inference-app.py

import streamlit as st
import pandas as pd
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import Dataset, DataLoader
from transformers import AutoTokenizer,AutoModel
import random
from bs4 import BeautifulSoup
import re


from transformers import AutoModelForSequenceClassification
import pytorch_lightning as pl

device = "cuda:0" if torch.cuda.is_available() else "cpu"

train_path = "train.csv"
test_path = "test.csv"
test_labels_paths = "test_labels.csv"
test_df = pd.read_csv(test_path)
test_labels_df = pd.read_csv(test_labels_paths)
test_df = pd.concat([test_df.iloc[:, 1], test_labels_df.iloc[:, 1:]], axis = 1)
test_df.to_csv("test-dataset.csv")
test_dataset_path = "test-dataset.csv"

#Lets make a new column labeled "healthy"

def healthy_filter(df):
  if (df["toxic"]==0) and (df["severe_toxic"]==0) and (df["obscene"]==0) and (df["threat"]==0) and (df["insult"]==0) and (df["identity_hate"]==0):
    return 1
  else:
    return 0

attributes = ['toxic', 'severe_toxic', 'obscene', 'threat',
       'insult', 'identity_hate', 'healthy']

class Comments_Dataset(Dataset):
  def __init__(self, data_path, tokenizer, attributes, max_token_len = 128, sample=5000):
    self.data_path = data_path
    self.tokenizer = tokenizer
    self.attributes = attributes
    self.max_token_len = max_token_len
    self.sample = sample
    self._prepare_data()

  def _prepare_data(self):
    data = pd.read_csv(self.data_path)
    data["healthy"] = data.apply(healthy_filter,axis=1)
    data["unhealthy"] = np.where(data['healthy']==1, 0, 1)
    if self.sample is not None:
      unhealthy = data.loc[data["healthy"] == 0]
      healthy = data.loc[data["healthy"] ==1]
      self.data = pd.concat([unhealthy, healthy.sample(self.sample, random_state=42)])
    else:
      self.data = data

  def __len__(self):
    return len(self.data)
  
  def __getitem__(self,index):
    item = self.data.iloc[index]
    comment = str(item.comment_text)
    attributes = torch.FloatTensor(item[self.attributes])
    tokens = self.tokenizer.encode_plus(comment,
                                      add_special_tokens=True,
                                      return_tensors='pt',
                                      truncation=True,
                                      padding='max_length',
                                      max_length=self.max_token_len,
                                      return_attention_mask = True)
    return {'input_ids': tokens.input_ids.flatten(), 'attention_mask': tokens.attention_mask.flatten(), 'labels': attributes}


class Comments_Data_Module(pl.LightningDataModule):

  def __init__(self, train_path, val_path, attributes, batch_size: int = 16, max_token_length: int = 128,  model_name='roberta-base'):
    super().__init__()
    self.train_path = train_path
    self.val_path = val_path
    self.attributes = attributes
    self.batch_size = batch_size
    self.max_token_length = max_token_length
    self.model_name = model_name
    self.tokenizer = AutoTokenizer.from_pretrained(model_name)

  def setup(self, stage = None):
    if stage in (None, "fit"):
      self.train_dataset = Comments_Dataset(self.train_path, attributes=self.attributes, tokenizer=self.tokenizer)
      self.val_dataset = Comments_Dataset(self.val_path, attributes=self.attributes, tokenizer=self.tokenizer, sample=None)
    if stage == 'predict':
      self.val_dataset = Comments_Dataset(self.val_path, attributes=self.attributes, tokenizer=self.tokenizer, sample=None)

  def train_dataloader(self):
    return DataLoader(self.train_dataset, batch_size = self.batch_size, num_workers=4, shuffle=True)

  def val_dataloader(self):
    return DataLoader(self.val_dataset, batch_size = self.batch_size, num_workers=4, shuffle=False)

  def predict_dataloader(self):
    return DataLoader(self.val_dataset, batch_size = self.batch_size, num_workers=4, shuffle=False)

comments_data_module = Comments_Data_Module(train_path, test_dataset_path, attributes=attributes)
comments_data_module.setup()
comments_data_module.train_dataloader()

class Comment_Classifier(pl.LightningModule):
#the config dict has the hugginface parameters in it
  def __init__(self, config: dict):
    super().__init__()
    self.config = config
    self.pretrained_model = AutoModel.from_pretrained(config['model_name'], return_dict = True)
    self.hidden = torch.nn.Linear(self.pretrained_model.config.hidden_size, self.pretrained_model.config.hidden_size)
    self.classifier = torch.nn.Linear(self.pretrained_model.config.hidden_size, self.config['n_labels'])
    torch.nn.init.xavier_uniform_(self.classifier.weight)
    self.loss_func = nn.CrossEntropyLoss()
    self.dropout = nn.Dropout()
    
  def forward(self, input_ids, attention_mask, labels=None):
    # roberta layer
    output = self.pretrained_model(input_ids=input_ids, attention_mask=attention_mask)
    pooled_output = torch.mean(output.last_hidden_state, 1)
    # final logits / classification layers
    pooled_output = self.dropout(pooled_output)
    pooled_output = self.hidden(pooled_output)
    pooled_output = F.relu(pooled_output)
    pooled_output = self.dropout(pooled_output)
    logits = self.classifier(pooled_output)
    # calculate loss
    loss = 0
    if labels is not None:
      loss = self.loss_func(logits.view(-1, self.config['n_labels']), labels.view(-1, self.config['n_labels']))
    return loss, logits

  def training_step(self, batch, batch_index):
    loss, outputs = self(**batch)
    self.log("train loss ", loss, prog_bar = True, logger=True)
    return {"loss":loss, "predictions":outputs, "labels": batch["labels"]}

  def validation_step(self, batch, batch_index):
    loss, outputs = self(**batch)
    self.log("validation loss ", loss, prog_bar = True, logger=True)
    return {"val_loss": loss, "predictions":outputs, "labels": batch["labels"]}

  def predict_step(self, batch, batch_index):
    loss, outputs = self(**batch)
    return outputs

  def configure_optimizers(self):
    optimizer = AdamW(self.parameters(), lr=self.config['lr'], weight_decay=self.config['weight_decay'])
    total_steps = self.config['train_size']/self.config['batch_size']
    warmup_steps = math.floor(total_steps * self.config['warmup'])
    scheduler = get_cosine_schedule_with_warmup(optimizer, warmup_steps, total_steps)
    return [optimizer],[scheduler]

    
config = {
    'model_name': 'distilroberta-base',
    'n_labels': len(attributes),
    'batch_size': 128,
    'lr': 1.5e-6,
    'warmup': 0.2, 
    'train_size': len(comments_data_module.train_dataloader()),
    'weight_decay': 0.001,
    'n_epochs': 100
}

model_name = 'distilroberta-base'
tokenizer = AutoTokenizer.from_pretrained(model_name)

model = Comment_Classifier(config=config)
model.load_state_dict(torch.load("model_state_dict.pt"))
model.eval()



def prepare_tokenized_review(raw_review):
  # Remove HTML tags with BS
  review_text = BeautifulSoup(raw_review).get_text()
  # Removing non-letters using a regular expression
  review_text = re.sub("[^a-zA-Z!?]"," ", review_text)
  # Convert words to lower case and split them
  words = review_text.lower().split()

  return " ".join(words)

def get_encodings(text):
    MAX_LEN=256
    encodings = tokenizer.encode_plus(
        text,
        None,
        add_special_tokens=True,
        max_length=MAX_LEN,
        padding='max_length',
        truncation=True,
        return_attention_mask=True,
        return_tensors='pt')
    return encodings

def run_inference(encoding):
  with torch.no_grad():
    input_ids = encoding['input_ids'].to(device, dtype=torch.long)
    attention_mask = encoding['attention_mask'].to(device, dtype=torch.long)
    output = model(input_ids, attention_mask)
    final_output = torch.softmax(output[1][0],dim=0).cpu()
    print(final_output.numpy().tolist())
    return final_output.numpy().tolist()



test_tweets = test_df["comment_text"].values
#streamlit section
models = ["distilroberta-base"]
model_pointers = ["default: distilroberta-base"]

# current_random_tweet = test_tweets[random.randint(0,len(test_tweets))]
# current_random_tweet = prepare_tokenized_review(current_random_tweet)
st.write("1. Hit the button to view and see the analyis of a random tweet")

with st.form(key="init_form"):
    current_random_tweet = test_tweets[random.randint(0,len(test_tweets))]
    current_random_tweet = prepare_tokenized_review(current_random_tweet)
 


    choice = st.selectbox("Choose Model", model_pointers)

    user_picked_model = models[model_pointers.index(choice)]
    with st.spinner("Analyzing..."):
        text_encoding = get_encodings(current_random_tweet)
        result = run_inference(text_encoding)
        df = pd.DataFrame({"Tweet":current_random_tweet}, index=[0])
        df["Highest Toxicity Class"] = attributes[result.index(max(result))]
        df["Sentiment Score"] = max(result)
        st.table(df)

    next_tweet = st.form_submit_button("Next Tweet")

if next_tweet:
    with st.spinner("Analyzing..."):
        st.write("")