|
import streamlit as st |
|
from transformers import pipeline |
|
from transformers import AutoTokenizer, AutoModelForSequenceClassification |
|
|
|
st.title("Milestone #3 offensive statement prediction with pre-trained models") |
|
st.write("in this basic demo you can select a model to judge whether or not the text below is offensive and how it's offensive. the options are[toxic, severe_toxic, obscene, threat, insult, identity_hate]") |
|
text = "I am going to mess you up" |
|
st.write(text) |
|
|
|
options = ["Greys/milestonemodel"] |
|
model = st.selectbox("Select a pre-trained model", options) |
|
|
|
con = st.button("Submit") |
|
if con: |
|
if model == "zero-shot-classification": |
|
classifier = pipeline(model) |
|
res = classifier(text, candidate_labels=["offensive"]) |
|
label = res['labels'][0] |
|
score = res['scores'][0] |
|
st.write(f"Prediction: {label}, Score: {score*100}% chance") |
|
|
|
if model == "cardiffnlp/twitter-roberta-base-offensive": |
|
classifier = pipeline('text-classification', model='cardiffnlp/twitter-roberta-base-offensive', tokenizer='cardiffnlp/twitter-roberta-base-offensive') |
|
result = classifier(text) |
|
label = result[0]['label'] |
|
score = result[0]['score'] |
|
st.write(f"Prediction: {label}, Score: {score*100}% chance") |
|
|
|
if model == "Greys/milestonemodel": |
|
|
|
tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased") |
|
model = AutoModelForSequenceClassification.from_pretrained("Greys/milestonemodel") |
|
my_list = [' ','toxic', 'severe_toxic', 'obscene', 'threat', 'insult', 'identity_hate'] |
|
def classify_sentence(text): |
|
inputs = tokenizer(text, return_tensors="pt") |
|
outputs = model(**inputs) |
|
probs = outputs.logits.softmax(dim=1) |
|
return probs.detach().numpy()[0] |
|
probs = classify_sentence(text) |
|
def find_largest_number(numbers): |
|
max_num = numbers[0] |
|
max_index = 0 |
|
for i in range(1, len(numbers)): |
|
if numbers[i] > max_num: |
|
max_num = numbers[i] |
|
max_index = i |
|
return max_index |
|
|
|
print(probs) |
|
index = find_largest_number(probs) |
|
st.write(my_list[index]) |
|
|
|
|
