milestone-3.py

import streamlit as st
import torch
from datasets import combine
from datasets import load_dataset
from transformers import AutoTokenizer
from transformers import pipeline

# Load HUPD dataset
dataset_dict = load_dataset(
    "HUPD/hupd",
    name="sample",
    data_files="https://huggingface.co/datasets/HUPD/hupd/blob/main/hupd_metadata_2022-02-22.feather",
    icpr_label=None,
    train_filing_start_date="2016-01-01",
    train_filing_end_date="2016-01-21",
    val_filing_start_date="2016-01-22",
    val_filing_end_date="2016-01-31",
)

# Process data
filtered_dataset = dataset_dict["validation"].filter(
    lambda e: e["decision"] == "ACCEPTED" or e["decision"] == "REJECTED"
)
seed = 88
accepted = filtered_dataset.filter(lambda e: e["decision"] == "ACCEPTED").shuffle(seed).select(range(5))
rejected = filtered_dataset.filter(lambda e: e["decision"] == "REJECTED").shuffle(seed).select(range(5))
dataset = combine.concatenate_datasets([accepted, rejected])
dataset = dataset.sort("patent_number")

# Create pipeline using model trainned on Colab
model = torch.load("patent_classifier_v4.pt", map_location=torch.device("cpu"))
tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
tokenizer_kwargs = {'padding':True,'truncation':True}

def load_data():
    selected_application = dataset.select([applications[st.session_state.id]])
    st.session_state.abstract = selected_application["abstract"][0]
    st.session_state.claims = selected_application["claims"][0]
    st.session_state.title = selected_application["title"][0]
    st.session_state.decision = selected_application["decision"][0]

st.title("CS-GY-6613 Project Milestone 3")

# List patent numbers for select box
applications = {}
for ds_index, example in enumerate(dataset):
    applications.update({example["patent_number"]: ds_index})
st.selectbox(
    "Select a sample patent application:", applications, on_change=load_data, key="id"
)

# Sample title/decision displayed for additional context only, not used with model
st.text_input("Sample Title", key="title", value=dataset[0]["title"],)
st.text_input("Sample Decision", key="decision", value=dataset[0]["decision"])

# Classifier input form
with st.form("Input Form"):
    abstract = st.text_area(
        "Abstract", key="abstract", value=dataset[0]["abstract"], height=200
    )
    claims = st.text_area(
        "Claims", key="claims", value=dataset[0]["abstract"], height=200
    )
    submitted = st.form_submit_button("Get Patentability Score")

if submitted:
    tokens = tokenizer(abstract, claims, return_tensors='pt', **tokenizer_kwargs)
    with torch.no_grad():
        output = model(**tokens)
    logits = output.logits
    pred = torch.softmax(logits, dim=1)
    score = pred[0][1] # index 1 of softmax output is probability that decision = ACCEPTED
    st.markdown(
        "This application's patentability score is  **{}**.".format(score)
    )