from sklearn.feature_extraction.text import TfidfVectorizer
from spacy.lang.en import English
import numpy as np
import gradio as gr

nlp = English()
nlp.create_pipe('sentencizer')


def summarizer(text, tokenizer, max_sent_in_summary=5):
    # Create spacy document for further sentence level tokenization
    doc = nlp(text.replace("\n", ""))
    sentences = [sent.text.strip() for sent in doc.sents]
    # Let's create an organizer which will store the sentence ordering to later reorganize the 
    # scored sentences in their correct order
    sentence_organizer = {k:v for v,k in enumerate(sentences)}
    # Let's now create a tf-idf (Term frequnecy Inverse Document Frequency) model
    tf_idf_vectorizer = TfidfVectorizer(min_df=2,  max_features=None, 
                                        strip_accents='unicode', 
                                        analyzer='word',
                                        token_pattern=r'\w{1,}',
                                        ngram_range=(1, 3), 
                                        use_idf=1,smooth_idf=1,
                                        sublinear_tf=1,
                                        stop_words = 'english')
    # Passing our sentences treating each as one document to TF-IDF vectorizer
    tf_idf_vectorizer.fit(sentences)
    # Transforming our sentences to TF-IDF vectors
    sentence_vectors = tf_idf_vectorizer.transform(sentences)
    # Getting sentence scores for each sentences
    sentence_scores = np.array(sentence_vectors.sum(axis=1)).ravel()
    # Getting top-n sentences
    N = max_sent_in_summary
    top_n_sentences = [sentences[ind] for ind in np.argsort(sentence_scores, axis=0)[::-1][:N]]
    # Let's now do the sentence ordering using our prebaked sentence_organizer
    # Let's map the scored sentences with their indexes
    mapped_top_n_sentences = [(sentence,sentence_organizer[sentence]) for sentence in top_n_sentences]
    # Ordering our top-n sentences in their original ordering
    mapped_top_n_sentences = sorted(mapped_top_n_sentences, key = lambda x: x[1])
    ordered_scored_sentences = [element[0] for element in mapped_top_n_sentences]
    # Our final summary
    summary = " ".join(ordered_scored_sentences)
    return summary

demo = gr.Interface(
    fn=summarizer,
    inputs=["text","text"],
    outputs="text",
)
demo.launch(debug=True)