Spaces:
Runtime error
Runtime error
| import streamlit as st | |
| from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification | |
| import torch | |
| import pandas as pd | |
| # Function to load the pre-trained model | |
| def load_model(model_name): | |
| tokenizer = AutoTokenizer.from_pretrained(model_name) | |
| model = AutoModelForSequenceClassification.from_pretrained(model_name) | |
| sentiment_pipeline = pipeline("sentiment-analysis", tokenizer=tokenizer, model=model) | |
| return sentiment_pipeline | |
| # Function to load the pre-trained model | |
| def load_finetune_model(model_name): | |
| tokenizer = AutoTokenizer.from_pretrained(model_name) | |
| model = AutoModelForSequenceClassification.from_pretrained(model_name) | |
| return tokenizer, model | |
| # | |
| def score(item): | |
| return item['score'] | |
| # Streamlit app | |
| st.title("Basic Sentiment Analysis App based on DistilBERT -- from hugging-face spaces ") | |
| st.write("Enter a text and select a pre-trained model to get the sentiment analysis.") | |
| # Input text | |
| default_text = "I love my dog, she's so cute." | |
| text = st.text_input("Enter your text:", value=default_text) | |
| # Model selection | |
| # distilbert loaded from hugging face and finetuned model built on training data | |
| model_option = { | |
| "distilbert-base-uncased-finetuned-sst-2-english": { | |
| "labels": ["NEGATIVE", "POSITIVE"], | |
| "description": "This model classifies text into positive or negative sentiment. It is based on DistilBERT and fine-tuned on the Stanford Sentiment Treebank (SST-2) dataset.", | |
| }, | |
| "emmaenglish/finetuned_distilbert": { | |
| "description": "This model detects different types of toxicity like threats, obscenity, insults, and identity-based hate in text.", | |
| }, | |
| } | |
| # user choses model | |
| model = st.selectbox("Choose a fine-tuned model:", model_option) | |
| # app displays model information description | |
| st.write("### Model Information") | |
| st.write(f"**Description:** {model_option[model]['description']}") | |
| # Load the model and perform sentiment analysis | |
| if st.button("Analyze"): | |
| # no text is entered | |
| if not text: | |
| st.write("Please enter a text.") | |
| else: | |
| with st.spinner("Analyzing toxicity..."): | |
| # user choses finetuned model trained on data in google cola b | |
| if model == "emmaenglish/finetuned_distilbert": | |
| classifier = AutoModelForSequenceClassification.from_pretrained(model) | |
| # tokenizer seperates text into smaller units | |
| tokenizer = AutoTokenizer.from_pretrained(model) | |
| text_token = tokenizer(text, return_tensors="pt") | |
| output = classifier(**text_token) | |
| prediction = torch.sigmoid(output.logits)*100 | |
| prediction = prediction.detach().numpy().tolist()[0] | |
| category_names = ["toxic", "severe toxic", "obscene", "threat", "insult", "identity hate"] | |
| output = [] | |
| for predict, category_names in (zip(prediction, category_names)): | |
| output.append({'label': category_names, 'score': predict}) | |
| labels = output | |
| labels.sort(key=score, reverse=True) | |
| # adding catagorical data for more indepth analysis | |
| df = pd.DataFrame([(text, labels[0]['label'], f"{round(labels[0]['score'], 3)}%", labels[1]['label'], f"{round(labels[1]['score'], 3)}%")], columns=('tweet/text','label 1', 'score 1', 'label 2', 'score 2')) | |
| st.table(df) | |
| else: | |
| # user chooses sentiment analysis of the model, no extranous model implementation nessasary | |
| classifier = pipeline(model=model) | |
| sentiment = classifier(text)[0]["label"] | |
| score = classifier(text)[0]['score'] | |
| st.write(f"The sentiment is {sentiment}.") | |
| st.write(f"The accuracty of this sentiment is {score}.") | |
| else: | |
| # nothing has been written yet auto display | |
| st.write("Enter a text and click 'Analyze' to perform toxicity analysis.") | |