app_stress.py

# ============================================================
#  HuggingFace Space — Stress Level Classifier Demo
#  File: app.py  (deploy as a Gradio Space)
# ============================================================
# How to deploy:
#   1. Go to https://huggingface.co/spaces → Create new Space
#   2. Choose SDK: Gradio
#   3. Upload this file as app.py + requirements.txt
#   4. Space auto-builds and launches
#
# requirements.txt:
#   torch>=2.1.0
#   gradio>=4.0.0
#   huggingface_hub>=0.22.0
#   numpy>=1.24.0
# ============================================================

import json
import zipfile
import shutil
import tempfile
import numpy as np
import torch
import torch.nn as nn
import gradio as gr
from huggingface_hub import hf_hub_download
from pathlib import Path

# ── Config ────────────────────────────────────────────────────────────────────
MODEL_REPO = "your-hf-username/stress-level-classifier-mlp"   # ← update this

FEATURE_COLS = [
    "anxiety_level", "self_esteem", "mental_health_history", "depression",
    "headache", "blood_pressure", "sleep_quality", "breathing_problem",
    "noise_level", "living_conditions", "safety", "basic_needs",
    "academic_performance", "study_load", "teacher_student_relationship",
    "future_career_concerns", "social_support", "peer_pressure",
    "extracurricular_activities", "bullying",
]

FEATURE_RANGES = {
    "anxiety_level":               (0, 21),
    "self_esteem":                 (0, 30),
    "mental_health_history":       (0, 1),
    "depression":                  (0, 27),
    "headache":                    (0, 5),
    "blood_pressure":              (1, 3),
    "sleep_quality":               (1, 5),
    "breathing_problem":           (1, 5),
    "noise_level":                 (0, 5),
    "living_conditions":           (1, 5),
    "safety":                      (1, 5),
    "basic_needs":                 (1, 5),
    "academic_performance":        (1, 5),
    "study_load":                  (1, 5),
    "teacher_student_relationship":(1, 5),
    "future_career_concerns":      (1, 5),
    "social_support":              (1, 5),
    "peer_pressure":               (1, 5),
    "extracurricular_activities":  (0, 5),
    "bullying":                    (1, 5),
}

STRESS_EMOJIS  = {"low": "🟢", "medium": "🟡", "high": "🔴"}
STRESS_LABELS  = ["low", "medium", "high"]
ID2LABEL       = {"0": "low", "1": "medium", "2": "high"}

# ── Model definition (must match training) ────────────────────────────────────
class StressMLP(nn.Module):
    def __init__(self, input_dim, num_classes, dropout=0.3):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(input_dim, 256), nn.BatchNorm1d(256), nn.ReLU(), nn.Dropout(dropout),
            nn.Linear(256, 128),       nn.BatchNorm1d(128), nn.ReLU(), nn.Dropout(dropout),
            nn.Linear(128, 64),        nn.BatchNorm1d(64),  nn.ReLU(), nn.Dropout(dropout / 2),
            nn.Linear(64, num_classes),
        )

    def forward(self, x):
        return self.net(x)

# ── Load model from Hub ───────────────────────────────────────────────────────
print(f"Loading model from {MODEL_REPO} …")

cfg_path     = hf_hub_download(MODEL_REPO, "config.json")
scaler_path  = hf_hub_download(MODEL_REPO, "scaler.json")
weights_path = hf_hub_download(MODEL_REPO, "model_weights.pt")

with open(cfg_path)    as f: cfg    = json.load(f)
with open(scaler_path) as f: scaler = json.load(f)

MEAN  = np.array(scaler["mean"],  dtype=np.float32)
SCALE = np.array(scaler["scale"], dtype=np.float32)

model = StressMLP(cfg["input_dim"], cfg["num_classes"], cfg["dropout"])
model.load_state_dict(torch.load(weights_path, map_location="cpu"))
model.eval()
print("Model loaded ✅")

# ── Build downloadable zip on startup ────────────────────────────────────────
DOWNLOAD_ZIP = Path("/tmp/stress_model_package.zip")

def build_download_zip():
    """Package the cached HF model files into a single zip for users."""
    if DOWNLOAD_ZIP.exists():
        return
    files_to_zip = {
        "model_weights.pt": weights_path,
        "config.json":      cfg_path,
        "scaler.json":      scaler_path,
    }
    with zipfile.ZipFile(DOWNLOAD_ZIP, "w", zipfile.ZIP_DEFLATED) as zf:
        for arcname, src in files_to_zip.items():
            zf.write(src, arcname=arcname)
    print(f"Download zip ready: {DOWNLOAD_ZIP}")

build_download_zip()

# ── Inference ─────────────────────────────────────────────────────────────────
def predict(*feature_values):
    arr = np.array(feature_values, dtype=np.float32).reshape(1, -1)
    arr = (arr - MEAN) / SCALE

    with torch.no_grad():
        logits = model(torch.tensor(arr))
        probs  = torch.softmax(logits, dim=-1).squeeze().numpy()

    pred_id    = int(probs.argmax())
    pred_label = STRESS_LABELS[pred_id]
    emoji      = STRESS_EMOJIS[pred_label]
    confidence = probs[pred_id] * 100

    result  = f"## {emoji} Stress Level: **{pred_label.upper()}**\n"
    result += f"Confidence: **{confidence:.1f}%**"

    scores = {
        f"{STRESS_EMOJIS[lbl]} {lbl.capitalize()}": float(probs[i]) * 100
        for i, lbl in enumerate(STRESS_LABELS)
    }
    return result, scores

def get_download():
    return str(DOWNLOAD_ZIP)

# ── Build sliders from feature ranges ────────────────────────────────────────
def make_label(col):
    return col.replace("_", " ").title()

sliders = [
    gr.Slider(
        minimum=lo, maximum=hi, value=(lo + hi) // 2, step=1,
        label=make_label(col),
    )
    for col, (lo, hi) in FEATURE_RANGES.items()
]

# ── Gradio UI ─────────────────────────────────────────────────────────────────
with gr.Blocks(
    theme=gr.themes.Soft(),
    title="Stress Level Classifier",
    css="""
        .result-md { font-size: 1.35rem; padding: 14px 18px;
                     border-radius: 10px; background: var(--block-background-fill); }
        .dl-btn { margin-top: 8px; }
        footer { display: none !important; }
    """,
) as demo:

    gr.Markdown(
        """
        # 🧠 Student Stress Level Classifier
        Predicts stress level (**Low / Medium / High**) from 20 psychosocial &
        physiological indicators using a fine-tuned MLP.
        """
    )

    with gr.Row():
        # ── Left: inputs ──────────────────────────────────────────────────────
        with gr.Column(scale=3):
            gr.Markdown("### 📋 Input Features")

            with gr.Accordion("🧬 Psychological Factors", open=True):
                s_anxiety  = sliders[0]
                s_esteem   = sliders[1]
                s_mh_hist  = sliders[2]
                s_depress  = sliders[3]

            with gr.Accordion("🩺 Physical Indicators", open=False):
                s_headache = sliders[4]
                s_bp       = sliders[5]
                s_sleep    = sliders[6]
                s_breath   = sliders[7]

            with gr.Accordion("🏠 Environmental Factors", open=False):
                s_noise    = sliders[8]
                s_living   = sliders[9]
                s_safety   = sliders[10]
                s_basic    = sliders[11]

            with gr.Accordion("🎓 Academic Factors", open=False):
                s_acad     = sliders[12]
                s_study    = sliders[13]
                s_teacher  = sliders[14]
                s_career   = sliders[15]

            with gr.Accordion("👥 Social Factors", open=False):
                s_social   = sliders[16]
                s_peer     = sliders[17]
                s_extra    = sliders[18]
                s_bully    = sliders[19]

            predict_btn = gr.Button("🔍 Predict Stress Level", variant="primary", size="lg")

        # ── Right: outputs ────────────────────────────────────────────────────
        with gr.Column(scale=2):
            gr.Markdown("### 📊 Results")
            result_md  = gr.Markdown(
                value="*Adjust sliders and click Predict.*",
                elem_classes=["result-md"],
            )
            scores_out = gr.Label(
                label="Confidence per Class (%)",
                num_top_classes=3,
            )

            gr.Markdown("---")
            gr.Markdown("### 📥 Download Model")
            gr.Markdown(
                "Download all model files (weights, config, scaler) as a single ZIP."
            )
            download_btn = gr.DownloadButton(
                label="⬇️ Download stress_model_package.zip",
                value=get_download,
                variant="secondary",
                elem_classes=["dl-btn"],
            )
            gr.Markdown(
                f"Or browse individual files on the "
                f"[HuggingFace model page 🤗](https://huggingface.co/{MODEL_REPO})."
            )

    # ── Examples ──────────────────────────────────────────────────────────────
    gr.Markdown("---")
    gr.Examples(
        examples=[
            # anxiety, esteem, mh_hist, depress, headache, bp, sleep, breath,
            # noise, living, safety, basic, acad, study, teacher, career,
            # social, peer, extra, bully
            [4,  26, 0, 6,  1, 2, 4, 1, 1, 4, 4, 4, 5, 1, 4, 1, 3, 2, 2, 1],  # Low stress
            [13, 22, 1, 12, 3, 1, 2, 4, 3, 3, 3, 3, 3, 3, 2, 3, 3, 3, 2, 2],  # Medium stress
            [20, 3,  1, 22, 5, 3, 1, 5, 3, 1, 1, 1, 1, 5, 2, 5, 1, 5, 4, 5],  # High stress
        ],
        inputs=sliders,
        label="Quick examples (Low / Medium / High stress)",
    )

    gr.Markdown(
        """
        ---
        **Features:** 20 psychosocial & physiological indicators  ·
        **Model:** PyTorch MLP fine-tuned for 10 epochs  ·
        **Classes:** 🟢 Low · 🟡 Medium · 🔴 High
        """
    )

    predict_btn.click(
        fn=predict,
        inputs=sliders,
        outputs=[result_md, scores_out],
    )

if __name__ == "__main__":
    demo.launch()