app.py

import csv
import gradio as gr
import pandas as pd
from sentiment_analyser import RandomAnalyser, RoBERTaAnalyser, ChatGPTAnalyser
import matplotlib.pyplot as plt
from sklearn.metrics import ConfusionMatrixDisplay, confusion_matrix


def plot_bar(value_counts):
    fig, ax = plt.subplots(figsize=(6, 6))
    value_counts.plot.barh(ax=ax)
    ax.bar_label(ax.containers[0])
    plt.title('Frequency of Predictions')
    return fig


def plot_confusion_matrix(y_pred, y_true):
    cm = confusion_matrix(y_true, y_pred, normalize='true')
    fig, ax = plt.subplots(figsize=(6, 6))
    labels = []
    for label in SENTI_MAPPING.keys():
        if (label in y_pred.values) or (label in y_true.values):
            labels.append(label)
    disp = ConfusionMatrixDisplay(confusion_matrix=cm,
                                  display_labels=labels)
    disp.plot(cmap="Blues", values_format=".2f", ax=ax, colorbar=False)
    plt.title("Normalized Confusion Matrix")
    return fig


def classify(num: int):
    samples_df = df.sample(num)
    X = samples_df['Text'].tolist()
    y = samples_df['Label']
    roberta = MODEL_MAPPING[OUR_MODEL]
    y_pred = pd.Series(roberta.predict(X), index=samples_df.index)
    samples_df['Predict'] = y_pred
    bar = plot_bar(y_pred.value_counts())
    cm = plot_confusion_matrix(y_pred, y)
    plt.close()
    return samples_df, bar, cm


def analysis(Text):
    keys = []
    values = []
    for name, model in MODEL_MAPPING.items():
        keys.append(name)
        values.append(SENTI_MAPPING[model.predict([Text])[0]])
    return pd.DataFrame([values], columns=keys)


def analyse_file(file):
    output_name = 'output.csv'
    with open(output_name, mode='w', newline='') as output:
        writer = csv.writer(output)
        header = ['Text', 'Label']
        writer.writerow(header)
        model = MODEL_MAPPING[OUR_MODEL]
        with open(file.name) as f:
            for line in f:
                text = line[:-1]
                sentiment = model.predict([text])
                writer.writerow([text, sentiment[0]])
    return output_name


MODEL_MAPPING = {
    'Random': RandomAnalyser(),
    'RoBERTa': RoBERTaAnalyser(),
    'ChatGPT': RandomAnalyser(),
}

OUR_MODEL = 'RoBERTa'

SENTI_MAPPING = {
    'negative': '😭',
    'neutral': '😶',
    'positive': '🥰'
}

TITLE = "Sentiment Analysis on Software Engineer Texts"

DESCRIPTION = {
    'en': (
        "This is the demo page for our model: "
        "[Cloudy1225/stackoverflow-roberta-base-sentiment]"
        "(https://huggingface.co/Cloudy1225/stackoverflow-roberta-base-sentiment)."
    ),
    'zh': (
        "这里是第16组“睿王和他的五个小跟班”软工三迭代三模型演示页面。"
        "模型链接：[Cloudy1225/stackoverflow-roberta-base-sentiment]"
        "(https://huggingface.co/Cloudy1225/stackoverflow-roberta-base-sentiment)."
    )
}

PROMPT1 = {
    'en': (
        "Enter text in the left text box and press Enter, and the sentiment analysis results will be output on the right. "
        "Here, we present three types of results, which come from random, our model, and ChatGPT."
    ),
    'zh': (
        "在左侧文本框中输入文本并按回车键，右侧将输出情感分析结果。"
        "这里我们展示了三种结果，分别是随机结果、模型结果和 ChatGPT 结果。"
    )
}

PROMPT2 = {
    'en': (
        "Upload a txt/csv file in the left file box, and the model will perform sentiment analysis on each line of the input text. "
        "You can download the output file on the right. "
        "The output file will be in CSV format with two columns: the original text, and the classification results."
    ),
    'zh': (
        "在左侧文件框中上传 txt/csv 文件，模型会对输入文本的每一行当作一个文本进行情感分析。"
        "可以在右侧下载输出文件，输出文件为两列 csv 格式，第一列为原始文本，第二列为分类结果。"
    )
}

PROMPT3 = {
    'en': (
        "Here we evaluate our model on the StackOverflow4423 dataset. "
        "Sliding the slider will sample a specified number of samples from the StackOverflow4423 dataset and predict their sentiment labels. "
        "Based on the prediction results, a label distribution chart and a confusion matrix will be plotted."
    ),
    'zh': (
        "这里是在 StackOverflow4423 数据集上评估我们的模型。"
        "滑动 Slider，将会从 StackOverflow4423 数据集中抽样出指定数量的样本，预测其情感标签。"
        "并根据预测结果绘制标签分布图和混淆矩阵。"
    )
}

DEFAULT_LANG = 'en'

MAX_SAMPLES = 64

df = pd.read_csv('./SOF4423.csv')


def set_language(lang):
    return DESCRIPTION[lang], PROMPT1[lang], PROMPT2[lang], PROMPT3[lang]


with gr.Blocks(title=TITLE) as demo:
    with gr.Row():
        with gr.Column():
            gr.HTML(f"<H1>{TITLE}</H1>")
        with gr.Column():
            language_selector = gr.Radio(
                ['en', 'zh'], label="Select Language", value=DEFAULT_LANG,
                interactive=True, show_label=False, container=False
            )

    description = gr.Markdown(DESCRIPTION[DEFAULT_LANG])
    gr.HTML("<H2>Model Inference</H2>")
    prompt1 = gr.Markdown(PROMPT1[DEFAULT_LANG])
    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(label='Input',
                                    placeholder="Enter a positive or negative sentence here...")
        with gr.Column():
            senti_output = gr.Dataframe(type="pandas", value=[['😋', '😋', '😋']],
                                        headers=list(MODEL_MAPPING.keys()), interactive=False)
    text_input.submit(analysis, inputs=text_input, outputs=senti_output, show_progress='full')

    prompt2 = gr.Markdown(PROMPT2[DEFAULT_LANG])
    with gr.Row():
        with gr.Column():
            file_input = gr.File(label='File',
                                 file_types=['.txt', '.csv'])
        with gr.Column():
            file_output = gr.File(label='Output')
    file_input.upload(analyse_file, inputs=file_input, outputs=file_output)

    gr.HTML("<H2>Model Evaluation</H2>")
    prompt3 = gr.Markdown(PROMPT3[DEFAULT_LANG])
    input_models = list(MODEL_MAPPING)
    input_n_samples = gr.Slider(
        minimum=4,
        maximum=MAX_SAMPLES,
        value=8,
        step=4,
        label='Number of samples'
    )

    with gr.Row():
        with gr.Column():
            bar_plot = gr.Plot(label='Predictions Frequency')
        with gr.Column():
            cm_plot = gr.Plot(label='Confusion Matrix')

    with gr.Row():
        dataframe = gr.Dataframe(type="pandas", wrap=True, headers=['Text', 'Label', 'Predict'])

    input_n_samples.change(fn=classify, inputs=input_n_samples, outputs=[dataframe, bar_plot, cm_plot])

    language_selector.change(fn=set_language, inputs=language_selector,
                             outputs=[description, prompt1, prompt2, prompt3])

    demo.launch()