app.py

import json
import re
from collections import defaultdict
from typing import Dict

import streamlit as st
import pandas as pd
import numpy as np
import torch
from transformers import AutoTokenizer, AutoModel
import umap
import plotly.express as px
import textwrap

st.title('Bible analysis with BERT')

@st.cache
def load_verses() -> Dict[str, str]:
    verses = {}
    count = 0
    with open('esv.txt', 'r', encoding='utf8') as f:
        lines = f.readlines()
        for line in lines:
            try:
                citation, raw_sentence = line.strip().split('\t')
                verses[citation] = raw_sentence
            except ValueError:
                count +=1
    print(count)
    return verses


@st.cache
def load_tags():
    index = defaultdict(list)
    with open("esv_tags.txt", encoding='utf8') as f:
        lines = f.readlines()
        for line in lines:
            verse, strongs = line.split("\t", maxsplit=1)
            tokens = strongs.strip().split("\t")
            for t in tokens:
                if "=" in t:
                    words, strongs = t.split("=")
                    words = [(verse, int(x)) for x in words.split("+")]
                    strongs = [x[1:-1] for x in strongs.split("+")]
                    for s in strongs:
                        index[s].extend(words)
    return index


@st.cache
def get_strong_defs():
    with open("strongs_defs.json", encoding='utf8') as f:
        return json.load(f)


def get_word_idx(sent: str, word: str):
    l = re.split('([ .,!?:;""()\'-])', sent)
    l = [x for x in l if x != " " and x != ""]
    return l.index(word)


@st.cache
def get_embedding(sent, word, layers=None):
    """Get a word vector by first tokenizing the input sentence, getting all token idxs
        that make up the word of interest, and then `get_hidden_states`."""
    layers = [-4, -3, -2, -1] if layers is None else layers
    tokenizer, model = get_models()
    encoded = tokenizer.encode_plus(sent, return_tensors="pt")

    idx = get_word_idx(sent, word)
    # get all token idxs that belong to the word of interest
    token_ids_word = np.where(np.array(encoded.word_ids()) == idx)

    with torch.no_grad():
        output = model(**encoded)

    # Get all hidden states
    states = output.hidden_states
    # Stack and sum all requested layers
    output = torch.stack([states[i] for i in layers]).sum(0).squeeze()
    # Only select the tokens that constitute the requested word
    word_tokens_output = output[token_ids_word]

    return word_tokens_output.mean(dim=0).numpy()



verses = load_verses()
strongs_tags = load_tags()
strongs_defs = get_strong_defs()
print(len(strongs_tags))
st.text('Loaded {} verses'.format(len(verses)))
st.text('Loaded {} tags'.format(len(strongs_tags)))

books = []
for k in verses:
    book = k[:k.index(" ", 2)]
    if book not in books:
        books.append(book)
print(books)

all_defs = {k: f"{k} - {strongs_defs[k]}" for k in strongs_defs}

def format_strong(number):
    return f"{number} - {strongs_defs[number]}"

option1 = st.multiselect('Select Strongs numbers for first concept', all_defs.keys(),  ['0025', '0026'], format_func=format_strong)
option2 = st.multiselect('Select Strongs numbers for second concept', all_defs.keys(),
                         ["5368", "5360", "5363", "5362", "5361", "5366", "5377"], format_func=format_strong)


@st.cache(allow_output_mutation=True)
def get_models():
    tokenizer = AutoTokenizer.from_pretrained('bert-base-cased')
    model = AutoModel.from_pretrained('bert-base-cased', output_hidden_states=True).eval()
    return tokenizer, model


@st.cache
def get_all_embeddings(greek_words):
    embeddings = []

    for word in greek_words:
        for number in greek_words[word]:
            if number in strongs_tags:
                gw = word
                for verse, idx in strongs_tags[number]:
                    if verse in verses:
                        text = verses[verse]
                        words = [x for x in re.split('([ \'])', text) if x != " " and x != "" and x != "'"]
                        if len(words) <= idx - 1:
                            continue
                        ew = words[idx-1].strip(",.!?;:()\"'-")
                        emb = get_embedding(text, ew)
                        embeddings.append((emb, f"{verse} {text}",  gw, book))
    return embeddings


def get_book_type(idx):
    if idx < 4:
        return 'Gospels'
    if idx == 4:
        return 'Acts'
    if idx < 19:
        return 'Pauline letters'
    if idx < 26:
        return 'Short lettters'
    return 'Revelation'



embeddings = get_all_embeddings({"concept1": option1, "concept2": option2})

mapper = umap.UMAP().fit([x[0] for x in embeddings])
ts = mapper.embedding_

x = ts[:,0]
y = ts[:,1]
df = pd.DataFrame({"x": x, "y": y, "verse": ["<br>".join(textwrap.wrap(x[1], 80)) for x in embeddings], "greek word": [x[2] for x in embeddings]})
fig = px.scatter(df, x="x", y="y",
                 hover_data=['verse'], color="greek word", )
# fig.write_html("book_love.html")
st.plotly_chart(fig)