exercise / scripture_compare.py
levimack's picture
Update scripture_compare.py
daf2dcd verified
raw
history blame
3.77 kB
import gradio as gr
import urllib.request, json
from transformers.utils import logging
logging.set_verbosity_error()
from sentence_transformers import SentenceTransformer
from sentence_transformers import util
# Search URL for Mackabee Ministries scriptures. Will fetch scriptures as a json object.
SEARCH_URL = 'https://dd4-biblical.appspot.com/_api/scriptures/v1/search?searchText={}&lang=en&version={}'
STANDARD_VERSION = 'RSKJ' # Normally RSKJ (Restored King James Version) as our standard text, may also use NRSV.
COMPARISON_VERSION = 'SEP' # Will use the Septuagint "SEP" version for most comparisons, also have DSS available for Isa
model = SentenceTransformer("all-MiniLM-L6-v2")
# Candidate helper class that holds a scripture, the comparison text and the resulting match score.
class Candidate:
compareText = ''
score = 0
def __init__(self, book, chapter, verse, standardText):
self.book = book
self.chapter = chapter
self.verse = verse
self.standardText = standardText
def reference(self):
return '{} {}:{}'.format(self.book, self.chapter, self.verse)
def __str__(self):
return '{},{:.4f},"{}","{}"'.format(self.reference(), self.score, self.standardText, self.compareText)
def __repr__(self):
return str(self)
def compare(reference):
candidates = []
candidateMap = {}
# Connect to the api and get the standard translation
with urllib.request.urlopen(SEARCH_URL.format(urllib.parse.quote(reference), STANDARD_VERSION)) as url:
response = json.load(url)
# print(response)
for result in response['items']:
# print(result)
candidate = Candidate(result['book'], result['chapter'], result['verse'], result['text'])
candidates.append(candidate)
candidateMap[candidate.reference()] = candidate
# Then fetch the comparison text
with urllib.request.urlopen(SEARCH_URL.format(urllib.parse.quote(reference), COMPARISON_VERSION)) as url:
response = json.load(url)
# print(response)
for result in response['items']:
# print(result)
candidate = candidateMap['{} {}:{}'.format(result['book'], result['chapter'], result['verse'])]
if candidate is None :
candidate = Candidate(result['book'], result['chapter'], result['verse'], '')
candidates.append(candidate)
candidateMap[candidate.reference()] = candidate
candidate.compareText = result['text']
# Isa 1:1 standardText: This is the book of Isaiah compareText: The book that Isaiah wrote.
# Isa 1:2 standardText: Isaiah was a good man compareText: Isaiah did what was right.
standardTexts = [] # 2 items This is the book of Isaiah, Isaiah was a good man
compareTexts = [] # 2 items The book that Isaiah wrote., Isaiah did what was right.
for candidate in candidates:
standardTexts.append(candidate.standardText)
compareTexts.append(candidate.compareText)
embeddings1 = model.encode(standardTexts, convert_to_tensor=True)
embeddings2 = model.encode(compareTexts, convert_to_tensor=True)
cosine_scores = util.cos_sim(embeddings1, embeddings2)
for i in range(len(candidates)):
candidates[i].score = cosine_scores[i][i]
# print(candidates[i])
return '\n\n'.join(str(c) for c in candidates)
demo = gr.Interface(
compare,
gr.Textbox(
label="Scriptures",
info="Enter a verse, verse range, chapter(s) or mix (e.g. Gen 2:3 or Gen 2:1-3 or Exo 12 or Lev 20-23 or Gen 2:3,Exo 12,Lev 20-23)",
lines=1,
),
gr.Textbox(label="Comparisons", lines=28))
demo.launch()