app.py

import torch
import gradio as gr
from datasets import load_dataset
from transformers import pipeline
from textwrap import dedent
from email import message_from_file
from email.header import decode_header

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

# load model
pipe = pipeline(model="1aurent/distilbert-base-multilingual-cased-finetuned-email-spam", device=device)

# fn to predict from text
def classify_raw(text):
    return pipe(text, top_k=2)

# fn to predict from form inputs
def classify_form(mailfrom, x_mailfrom, to, reply_to, subject):
    text = dedent(f"""
        From: {mailfrom}
        X-MailFrom: {x_mailfrom}
        To: {to}
        Reply-To: {reply_to}
        Subject: {subject}
    """).strip()
    return pipe(text, top_k=2)

# helper to extract header from email
def get_header(message, header_name: str) -> str:
    try:
        for payload, _ in decode_header(message[header_name]):
            if type(payload) == bytes:
                payload = payload.decode(errors="ignore")
            header = payload
        header = header.replace("\n", " ")
        header = header.strip()
        return header
    except:
        return ""

# fn to predict from email file
def classify_file(file):
    message = message_from_file(open(file.name))

    return classify_form(
        mailfrom=get_header(message, "From"),
        x_mailfrom=get_header(message, "X-MailFrom"),
        to=get_header(message, "To"),
        reply_to=get_header(message, "Reply-To"),
        subject=get_header(message, "Subject"),
    )
    

title = "Email Spam Classifier"
description = """
Spam or ham ?
"""

demo = gr.Blocks()

raw_interface = gr.Interface(
    fn=classify_raw,
    inputs=gr.Textbox(
        label="Formatted Email Header",
        lines=5,
        placeholder=dedent("""
          From: Laurent Fainsin <[email protected]>
          X-MailFrom: Laurent Fainsin <[email protected]>
          To: net7 <[email protected]>
          Reply-To: Laurent Fainsin <[email protected]>
          Subject: Re: Demande d'un H24 net7
        """).strip(),
    ),
    outputs="json",
    api_name="predict_raw_text",
)

form_interface = gr.Interface(
    fn=classify_form,
    inputs=[
        gr.Textbox(
            label="From",
            placeholder="Laurent Fainsin <[email protected]>",
        ),
        gr.Textbox(
            label="X-MailFrom",
            placeholder="Laurent Fainsin <[email protected]>",
        ),
        gr.Textbox(
            label="To",
            placeholder="net7 <[email protected]>",
        ),
        gr.Textbox(
            label="Reply-To",
            placeholder="Laurent Fainsin <[email protected]>",
        ),
        gr.Textbox(
            label="Subject",
            placeholder="Re: Demande d'un H24 net7",
        ),
    ],
    outputs="json",
    api_name="predict_form",
)

file_interface = gr.Interface(
    fn=classify_file,
    inputs=gr.File(
        label="Email File",
        file_types=[".eml"],
    ),
    outputs="json",
    api_name="predict_file",
)

with demo:
    gr.TabbedInterface(
        interface_list=[
            raw_interface,
            form_interface,
            file_interface
        ],
        tab_names=[
            "Raw Text",
            "Form",
            "File"
        ]
    )

demo.launch()