app.py

# imports
import gradio as gr
import pandas as pd
import tempfile
import itertools
import torch
import numpy as np
from numpy import dot
from numpy.linalg import norm, multi_dot
from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer

# compute dot product of inputs
# summary function - test for single gradio function interfrace
def gr_cosine_similarity(sentence1, sentence2):
  # Create class for data preparation
  class SimpleDataset:
      def __init__(self, tokenized_texts):
          self.tokenized_texts = tokenized_texts
      
      def __len__(self):
          return len(self.tokenized_texts["input_ids"])
      
      def __getitem__(self, idx):
          return {k: v[idx] for k, v in self.tokenized_texts.items()}

  # load tokenizer and model, create trainer
  model_name = "j-hartmann/emotion-english-distilroberta-base"
  tokenizer = AutoTokenizer.from_pretrained(model_name)
  model = AutoModelForSequenceClassification.from_pretrained(model_name)
  trainer = Trainer(model=model)  
    
  # sentences in list
  lines_s = [sentence1, sentence2]
  print(type(sentence1), type(sentence2))
  print(sentence1, sentence2)
  print(lines_s)
 
    # Tokenize texts and create prediction data set
  tokenized_texts = tokenizer(lines_s, truncation=True, padding=True)
  pred_dataset = SimpleDataset(tokenized_texts)

    # Run predictions -> predict whole df
  predictions = trainer.predict(pred_dataset)

    # Transform predictions to labels
  preds = predictions.predictions.argmax(-1)
  labels = pd.Series(preds).map(model.config.id2label)
  scores = (np.exp(predictions[0])/np.exp(predictions[0]).sum(-1,keepdims=True)).max(1)
    # scores raw
  temp = (np.exp(predictions[0])/np.exp(predictions[0]).sum(-1, keepdims=True)).tolist()


    # work in progress
  # container
  anger = []
  disgust = []
  fear = []
  joy = []
  neutral = []
  sadness = []
  surprise = []

  print(temp)
  # extract scores (as many entries as exist in pred_texts)
  for i in range(len(lines_s)):
    anger.append(temp[i][0])
    disgust.append(temp[i][1])
    fear.append(temp[i][2])
    joy.append(temp[i][3])
    neutral.append(temp[i][4])
    sadness.append(temp[i][5])
    surprise.append(temp[i][6])

  # define both vectors for the dot product
  # each include all values for both predictions
  v1 = temp[0]
  v2 = temp[1]
  print(type(v1), type(v2))
  # compute dot product of all 
  dot_product = dot(v1, v2)

   # define df
  df = pd.DataFrame(list(zip(lines_s,labels, anger, disgust, fear, joy, neutral, sadness, surprise)), columns=['text','label', 'anger', 'disgust', 'fear', 'joy', 'neutral', 'sadness', 'surprise'])

  # compute cosine similarity
  # is dot product of vectors n / norms 1*..*n vectors
  cosine_similarity = dot_product / (norm(v1) * norm(v2))


  # return dataframe for space output
  return  df, cosine_similarity
  
 

gr.Interface(gr_cosine_similarity,
                   [
                    gr.inputs.Textbox(lines=1, placeholder="This movie always makes me cry..", default="", label="Text 1"),
                    gr.inputs.Textbox(lines=1, placeholder="Her dog is sad.", default="", label="Text 2"),

                    #gr.outputs.Textbox(type="auto", label="Cosine similarity"),
                                         ],
                   ["dataframe","text"]
                   ).launch(debug=True)