lyrical-classification / src /streamlit_app.py
bkusuma's picture
Update src/streamlit_app.py
6e1ba52 verified
Raw History Blame Contribute Delete
5.72 kB
import streamlit as st
import pandas as pd
import numpy as np
"""
# Welcome to Streamlit!
Edit `/streamlit_app.py` to customize this app to your heart's desire :heart:.
If you have any questions, checkout our [documentation](https://docs.streamlit.io) and [community
forums](https://discuss.streamlit.io).
"""
# --- Helper Functions ---
def mock_predict(lyrics):
"""
Mock prediction function for demo purposes.
In a production environment, this would load a trained .pkl model
(e.g., Logistic Regression or XGBoost) and apply TF-IDF vectorization.
"""
lyrics_lower = lyrics.lower()
if any(word in lyrics_lower for word in ["love", "baby", "heart", "dance"]):
return "Pop"
elif any(word in lyrics_lower for word in ["blood", "death", "kill", "dark"]):
return "Metal"
elif any(word in lyrics_lower for word in ["truck", "beer", "dirt", "whiskey"]):
return "Country"
elif any(word in lyrics_lower for word in ["rhyme", "beat", "flow", "street"]):
return "Hip Hop"
else:
return np.random.choice(["Rock", "Pop", "Indie", "Jazz"])
# --- Main App Configuration ---
st.set_page_config(page_title="Lyric Genre Classifier", page_icon="🎡")
# --- Sidebar Navigation ---
st.sidebar.title("Navigation")
st.sidebar.write("Explore the project:")
page = st.sidebar.radio("Go to", ["Project Overview", "Data Exploration", "Model Results", "Interactive Demo"])
st.sidebar.markdown("---")
st.sidebar.write("**Author:** Data Science Portfolio")
st.sidebar.write("**Tech Stack:** Python, Pandas, Scikit-Learn, XGBoost, Streamlit")
# --- Page 1: Project Overview ---
if page == "Project Overview":
st.title("🎡 Lyric Frequency Genre Classifier")
st.write("""
### Welcome to my project!
This application attempts to classify the musical genre of a song based purely on the text of its lyrics.
By treating lyrics as unstructured text data, we can apply Natural Language Processing (NLP) techniques
like Bag-of-Words and TF-IDF to find patterns in how different genres use vocabulary.
""")
st.markdown("""
**Project Goals:**
* Clean and preprocess thousands of song lyrics.
* Perform Exploratory Data Analysis (EDA) to find distinctive words per genre.
* Train machine learning classification models (Logistic Regression, XGBoost, HistGradientBoosting).
* Evaluate model accuracy and build a simple interface.
""")
st.info("πŸ‘ˆ Use the sidebar to navigate through the data, model results, and try the demo!")
# --- Page 2: Data Exploration ---
elif page == "Data Exploration":
st.title("πŸ“Š Exploratory Data Analysis")
st.write("Before modeling, I explored the dataset to understand the class balance and vocabulary distribution.")
st.subheader("Genre Distribution")
st.write("This chart shows the balance of our target classes across the dataset.")
# Mock data for demonstration purposes
genre_counts = pd.DataFrame({
"Genre": ["Rock", "Pop", "Hip Hop", "Metal", "Country"],
"Song Count": [1250, 1100, 950, 800, 750]
}).set_index("Genre")
st.bar_chart(genre_counts)
st.markdown("---")
st.subheader("Distinctive Words by Genre")
st.write("Using frequency counts and TF-IDF, I identified words that heavily index toward specific genres.")
# Mock top words dataframe
top_words = pd.DataFrame({
"Genre": ["Rock", "Pop", "Hip Hop", "Metal", "Country"],
"Top Distinctive Words": ["guitar, band, night, roll", "love, baby, dance, tonight", "beat, flow, street, money", "death, blood, dark, soul", "truck, whiskey, dirt, road"]
})
st.table(top_words)
# --- Page 3: Model Results ---
elif page == "Model Results":
st.title("πŸ“ˆ Model Performance")
st.write("""
I experimented with several models to classify the text data.
The primary metrics used for evaluation were Accuracy and F1-Score (due to slight class imbalances).
""")
st.subheader("Classification Metrics")
# Mock metrics dataframe
metrics = pd.DataFrame({
"Model": ["Logistic Regression", "HistGradientBoosting", "XGBoost"],
"Accuracy": [0.72, 0.78, 0.81],
"Weighted F1-Score": [0.71, 0.76, 0.80]
})
st.dataframe(metrics, use_container_width=True, hide_index=True)
st.subheader("Key Takeaways")
st.markdown("""
* **XGBoost** handled the sparse text matrices well but required careful hyperparameter tuning.
* **Logistic Regression** served as a surprisingly strong and interpretable baseline.
* Genres with highly specific jargon (like Metal or Country) were easier to classify than broad categories (like Pop/Rock) which often shared overlapping vocabulary.
""")
# --- Page 4: Interactive Demo ---
elif page == "Interactive Demo":
st.title("🎀 Live Genre Predictor")
st.write("Paste some lyrics below to see what genre the model predicts!")
# Input text area
user_lyrics = st.text_area("Enter lyrics here:", height=200, placeholder="e.g., I'm driving my truck down the dirt road...")
if st.button("Predict Genre"):
if user_lyrics.strip() == "":
st.error("Please enter some lyrics first.")
else:
with st.spinner("Analyzing lyric frequencies..."):
# Call mock prediction
prediction = mock_predict(user_lyrics)
st.success(f"**Predicted Genre:** {prediction}")
st.write("*Note: This is a lightweight demo version utilizing a mock predictor. Full model inferences and vectorizers are available in the repository notebooks.*")