import streamlit as st
from transformers import pipeline
import torch
import matplotlib.pyplot as plt
import numpy as np
#from PIL import Image


#pipe = pipeline(model="RuudVelo/dutch_news_classifier_bert_finetuned")
#text = st.text_area('Please type/copy/paste the Dutch article')

#labels = ['Binnenland' 'Buitenland' 'Cultuur & Media' 'Economie' 'Koningshuis'
# 'Opmerkelijk' 'Politiek' 'Regionaal nieuws' 'Tech']

#if text:
#   out = pipe(text)
#   st.json(out)
   
   
   # load tokenizer and model, create trainer
  #model_name = "RuudVelo/dutch_news_classifier_bert_finetuned"
  #tokenizer = AutoTokenizer.from_pretrained(model_name)
  #model = AutoModelForSequenceClassification.from_pretrained(model_name)
  #trainer = Trainer(model=model)  
  #print(filename, type(filename))
  #print(filename.name)
  
from transformers import BertForSequenceClassification, BertTokenizer

model = BertForSequenceClassification.from_pretrained("RuudVelo/dutch_news_clf_bert_finetuned")
#from transformers import BertTokenizer

tokenizer = BertTokenizer.from_pretrained("RuudVelo/dutch_news_clf_bert_finetuned")


# Title
st.title("Dutch news article classification")

st.write("This app classifies a Dutch news article into one of 9 pre-defined* article categories")

#image = Image.open('dataset-cover_articles.jpg')
st.image('dataset-cover_articles.jpeg', width=150)

text = st.text_area('Please type/copy/paste text of the Dutch article and click Submit')

#if text:
#   encoding = tokenizer(text, return_tensors="pt")
#   outputs = model(**encoding)
#   predictions = outputs.logits.argmax(-1)
#   probabilities = torch.nn.functional.softmax(outputs.logits, dim=-1)
   
##   fig = plt.figure()
#   ax = fig.add_axes([0,0,1,1])
#   labels_plot = ['Binnenland', 'Buitenland' ,'Cultuur & Media' ,'Economie' ,'Koningshuis',
# 'Opmerkelijk' ,'Politiek', 'Regionaal nieuws', 'Tech']
#   probs_plot = probabilities[0].cpu().detach().numpy()
   
#   ax.barh(labels_plot,probs_plot )
#   st.pyplot(fig)


#input = st.text_input('Context')

if st.button('Submit'):
    with st.spinner('Generating a response...'):
        encoding = tokenizer(text, return_tensors="pt")
        outputs = model(**encoding)
        predictions = outputs.logits.argmax(-1)
        number = predictions[0].cpu().detach().numpy()
        probabilities = torch.nn.functional.softmax(outputs.logits, dim=-1)
   
        fig = plt.figure(figsize=(10,4))
        ax = fig.add_axes([0,0,1,1])
        labels_plot = ['Binnenland', 'Buitenland' ,'Cultuur & Media' ,'Economie' ,'Koningshuis',
 'Opmerkelijk' ,'Politiek', 'Regionaal nieuws', 'Tech']
        probs_plot = probabilities[0].cpu().detach().numpy()*100
   
        ax.barh(labels_plot,probs_plot)
        ax.set_title("Predicted article category probability", fontsize=20)
        ax.set_xlabel("Probability (%)", fontsize=16)
        ax.set_ylabel("Predicted category", fontsize=16)
        
        # change the fontsize
        #ax.set_xticklabels(fontsize=14)
        ax.set_yticklabels(labels_plot, fontsize=14)
        
        st.pyplot(fig)
        
        st.write('The predicted category is: **{}** with a probability of: **{:.1f}%**'.format(labels_plot[number],(probs_plot[predictions])*1))
   #     output = genQuestion(option, input)
   #     print(output)
   # st.write(output)   
#encoding = tokenizer(text, return_tensors="pt")
#import numpy as np
st.write("The pre-defined categories are Binnenland, Buitenland, Cultuur & Media, Economie , Koningshuis, Opmerkelijk, Politiek, 'Regionaal nieuws en Tech")
st.write("The model for this app has been trained using data from Dutch news articles published by NOS. For more information regarding the dataset can be found at https://www.kaggle.com/maxscheijen/dutch-news-articles")
#st.write('\n')
st.write('The model performance details can be found at https://huggingface.co/RuudVelo/dutch_news_classifier_bert_finetuned')