src/streamlit_app.py

import matplotlib.pyplot as plt
import nltk
import numpy as np
import pandas as pd
import seaborn as sns
import streamlit as st
from nltk.tokenize import word_tokenize
import os

# Use a writable temporary directory
nltk_data_dir = "/tmp/nltk_data"

# Create the directory if it doesn't exist
os.makedirs(nltk_data_dir, exist_ok=True)

# Add it to NLTK's data path
nltk.data.path.append(nltk_data_dir)

# Download 'punkt' only if not already downloaded
try:
    nltk.data.find("tokenizers/punkt")
except LookupError:
    nltk.download("punkt", download_dir=nltk_data_dir)
    nltk.download("punkt_tab", download_dir=nltk_data_dir)

st.title("📊 Bayesian Token Co-occurrence Simulator")

# User input
user_input = st.text_area(
    "✍️ Enter your training sentences (one per line):",
    """
fido loves the red ball
timmy and fido go to the park
fido and timmy love to play
the red ball is timmy's favorite toy
""",
)

sentences = user_input.strip().split("\n")
tokenized = [word_tokenize(s.lower()) for s in sentences if s.strip()]
vocab = sorted(set(word for sentence in tokenized for word in sentence))
token2idx = {word: i for i, word in enumerate(vocab)}
idx2token = {i: word for word, i in token2idx.items()}

# Co-occurrence matrix
window_size = 2
matrix = np.zeros((len(vocab), len(vocab)))

for sentence in tokenized:
    for i, word in enumerate(sentence):
        for j in range(
            max(0, i - window_size), min(len(sentence), i + window_size + 1)
        ):
            if i != j:
                matrix[token2idx[word]][token2idx[sentence[j]]] += 1

alpha = st.slider("🔧 Set Bayesian Prior (α smoothing)", 0.0, 2.0, 0.1)
posterior = matrix + alpha

df = pd.DataFrame(posterior, index=vocab, columns=vocab)
st.subheader("📈 Co-occurrence Heatmap")
fig, ax = plt.subplots(figsize=(10, 8))
sns.heatmap(df, annot=True, cmap="Blues", fmt=".1f", ax=ax)
st.pyplot(fig)

# Next-token prediction
selected_word = st.selectbox("🔮 Predict next token after:", vocab)
row = posterior[token2idx[selected_word]]
probs = row / row.sum()
prediction = np.random.choice(vocab, p=probs)

st.markdown(f"**Predicted next token:** `{prediction}`")