(fig)42  43# Streamlit text boxes44text = st.text_area('Enter sentences:', value="Behavior right this is a kind of Heisenberg uncertainty principle situation if I told you, then you behave differently. What would be the impressive thing is you have talked about winning a nobel prize in a system winning a nobel prize. Adjusting it and then making your own.  That is when I fell in love with computers.  I realized that they were a very magical device.  Can go to sleep come back the next day and it is solved.  You know that feels magical to me.")45 46nc = st.slider('Select a number of clusters:', min_value=1, max_value=15, value=3)47 48model_type = st.radio("Choose model:", ('Sentence Transformer', 'Universal Sentence Encoder'), index=0)49 50# Model setup51if model_type == "Sentence Transformer":52    model = SentenceTransformer('paraphrase-distilroberta-base-v1')53elif model_type == "Universal Sentence Encoder":54    model_url = "https://tfhub.dev/google/universal-sentence-encoder-large/5"55    model = hub.load(model_url)56 57nltk.download('punkt')58 59# Run model60if text:61    sentences = nltk.tokenize.sent_tokenize(text)62    if model_type == "Sentence Transformer":63        embed = model.encode(sentences)64    elif model_type == "Universal Sentence Encoder":65        embed = model(sentences).numpy()66    sim = np.zeros([len(embed), len(embed)])67    for i,em in enumerate(embed):68        for j,ea in enumerate(embed):69            sim[i][j] = 1.0-cosine(em,ea)70    st.subheader("Similarity Heatmap")71    plot_heatmap(sentences, sim)72    st.subheader("Results from K-Means Clustering")73    cluster_examples(sentences, embed, nc)74