import sys
import gradio as gr
from transformers import AutoTokenizer
import torch
import json, re
tokenizer = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-ar")
model = torch.load("helsinki_fineTuned.pt", map_location=torch.device('cpu'))
model.eval()

# Open Keywords
with open('merged.json', encoding='utf8') as merged:
    data = json.load(merged)
keyword_map = {}
for en in data:
    keyword_map[en.lower()] = data[en]
merged.close()

# Getting keywords from the file
def getKeywords(word):
    words = word.lower().rstrip()
    if keyword_map.get(words):
        return keyword_map[words], True
    return word, False
    
# Replace keywords with the translation
def final_output(text):
    reg = re.compile('[a-zA-Z][ \-()a-zA-Z]+')
    keywords = re.findall(reg, text)
    text_split = re.split(reg, text)
    n=len(keywords)
    for i in range(n):
        word, found = getKeywords(keywords[i])
        if found:
            text_split[i]+=(word + " (" + keywords[i].rstrip() + ")")
        else:
            text_split[i]+=word
    return ' '.join(text_split)
  
  
  
def translate(input):
    translation = []
    text_list = input.split('.')
    for i in range(len(text_list)):
        encode = model.generate(**tokenizer.prepare_seq2seq_batch(text_list[i],return_tensors='pt'))
        text_ar = tokenizer.batch_decode(encode,skip_special_tokens=True)[0]
        text_post = final_output(text_ar)
        translation.append(text_post)
    article_ar = ".".join(translation)
    return article_ar
# Wrap up function
translate_interface = gr.Interface(fn = translate,
                                   allow_flagging = True,
                                   flagging_dir = 'flagged/logs',
                                   title = 'Translating "English Data Science" content into Arabic',
                                   inputs=gr.inputs.Textbox(lines = 7, label = 'English content'),
                                   outputs="text",
                                   examples = [
                                   ['In the last few years the RNN-based architectures have shown the best performance in machine translation problems, but still they have some problems that had to be solved. First, they have a difficulty to cope with long-range dependencies (also LSTM when it has to deal with really long sentences). Secondly, each hidden state depends on the previous one which impossible to parallelize and makes it inefficient on GPUs.']
                                   , 
                                   ['Data science is an interdisciplinary field that uses scientific methods, processes, algorithms and systems to extract knowledge and insights from noisy, structured and unstructured data,[1][2] and apply knowledge and actionable insights from data across a broad range of application domains. Data science is related to data mining, machine learning and big data.']
                                   ]
                                                               )
                                                               
translate_interface.launch()