app.py

# Hint: this cheatsheet is magic! https://cheat-sheet.streamlit.app/
import constants
import pandas as pd
import streamlit as st
import matplotlib.pyplot as plt
from transformers import BertForSequenceClassification, AutoTokenizer

import altair as alt
from altair import X, Y, Scale
import base64

import re


def preprocess_text(arabic_text, remove_urls, remove_latin):
    """Apply preprocessing to the given Arabic text.

    Args:
        arabic_text: The Arabic text to be preprocessed.
        remove_urls: Boolean indicating whether to remove URLs.
        remove_latin: Boolean indicating whether to remove Latin characters.

    Returns:
        The preprocessed Arabic text.
    """
    if remove_urls:
        arabic_text = re.sub(
            r"(https|http)?:\/\/(\w|\.|\/|\?|\=|\&|\%)*\b",
            "",
            arabic_text,
            flags=re.MULTILINE,
        )

    if remove_latin:
        arabic_text = re.sub(r"[a-zA-Z]", "", arabic_text)

    arabic_text = arabic_text.strip()

    return arabic_text


@st.cache_data
def render_svg(svg):
    """Renders the given svg string."""
    b64 = base64.b64encode(svg.encode("utf-8")).decode("utf-8")
    html = rf'<p align="center"> <img src="data:image/svg+xml;base64,{b64}"/> </p>'
    c = st.container()
    c.write(html, unsafe_allow_html=True)


@st.cache_data
def convert_df(df):
    # IMPORTANT: Cache the conversion to prevent computation on every rerun
    return df.to_csv(index=None).encode("utf-8")


@st.cache_resource
def load_model(model_name):
    model = BertForSequenceClassification.from_pretrained(model_name)
    return model


tokenizer = AutoTokenizer.from_pretrained(constants.MODEL_NAME)
model = load_model(constants.MODEL_NAME)


def compute_ALDi(sentences, remove_urls=True, remove_latin=True):
    """Computes the ALDi score for the given sentences.

    Args:
        sentences: A list of Arabic sentences.

    Returns:
        A list of ALDi scores for the given sentences.
    """
    progress_text = "Computing ALDi..."
    my_bar = st.progress(0, text=progress_text)

    BATCH_SIZE = 4
    output_logits = []

    preprocessed_sentences = [
        preprocess_text(s, remove_urls, remove_latin) for s in sentences
    ]

    for first_index in range(0, len(preprocessed_sentences), BATCH_SIZE):
        inputs = tokenizer(
            preprocessed_sentences[first_index : first_index + BATCH_SIZE],
            return_tensors="pt",
            padding=True,
        )
        outputs = model(**inputs).logits.reshape(-1).tolist()
        output_logits = output_logits + [max(min(o, 1), 0) for o in outputs]
        my_bar.progress(
            min((first_index + BATCH_SIZE) / len(preprocessed_sentences), 1),
            text=progress_text,
        )
    my_bar.empty()
    return output_logits


@st.cache_data
def render_metadata():
    """Renders the metadata."""
    html = r"""<p align="center">
        <a href="https://huggingface.co/AMR-KELEG/Sentence-ALDi"><img alt="HuggingFace Model" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-8A2BE2"></a>
        <a href="https://github.com/AMR-KELEG/ALDi"><img alt="GitHub" src="https://img.shields.io/badge/%F0%9F%93%A6%20GitHub-orange"></a>
        <a href="https://arxiv.org/abs/2310.13747"><img alt="arXiv" src="https://img.shields.io/badge/arXiv-2310.13747-b31b1b.svg"></a>
        </p>"""
    c = st.container()
    c.write(html, unsafe_allow_html=True)


render_svg(open("assets/ALDi_logo.svg").read())
render_metadata()

tab1, tab2 = st.tabs(["Input a Sentence", "Upload a File"])

with tab1:
    sent = st.text_input(
        "Arabic Sentence:", placeholder="Enter an Arabic sentence.", on_change=None
    )

    # TODO: Check if this is needed!
    clicked = st.button("Submit")

    remove_urls = st.toggle("Remove urls")
    remove_latin = st.toggle("Remove Latin characters")

    if sent:
        ALDi_score = compute_ALDi(
            [sent], remove_urls=remove_urls, remove_latin=remove_latin
        )[0]

        ORANGE_COLOR = "#FF8000"
        fig, ax = plt.subplots(figsize=(8, 1))
        fig.patch.set_facecolor("none")
        ax.set_facecolor("none")

        ax.spines["left"].set_color(ORANGE_COLOR)
        ax.spines["bottom"].set_color(ORANGE_COLOR)
        ax.tick_params(axis="x", colors=ORANGE_COLOR)

        ax.spines[["right", "top"]].set_visible(False)

        ax.barh(y=[0], width=[ALDi_score], color=ORANGE_COLOR)
        ax.set_xlim(0, 1)
        ax.set_ylim(-1, 1)
        ax.set_title(f"ALDi score is: {round(ALDi_score, 3)}", color=ORANGE_COLOR)
        ax.get_yaxis().set_visible(False)
        ax.set_xlabel("ALDi score", color=ORANGE_COLOR)
        st.pyplot(fig)

        print(sent)
        with open("logs.txt", "a") as f:
            f.write(sent + "\n")

with tab2:
    file = st.file_uploader("Upload a file", type=["txt"])
    if file is not None:
        df = pd.read_csv(file, sep="\t", header=None)
        df.columns = ["Sentence"]
        df.reset_index(drop=True, inplace=True)

        # TODO: Run the model
        df["ALDi"] = compute_ALDi(df["Sentence"].tolist())

        # A horizontal rule
        st.markdown("""---""")

        chart = (
            alt.Chart(df.reset_index())
            .mark_area(color="darkorange", opacity=0.5)
            .encode(
                x=X(field="index", title="Sentence Index"),
                y=Y("ALDi", scale=Scale(domain=[0, 1])),
            )
        )
        st.altair_chart(chart.interactive(), use_container_width=True)

        col1, col2 = st.columns([4, 1])

        with col1:
            # Display the output
            st.table(
                df,
            )

        with col2:
            # Add a download button
            csv = convert_df(df)
            st.download_button(
                label=":file_folder: Download predictions as CSV",
                data=csv,
                file_name="ALDi_scores.csv",
                mime="text/csv",
            )