import pandas as pd import matplotlib.pyplot as plt from sklearn.preprocessing import MultiLabelBinarizer from sklearn.feature_extraction.text import TfidfVectorizer from sklearn.model_selection import train_test_split from sklearn.linear_model import LogisticRegression from sklearn.multioutput import MultiOutputClassifier from sklearn.metrics import classification_report, f1_score, accuracy_score, hamming_loss import gradio as gr # Load dataset splits = {'train': 'simplified/train-00000-of-00001.parquet'} df = pd.read_parquet("hf://datasets/google-research-datasets/go_emotions/" + splits["train"]) emotion_labels = [ "admiration", "amusement", "anger", "annoyance", "approval", "caring", "confusion", "curiosity", "desire", "disappointment", "disapproval", "disgust", "embarrassment", "excitement", "fear", "gratitude", "grief", "joy", "love", "nervousness", "optimism", "pride", "realization", "relief", "remorse", "sadness", "surprise", "neutral" ] index_to_emotion = {i: label for i, label in enumerate(emotion_labels)} mlb = MultiLabelBinarizer(classes=range(28)) y = mlb.fit_transform(df['labels']) vectorizer = TfidfVectorizer(max_features=5000) X = vectorizer.fit_transform(df['text']) # Placeholder for trained model model = None metrics_report = "" def train_model(test_size=0.2, max_iter=1000, random_state=42): global model, metrics_report X_train, X_test, y_train, y_test = train_test_split( X, y, test_size=test_size, random_state=random_state ) model = MultiOutputClassifier(LogisticRegression(max_iter=max_iter)) model.fit(X_train, y_train) y_pred = model.predict(X_test) # Calculate standard classification report + other metrics report = classification_report( y_test, y_pred, target_names=[str(i) for i in range(28)] ) micro_f1 = f1_score(y_test, y_pred, average="micro") macro_f1 = f1_score(y_test, y_pred, average="macro") acc = accuracy_score(y_test, y_pred) hamming = hamming_loss(y_test, y_pred) metrics_summary = f""" Micro F1-score: {micro_f1:.4f} Macro F1-score: {macro_f1:.4f} Accuracy (Exact Match): {acc:.4f} Hamming Loss: {hamming:.4f} """ # Save the full report metrics_report = metrics_summary.strip() + "\n\n" + report return "Training Complete!" def predict_emotions(text): if model is None: return "Please train the model first.", "" vectorized = vectorizer.transform([text]) probas = model.predict_proba(vectorized) result = {} for i, emotion in enumerate(mlb.classes_): prob_class_1 = probas[i][0][1] result[emotion] = round(prob_class_1 * 100, 2) sorted_result = sorted(result.items(), key=lambda x: x[1], reverse=True) return sorted_result def predict_and_display(sentence): predictions = predict_emotions(sentence) if isinstance(predictions, str): return predictions, "" max_len = max(len(index_to_emotion[emo_id]) for emo_id, _ in predictions) result = "```" + "\nEmotion Predictions:\n\n" for emo_id, score in predictions: emo_name = index_to_emotion[emo_id] result += f"{emo_name.ljust(max_len)} → {score}%\n" result += "```" top_emotion = index_to_emotion[predictions[0][0]] return result, top_emotion # Gradio App with gr.Blocks(title="Interactive Emotion Detector", theme=gr.themes.Soft()) as demo: with gr.Tabs(): with gr.Tab("Emotion Detection"): gr.Markdown("## Emotion Detection") with gr.Row(): with gr.Column(): input_text = gr.Textbox( lines=3, placeholder="Enter a sentence...", label="Input Sentence" ) submit_btn = gr.Button("Analyze Emotion") with gr.Column(): output_text = gr.Markdown(label="Prediction Results") top_emotion = gr.Label(label="Top Emotion") submit_btn.click( fn=predict_and_display, inputs=input_text, outputs=[output_text, top_emotion] ) with gr.Tab("Dataset"): gr.Markdown("## Dataset Information") def dataset_info(): df = pd.read_parquet("hf://datasets/google-research-datasets/go_emotions/simplified/train-00000-of-00001.parquet") total_samples = len(df) emotions = sorted(set(e for label in df['labels'] for e in label)) emotion_names = [emotion_labels[i] for i in emotions] # Count distribution all_labels = [emotion_labels[i] for sublist in df['labels'] for i in sublist] label_counts = pd.Series(all_labels).value_counts().sort_index() label_df = pd.DataFrame({ "Emotion": label_counts.index, "Count": label_counts.values }) stats = f""" **Total Samples**: {total_samples} **Emotion Classes**: {', '.join(emotion_names)} """ return stats, label_df stats_display = gr.Markdown() dist_table = gr.Dataframe(headers=["Emotion", "Count"], interactive=False) load_btn = gr.Button("Load Dataset Info") load_btn.click(fn=dataset_info, inputs=[], outputs=[stats_display, dist_table]) with gr.Tab("EDA"): gr.Markdown("## Exploratory Data Analysis") eda_btn = gr.Button("Run EDA") eda_output = gr.Plot(label="EDA Output") def run_eda(): import matplotlib.pyplot as plt from collections import Counter import re # Define the label map inside the function label_map = [ 'admiration', 'amusement', 'anger', 'annoyance', 'approval', 'caring', 'confusion', 'curiosity', 'desire', 'disappointment', 'disapproval', 'disgust', 'embarrassment', 'excitement', 'fear', 'gratitude', 'grief', 'joy', 'love', 'nervousness', 'optimism', 'pride', 'realization', 'relief', 'remorse', 'sadness', 'surprise', 'neutral' ] fig, axs = plt.subplots(2, 2, figsize=(18, 10)) # Label distribution label_counts = df['labels'].explode().value_counts().sort_index() axs[0, 0].bar(label_map, label_counts) axs[0, 0].set_title("Label Frequency Distribution") axs[0, 0].tick_params(axis='x', rotation=45) # Labels per example df['num_labels'] = df['labels'].apply(len) df['num_labels'].value_counts().sort_index().plot(kind='bar', ax=axs[0, 1]) axs[0, 1].set_title("Number of Labels per Example") # Text length distribution df['text_length'] = df['text'].apply(len) df['text_length'].hist(bins=50, ax=axs[1, 0]) axs[1, 0].set_title("Distribution of Text Lengths") axs[1, 0].set_xlabel("Text Length (characters)") axs[1, 0].set_ylabel("Frequency") # Most common words all_words = " ".join(df['text']).lower() tokens = re.findall(r'\b\w+\b', all_words) common_words = Counter(tokens).most_common(20) words, freqs = zip(*common_words) axs[1, 1].bar(words, freqs) axs[1, 1].set_title("Top 20 Most Common Words") axs[1, 1].tick_params(axis='x', rotation=45) plt.tight_layout() return fig eda_btn.click(fn=run_eda, inputs=[], outputs=eda_output) with gr.Tab("Train Model"): gr.Markdown("## Train Your Emotion Model") test_size = gr.Slider(0.1, 0.5, step=0.05, value=0.2, label="Test Size") max_iter = gr.Slider(100, 5000, step=100, value=1000, label="Max Iterations") random_state = gr.Number(value=42, label="Random State") train_button = gr.Button("Train Model") train_status = gr.Textbox(label="Training Status") train_button.click( fn=train_model, inputs=[test_size, max_iter, random_state], outputs=train_status ) with gr.Tab("Results"): gr.Markdown("## Evaluation Metrics") results_output = gr.Markdown(label="Classification Report") def get_report(): return "```\n" + metrics_report + "\n```" refresh_btn = gr.Button("Refresh Report") refresh_btn.click( fn=get_report, inputs=[], outputs=results_output ) demo.launch()