app.py

import streamlit as st
from lime.lime_text import LimeTextExplainer
from nltk.corpus import stopwords

from deployment_utils import DataPreparator, Predictor, generate_random_sample, generate_highlighted_words, extract_case_information


data_preparator = DataPreparator()
predictor = Predictor()
eng_stop_words = stopwords.words("english")

st.set_page_config(
    page_title="JudgerAI",
    page_icon="🧊",
    layout="wide")

with open("./src/style.css") as f:
    st.markdown(f"<style>{f.read()}</style>", unsafe_allow_html=True)

# application header
left_col, right_col = st.columns(2)

with left_col:
    st.header("Summarize your Case")
    with st.expander(label="Case Summarizer", expanded=True):
        option = st.selectbox(
            'Choose a Method for Entering your Case Facts',
            ('Upload a File', 'Write it Myself'))

        if option == "Upload a File":
            uploaded_file = st.file_uploader(
                label='Upload your Case File (.txt)', type=['txt'])
            if uploaded_file is not None:
                content = uploaded_file.getvalue().decode("utf-8")
                petitioner, respondent, case_facts = extract_case_information(
                    content)

                col1, col2 = st.columns(2)

                with col1:
                    st.write(
                        '<p class="bold-text"> Petitioner </p>', unsafe_allow_html=True)
                    st.info(petitioner)
                with col2:
                    st.write(
                        '<p class="bold-text"> Respondent </p>', unsafe_allow_html=True)
                    st.info(respondent)
                st.write(
                    '<p class="bold-text"> Case Facts </p>', unsafe_allow_html=True)
                st.info(case_facts)

        else:
            case_facts = st.text_area(
                label="Enter your Case Facts youself", height=250)

        submitted = st.button(label="Summarize")

        if submitted:
            with st.spinner("Your Case is being Summarized..."):
                summarized_case_facts = predictor.summarize_facts(case_facts)
                st.write(
                    '<p class="bold-text"> Your Summarized Case Facts </p>', unsafe_allow_html=True)
                st.success(summarized_case_facts)

                summarized_case_facts_file = "petitioner: " + petitioner + "\n" + \
                    "respondent: " + respondent + "\n" + "facts: " + summarized_case_facts

                btn = st.download_button(
                    label="Download",
                    data=summarized_case_facts_file,
                    file_name="summarized_case_facts.txt",
                    mime="file/txt"
                )

with right_col:
    st.header("Predict the Outcome")

    # get_random_case_button = st.button(label="Get Random Sample")

    # input form
    with st.expander(label="Case Outcome Predictor", expanded=True):
        option = st.selectbox(
            'Choose a Method for Entering your Case Information',
            ('Upload a File', 'Write it Myself'))

        if option == 'Upload a File':
            uploaded_file = st.file_uploader(
                label='Upload your Case File (.txt)', type=['txt'], key="prediction_case_uploader")
            if uploaded_file is not None:
                content = uploaded_file.getvalue().decode("utf-8")
                petitioner, respondent, case_facts = extract_case_information(
                    content)

                st.session_state["petitioner"] = petitioner.strip()
                st.session_state["respondent"] = respondent.strip()
                st.session_state["facts"] = case_facts.strip()

        option = st.selectbox(
            "Select Model",
            ("1D Convolutional","BERT","Doc2Vec")
        )

        col1, col2 = st.columns(2)

        with col1:
            petitioner = st.text_input(
                label="Petitioner", key="petitioner")

        with col2:
            respondent = st.text_input(
                label="Respondent", key="respondent")

        global facts
        facts = st.text_area(label="Case Facts",
                             height=300, key="facts")

        # remove stopwords to not highlight it
        facts = " ".join([word for word in facts.split()
                          if word not in eng_stop_words])

        # create `LimeTextExplainer` for models interpretation
        class_names = [petitioner, respondent]
        explainer = LimeTextExplainer(class_names=class_names)

        submitted = st.button(label="Predict")

        if submitted:
            if petitioner and respondent and facts:
                with st.spinner("Analyzing Case Facts ..."):
                    # get predcitions
                    if option == "Doc2Vec":
                        predictions = predictor.predict_doc2vec(facts)
                        output = explainer.explain_instance(
                            facts, predictor.predict_doc2vec)
                        important_words = output.as_list()

                    elif option == "TF-IDF":
                        anonymized_facts = data_preparator._anonymize_facts(
                            petitioner, respondent, facts)
                        predictions = predictor.predict_tf_idf(
                            anonymized_facts)
                        output = explainer.explain_instance(
                            anonymized_facts, predictor.predict_tf_idf)
                        important_words = output.as_list()

                    elif option == "1D Convolutional":
                        predictions = predictor.predict_cnn(facts)
                        output = explainer.explain_instance(
                            facts, predictor.predict_cnn)
                        important_words = output.as_list()

                    elif option == "GloVe":
                        predictions = predictor.predict_glove(facts)
                        output = explainer.explain_instance(
                            facts, predictor.predict_glove)
                        important_words = output.as_list()

                    elif option == "LSTM":
                        predictions = predictor.predict_lstm(facts)
                        output = explainer.explain_instance(
                            facts, predictor.predict_lstm)
                        important_words = output.as_list()

                    elif option == "BERT":
                        predictions = predictor.predict_bert(facts)

                    elif option == "FastText":
                        predictions = predictor.predict_fasttext(facts)

                    elif option == "Ensemble (Doc2Vec + TF-IDF)":
                        doc2vec_predictions = predictor.predict_doc2vec(facts)
                        tf_idf_predictions = predictor.predict_tf_idf(facts)

                        predictions = (doc2vec_predictions +
                                       tf_idf_predictions) / 2

                        doc2vec_output = explainer.explain_instance(
                            facts, predictor.predict_doc2vec)
                        doc2vec_important_words = doc2vec_output.as_list()

                        tf_idf_output = explainer.explain_instance(
                            facts, predictor.predict_tf_idf)
                        tf_idf_important_words = tf_idf_output.as_list()

                        important_words = doc2vec_important_words + tf_idf_important_words

                # displaying predictions
                col1, col2 = st.columns(2)

                with col1:
                    st.write("Percentage of petitioner winning:")
                    st.warning(f"{predictions[0, 0] * 100:.3f}%")

                with col2:
                    st.write("Percentage of respondent winning:")
                    st.info(f"{predictions[0, 1] * 100:.3f}%")

                st.write("Winning party:")
                if predictions[0, 0] > predictions[0, 1]:
                    st.success(petitioner)
                else:
                    st.success(respondent)

                # displaying highlighted words
                st.write(
                    '<p class="bold-text"> Top Words for Model\'s Decision: </p>', unsafe_allow_html=True)

                if option not in ["BERT", "FastText"]:
                    petitioner_words = [word for word,
                                        score in important_words if score < 0]
                    respondent_words = [word for word,
                                        score in important_words if score > 0]

                    for name in petitioner.split(" "):
                        if name in petitioner_words:
                            petitioner_words.remove(name)
                        elif name in respondent_words:
                            respondent_words.remove(name)

                    for name in respondent.split(" "):
                        if name in petitioner_words:
                            petitioner_words.remove(name)
                        elif name in respondent_words:
                            respondent_words.remove(name)

                    rendered_text = generate_highlighted_words(
                        facts, petitioner_words, respondent_words)

                    st.write(rendered_text, unsafe_allow_html=True)

                else:
                    st.warning(
                        "Sadly, this feature is not supported in BERT")

            else:
                st.error("Please, fill in all fields!")