app.py

import numpy as np
import pandas as pd
import torch
from torch import nn
from torch.utils.data import Dataset, DataLoader
from transformers import AutoTokenizer, BertModel
from sklearn import metrics
import streamlit as st

# Define constants. Enable CUDA if available.
MAX_LENGTH = 100
INFER_BATCH_SIZE = 128
HEAD_DROP_OUT = 0.4

device = "cuda" if torch.cuda.is_available() else "cpu"
bert_path = "bert-base-uncased"
tokenizer = AutoTokenizer.from_pretrained(bert_path)

# Read and format data.
tweets_raw = pd.read_csv("test.csv", nrows=50)
labels_raw = pd.read_csv("test_labels.csv", nrows=50)

label_set = ["toxic", "severe_toxic", "obscene", "threat", "insult", "identity_hate"]
label_vector = labels_raw[label_set].values.tolist()

tweet_df = tweets_raw[["comment_text"]]
tweet_df["labels"] = label_vector

# Dataset for loading tables into DataLoader
class ToxicityDataset(Dataset):
    def __init__(self, dataframe, tokenizer, max_len):
        self.tokenizer = tokenizer
        self.data = dataframe
        self.text = self.data.comment_text
        self.targets = self.data.labels
        self.max_len = max_len

    def __len__(self):
        return len(self.text)

    def __getitem__(self, index):
        text = str(self.text[index])
        text = " ".join(text.split())

        inputs = self.tokenizer.encode_plus(
            text,
            None,
            add_special_tokens=True,
            max_length=self.max_len,
            padding="max_length",
            truncation=True,
            return_token_type_ids=True,
        )
        ids = inputs["input_ids"]
        mask = inputs["attention_mask"]
        token_type_ids = inputs["token_type_ids"]
        return {
            "ids": torch.tensor(ids, dtype=torch.long),
            "mask": torch.tensor(mask, dtype=torch.long),
            "token_type_ids": torch.tensor(token_type_ids, dtype=torch.long),
            "targets": torch.tensor(self.targets[index], dtype=torch.float),
        }

# Based on user model selection, prepare Dataset and DataLoader
infer_dataset = ToxicityDataset(tweet_df, tokenizer, MAX_LENGTH)
infer_params = {"batch_size": INFER_BATCH_SIZE, "shuffle": False}
infer_loader = DataLoader(infer_dataset, **infer_params)

class BertClass(torch.nn.Module):
    def __init__(self):
        super(BertClass, self).__init__()
        self.l1 = BertModel.from_pretrained(bert_path)
        self.dropout = torch.nn.Dropout(HEAD_DROP_OUT)
        self.classifier = torch.nn.Linear(768, 6)

    # return_dict must equal False for Huggingface Transformers v4+
    def forward(self, input_ids, attention_mask, token_type_ids):
        output_1 = self.l1(
            input_ids=input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            return_dict=False,
        )
        hidden_state = output_1[0]
        pooler = hidden_state[:, 0]
        pooler = self.dropout(pooler)
        output = self.classifier(pooler)
        return output

class PretrainedBertClass(torch.nn.Module):
    def __init__(self):
        super(PretrainedBertClass, self).__init__()
        self.l1 = BertModel.from_pretrained(bert_path)
        self.l2 = torch.nn.Dropout(HEAD_DROP_OUT)
        self.l3 = torch.nn.Linear(768, 6)
    
    def forward(self, ids, mask, token_type_ids):
        _, output_1= self.l1(ids, attention_mask = mask, token_type_ids = token_type_ids, return_dict=False)
        output_2 = self.l2(output_1)
        output = self.l3(output_2)
        return output

# User selects model for front-end.
option = st.selectbox("Select a text analysis model:", ("BERT", "Fine-tuned BERT"))
if option == "BERT":
    model = PretrainedBertClass()
else:
    model = torch.load("pytorch_bert_toxic.bin", map_location=torch.device("cpu"))

# Freeze model and input tokens
def inference():
    model.eval()
    final_targets = []
    final_outputs = []
    with torch.no_grad():
        for _, data in enumerate(infer_loader, 0):
            ids = data["ids"].to(device, dtype=torch.long)
            mask = data["mask"].to(device, dtype=torch.long)
            token_type_ids = data["token_type_ids"].to(device, dtype=torch.long)
            targets = data["targets"].to(device, dtype=torch.float)
            outputs = model(ids, mask, token_type_ids)
            final_targets.extend(targets.cpu().detach().numpy().tolist())
            final_outputs.extend(torch.sigmoid(outputs).cpu().detach().numpy().tolist())
    return final_outputs, final_targets

# Get predictions!
prediction, targets = inference()

# Format and present findings.
best_preds = []
best_labels = []
for example in prediction:
    i = np.argmax(example)
    best_prediction = example[i]
    best_label = label_set[i]
    best_preds.append(best_prediction)
    best_labels.append(best_label)

st.write("Toxicity Classification Result:")
display_table = tweets_raw[["comment_text"]]
display_table["Toxicity Classification"] = best_labels
display_table["Probability"] = best_preds
st.write(display_table)