import pickle
import numpy as np
import torch
import torch.nn as nn
import gradio as gr
# ══════════════════════════════════════════════════════════════════════════════
# PART 1 — MODEL ARCHITECTURE (must match training)
# ══════════════════════════════════════════════════════════════════════════════
class MultiTaskNet(nn.Module):
def __init__(self, n_features: int, n_classes: int = 3,
dropout: float = 0.25):
super().__init__()
self.backbone = nn.Sequential(
nn.Linear(n_features, 128),
nn.BatchNorm1d(128),
nn.ReLU(),
nn.Dropout(dropout),
nn.Linear(128, 64),
nn.BatchNorm1d(64),
nn.ReLU(),
nn.Dropout(dropout),
nn.Linear(64, 32),
nn.ReLU(),
)
self.clf_head = nn.Linear(32, n_classes)
self.reg_head = nn.Linear(32, 1)
def forward(self, x):
shared = self.backbone(x)
return self.clf_head(shared), self.reg_head(shared)
# ══════════════════════════════════════════════════════════════════════════════
# PART 2 — SKLEARN-COMPATIBLE WRAPPERS
# These mimic the exact API the UI expects:
# apnea_model.predict(row)
# apnea_model.predict_proba(row)
# stress_model.predict(row)
# ══════════════════════════════════════════════════════════════════════════════
class _ApneaWrapper:
"""
Wraps the classification head of MultiTaskNet.
Input row has 12 columns (with Stress Level placeholder at index 6).
We drop that column before inference since the DL model uses 11 features.
"""
def __init__(self, net, scaler):
self.net = net
self.scaler = scaler
def _prepare(self, X):
X = np.array(X, dtype=np.float32)
if X.ndim == 1:
X = X.reshape(1, -1)
# Drop Stress Level column (index 6) — the UI always passes it
if X.shape[1] == 12:
X = np.delete(X, 6, axis=1) # now 11 features
X = self.scaler.transform(X).astype(np.float32)
return torch.tensor(X)
def predict(self, X):
self.net.eval()
with torch.no_grad():
logits, _ = self.net(self._prepare(X))
return logits.argmax(dim=1).numpy()
def predict_proba(self, X):
self.net.eval()
with torch.no_grad():
logits, _ = self.net(self._prepare(X))
return torch.softmax(logits, dim=1).numpy()
class _StressWrapper:
"""
Wraps the regression head of MultiTaskNet.
Same column-drop logic as _ApneaWrapper.
"""
def __init__(self, net, scaler):
self.net = net
self.scaler = scaler
def _prepare(self, X):
X = np.array(X, dtype=np.float32)
if X.ndim == 1:
X = X.reshape(1, -1)
if X.shape[1] == 12:
X = np.delete(X, 6, axis=1)
X = self.scaler.transform(X).astype(np.float32)
return torch.tensor(X)
def predict(self, X):
self.net.eval()
with torch.no_grad():
_, stress = self.net(self._prepare(X))
return stress.squeeze().numpy()
# ══════════════════════════════════════════════════════════════════════════════
# PART 3 — LOAD MODEL (replaces pickle.load)
# ══════════════════════════════════════════════════════════════════════════════
with open('encoders.pkl', 'rb') as f:
enc = pickle.load(f)
occ_le = enc['occ_le']
scaler = enc['scaler']
n_features = enc['n_features'] # 11
_net = MultiTaskNet(n_features=n_features)
_net.load_state_dict(torch.load('model.pth', map_location='cpu'))
_net.eval()
# Drop-in replacements — same names the UI uses
apnea_model = _ApneaWrapper(_net, scaler)
stress_model = _StressWrapper(_net, scaler)
print('✅ Deep learning model loaded successfully')
# ══════════════════════════════════════════════════════════════════════════════
# PART 4 — METADATA (unchanged)
# ══════════════════════════════════════════════════════════════════════════════
CLASS_NAMES = ['None', 'Insomnia', 'Sleep Apnea']
OCCUPATION_LIST = [
'Accountant', 'Doctor', 'Engineer', 'Lawyer', 'Manager',
'Nurse', 'Sales Representative', 'Salesperson',
'Scientist', 'Software Engineer', 'Teacher'
]
BMI_MAP = {'Normal': 0, 'Overweight': 1, 'Obese': 2}
BMI_LIST = ['Normal', 'Overweight', 'Obese']
BENCHMARKS = {
'18-30': {'sleep_duration': 7.4, 'quality': 7.1, 'steps': 8200, 'hr': 68},
'31-45': {'sleep_duration': 7.1, 'quality': 6.8, 'steps': 7100, 'hr': 71},
'46-65': {'sleep_duration': 6.8, 'quality': 6.3, 'steps': 6000, 'hr': 73},
}
def get_benchmark(age):
if age <= 30: return BENCHMARKS['18-30']
elif age <= 45: return BENCHMARKS['31-45']
else: return BENCHMARKS['46-65']
def run_bias_audit():
groups = {
'Young Male': [25, 1, 2, 7, 7, 40, 0, 120, 80, 70, 6000],
'Young Female': [25, 0, 2, 7, 7, 40, 0, 120, 80, 70, 6000],
'Older Male': [55, 1, 2, 7, 7, 40, 0, 130, 85, 75, 5000],
'Older Female': [55, 0, 2, 7, 7, 40, 0, 130, 85, 75, 5000],
'Overweight': [35, 1, 2, 7, 7, 40, 1, 125, 82, 72, 5500],
'Obese': [35, 1, 2, 7, 7, 40, 2, 135, 88, 78, 4500],
}
results = {}
for name, features in groups.items():
# 11-feature row (no stress placeholder needed here)
row = np.array([features])
pred = apnea_model.predict(row)[0]
proba = apnea_model.predict_proba(row)[0]
results[name] = {
'prediction': CLASS_NAMES[pred],
'confidence': round(max(proba) * 100, 1)
}
return results
# ══════════════════════════════════════════════════════════════════════════════
# PART 5 — PREDICT FUNCTION (signature 100% unchanged)
# ══════════════════════════════════════════════════════════════════════════════
def predict(
age, gender, occupation,
sleep_duration, quality_of_sleep,
physical_activity, bmi_category,
bp_systolic, bp_diastolic,
heart_rate, daily_steps
):
gender_enc = 1 if gender == 'Male' else 0
occ_enc = OCCUPATION_LIST.index(occupation)
bmi_enc = BMI_MAP[bmi_category]
bm = get_benchmark(age)
sleep_delta = sleep_duration - bm['sleep_duration']
# stress_est is still computed for the HTML display heuristic;
# the model no longer uses it as input (no leakage)
stress_est = 8 - (quality_of_sleep * 0.6) - (physical_activity / 30) + (1 if bmi_enc >= 1 else 0)
stress_est = int(np.clip(round(stress_est), 3, 8))
# 11-feature row — Stress Level column NOT included
row = np.array([[
age, gender_enc, occ_enc,
sleep_duration, quality_of_sleep,
physical_activity,
bmi_enc, bp_systolic, bp_diastolic,
heart_rate, daily_steps
]])
# ── Sleep Apnea Prediction ────────────────────────────────────────────────
apnea_idx = apnea_model.predict(row)[0]
apnea_proba = apnea_model.predict_proba(row)[0] if hasattr(apnea_model, 'predict_proba') else [0, 0, 0]
apnea_label = CLASS_NAMES[apnea_idx]
risk_map = {
'None': ('No Risk', 'LOW'),
'Insomnia': ('Insomnia Risk', 'MEDIUM'),
'Sleep Apnea': ('Sleep Apnea Risk', 'HIGH'),
}
result_text, risk_level = risk_map[apnea_label]
# ── Stress Prediction ─────────────────────────────────────────────────────
stress_raw = stress_model.predict(row)
stress_score = int(np.clip(round(float(stress_raw)), 3, 8))
stress_label = 'Low' if stress_score <= 4 else ('Medium' if stress_score <= 6 else 'High')
stress_pct = int(stress_score / 8 * 100)
stress_col = '#1a9e75' if stress_score <= 4 else ('#e5a020' if stress_score <= 6 else '#e05252')
stress_bg = '#f0fdf4' if stress_score <= 4 else ('#fffbeb' if stress_score <= 6 else '#fef2f2')
# ── Overall Sleep Score ───────────────────────────────────────────────────
sleep_score = int(np.clip(
(quality_of_sleep / 10 * 100) * 0.35 +
(min(sleep_duration, 9) / 9 * 100) * 0.30 +
((8 - stress_score) / 8 * 100) * 0.20 +
(physical_activity / 90 * 100) * 0.15,
0, 100
))
if sleep_score >= 80:
score_msg = "That's a great score!"
score_sub = "You're well on your way to improving your health."
elif sleep_score >= 60:
score_msg = 'Moderate score.'
score_sub = 'Some areas need attention for better sleep health.'
else:
score_msg = 'Low score.'
score_sub = 'Please follow the recommendations below carefully.'
dur_score = int(min(sleep_duration / 9 * 70, 70))
qual_score = int(quality_of_sleep * 2)
activity_pts = int(min(physical_activity / 90 * 5, 5))
stress_pts = int(max(0, (8 - stress_score) / 5 * 5))
none_pct = round(apnea_proba[0] * 100, 1)
ins_pct = round(apnea_proba[1] * 100, 1)
apnea_pct = round(apnea_proba[2] * 100, 1)
rcolour = {'LOW': '#1a9e75', 'MEDIUM': '#e5a020', 'HIGH': '#e05252'}[risk_level]
ring_col = '#1a9e75' if sleep_score >= 80 else ('#e5a020' if sleep_score >= 60 else '#e05252')
R = 54
circ = 2 * 3.14159 * R
dash = circ * sleep_score / 100
gap = circ - dash
# ════════════════════════════════════════
# Sleep Apnea Score Card HTML
# ════════════════════════════════════════
apnea_result_html = f"""
NeuroRest Score
{sleep_score}
100 points
{score_msg}
{score_sub}
{sleep_duration:.1f}h sleep duration
{dur_score} / 70
Sleep quality rating
{qual_score} / 20
{apnea_label} detected
{activity_pts} / 5
Stress — {stress_label}
{stress_pts} / 5
"""
# ════════════════════════════════════════
# Stress Output HTML
# ════════════════════════════════════════
stress_output = f"""
Stress Level
{stress_score}/8
Category: {stress_label}
"""
# ════════════════════════════════════════
# Recommendations HTML
# ════════════════════════════════════════
recs = []
if sleep_duration < bm['sleep_duration']:
recs.append(('🛌', 'Increase Sleep Duration',
f"You sleep {sleep_duration:.1f}h vs the {bm['sleep_duration']}h benchmark for your age. "
f"Aim for 7–9 hours nightly."))
if quality_of_sleep < bm['quality']:
recs.append(('🌙', 'Improve Sleep Quality',
'Maintain a consistent sleep schedule, keep your room cool and dark.'))
if stress_score >= 6:
recs.append(('🧘', 'Manage Stress',
'High stress detected. Consider meditation, journaling, or speaking with a professional.'))
if physical_activity < 30:
recs.append(('🏃', 'Increase Physical Activity',
'At least 30 minutes of moderate exercise daily improves sleep quality significantly.'))
if bmi_enc >= 1:
recs.append(('⚖️', 'Monitor BMI',
'Elevated BMI increases sleep apnea risk. Consider dietary adjustments.'))
if daily_steps < bm['steps']:
recs.append(('👟', 'Increase Daily Steps',
f"You average {int(daily_steps):,} steps vs benchmark {bm['steps']:,}. "
f"Aim for 7,000–10,000 steps."))
if bp_systolic > 130 or bp_diastolic > 85:
recs.append(('❤️', 'Monitor Blood Pressure',
'Elevated BP detected. Reduce sodium, increase potassium, and consult your doctor.'))
if not recs:
recs.append(('✅', 'Great Health Indicators',
'Your metrics are within healthy ranges. Keep maintaining your current lifestyle!'))
rec_items = ''.join(f"""
""" for icon, title, body in recs)
rec_html = f"""
"""
# ════════════════════════════════════════
# AI Coaching HTML
# ════════════════════════════════════════
coaching_tips = []
if apnea_label == 'Sleep Apnea':
coaching_tips.append(('🩺', 'Consult a Sleep Specialist',
'Sleep Apnea detected. A polysomnography test can confirm and guide CPAP therapy.'))
if apnea_label == 'Insomnia':
coaching_tips.append(('🧠', 'Try CBT-I',
'Cognitive Behavioural Therapy for Insomnia is the most effective non-drug treatment.'))
if stress_score >= 5:
coaching_tips.append(('🌿', 'Wind-down Routine',
'Start a 30-minute pre-sleep routine: dim lights, no screens, light stretching.'))
coaching_tips.append(('📱', 'Track Your Sleep',
'Use a wearable or sleep-tracking app to monitor trends over 2–4 weeks.'))
coaching_tips.append(('☀️', 'Morning Light Exposure',
'Get 10–15 min of natural light within an hour of waking to regulate your circadian rhythm.'))
tips_html = ''.join(f"""
""" for icon, title, body in coaching_tips)
coaching_html = f"""
🤖
AI Sleep Coach
Personalized insights based on your vitals
{tips_html}
💬 Re-analyze weekly to get updated coaching as your health data changes.
"""
return apnea_result_html, stress_output, rec_html, coaching_html
# ══════════════════════════════════════════════════════════════════════════════
# PART 6 — GRADIO UI (100% original — zero changes)
# ══════════════════════════════════════════════════════════════════════════════
CSS = """
@import url('https://fonts.googleapis.com/css2?family=DM+Sans:opsz,wght@9..40,400;9..40,500;9..40,600;9..40,700&display=swap');
body, .gradio-container {
background: #eef3f8 !important;
font-family: 'DM Sans', system-ui, sans-serif !important;
}
.gradio-container { max-width: 1340px !important; margin: 0 auto !important; }
.gap, .form, .block, .contain,
.gradio-group, .gr-group, .gr-box,
div.svelte-vt1mxs, div.svelte-1gfkn6j,
.wrap.svelte-z7cif2, .padded {
background: transparent !important;
border: none !important;
box-shadow: none !important;
}
#nr-header {
background: linear-gradient(135deg, #0d1b2a 0%, #123d5e 55%, #0f5c42 100%);
border-radius: 18px; padding: 26px 36px; margin-bottom: 20px;
box-shadow: 0 4px 24px rgba(13,27,42,0.2);
}
#nr-header h1 {
font-size: 1.8rem !important; font-weight: 700 !important;
color: white !important; margin: 0 0 4px !important; letter-spacing: -0.4px;
}
#nr-header p { color: rgba(255,255,255,0.58) !important; font-size: 0.87rem !important; margin: 0 !important; }
#nr-dot {
display: inline-block; width: 9px; height: 9px;
background: #2ecc8f; border-radius: 50%; margin-right: 9px;
animation: blink 2s ease-in-out infinite;
}
@keyframes blink { 0%,100%{opacity:1} 50%{opacity:0.25} }
.nr-lbl {
font-size: 0.67rem !important; font-weight: 700 !important;
letter-spacing: 1.8px !important; text-transform: uppercase !important;
color: #1a6b8a !important; margin-bottom: 8px !important; padding-left: 2px !important;
}
.nr-card {
background: white !important;
border-radius: 16px !important;
border: 1px solid #d1e2ed !important;
padding: 20px 22px !important;
box-shadow: 0 2px 12px rgba(15,82,110,0.07) !important;
margin-bottom: 14px !important;
}
input[type=range] {
accent-color: #1a6b8a !important;
height: 4px !important;
cursor: pointer !important;
}
.nr-card label > span,
.nr-card .block > label > span,
label > span {
font-size: 0.8rem !important;
font-weight: 600 !important;
color: #1a4b6a !important;
letter-spacing: 0.1px !important;
}
.nr-card input[type=number] {
background: #f0f7fb !important;
border: 1px solid #c4d9e8 !important;
border-radius: 8px !important;
color: #1a6b8a !important;
font-weight: 700 !important;
font-size: 0.85rem !important;
text-align: center !important;
}
.nr-card .wrap label {
background: #f4f8fb !important;
border: 1.5px solid #c4d9e8 !important;
border-radius: 10px !important;
padding: 7px 14px !important;
font-size: 0.82rem !important;
font-weight: 600 !important;
color: #2d4a5e !important;
cursor: pointer !important;
transition: all 0.18s ease !important;
}
.nr-card .wrap label:hover {
border-color: #1a6b8a !important;
background: #e8f4fb !important;
}
.nr-card .wrap input[type=radio]:checked + label,
.nr-card .wrap label.selected {
background: linear-gradient(135deg, #1a6b8a, #0f5c42) !important;
color: white !important;
border-color: transparent !important;
box-shadow: 0 2px 8px rgba(26,107,138,0.25) !important;
}
.nr-card select, select {
border-radius: 10px !important;
border: 1.5px solid #c4d9e8 !important;
background: #f4f8fb !important;
font-size: 0.85rem !important;
color: #1e293b !important;
padding: 8px 12px !important;
font-weight: 500 !important;
transition: border-color 0.2s !important;
}
.nr-card select:focus, select:focus {
border-color: #1a6b8a !important;
outline: none !important;
box-shadow: 0 0 0 3px rgba(26,107,138,0.1) !important;
}
#nr-btn button {
background: linear-gradient(135deg, #1a6b8a 0%, #0f5c42 100%) !important;
border: none !important;
border-radius: 14px !important;
color: white !important;
font-size: 1rem !important;
font-weight: 700 !important;
padding: 16px !important;
width: 100% !important;
box-shadow: 0 4px 20px rgba(26,107,138,0.35) !important;
transition: opacity 0.2s, transform 0.1s !important;
letter-spacing: 0.3px !important;
}
#nr-btn button:hover { opacity: 0.88 !important; transform: translateY(-2px) !important; }
#nr-btn button:active { transform: translateY(0) !important; }
.nr-html-out > div { border-radius: 14px !important; }
#nr-footer {
text-align: center; padding: 18px; font-size: 0.75rem;
color: #94a3b8; margin-top: 4px;
}
"""
with gr.Blocks(css=CSS, title='NeuroRest AI') as demo:
gr.HTML("""
""")
with gr.Row(equal_height=False):
with gr.Column(scale=5):
gr.HTML('👤 Personal Information
')
with gr.Group(elem_classes='nr-card'):
with gr.Row():
age = gr.Slider(18, 65, value=30, step=1, label='Age')
gender = gr.Radio(['Male', 'Female'], value='Male', label='Gender')
occupation = gr.Dropdown(OCCUPATION_LIST, value='Engineer', label='Occupation')
gr.HTML('😴 Sleep Metrics
')
with gr.Group(elem_classes='nr-card'):
with gr.Row():
sleep_duration = gr.Slider(4, 10, value=7, step=0.5, label='Sleep Duration (hrs)')
quality_of_sleep = gr.Slider(1, 10, value=7, step=1, label='Sleep Quality (1–10)')
gr.HTML('🏃 Physical Health
')
with gr.Group(elem_classes='nr-card'):
with gr.Row():
physical_activity = gr.Slider(0, 90, value=40, step=5, label='Physical Activity (min/day)')
bmi_category = gr.Radio(BMI_LIST, value='Normal', label='BMI Category')
gr.HTML('❤️ Vitals
')
with gr.Group(elem_classes='nr-card'):
with gr.Row():
bp_systolic = gr.Slider(90, 180, value=120, step=1, label='Systolic BP (mmHg)')
bp_diastolic = gr.Slider(60, 110, value=80, step=1, label='Diastolic BP (mmHg)')
with gr.Row():
heart_rate = gr.Slider(55, 100, value=70, step=1, label='Heart Rate (bpm)')
daily_steps = gr.Slider(1000, 12000, value=6000, step=500, label='Daily Steps')
with gr.Row(elem_id='nr-btn'):
analyze_btn = gr.Button('🔍 Analyze My Sleep & Stress', variant='primary')
with gr.Column(scale=5):
gr.HTML('📊 Sleep Apnea Score
')
apnea_out = gr.HTML(
value=""
"Run analysis to see your sleep score
"
)
gr.HTML('😓 Stress Level
')
stress_out = gr.HTML(
value=""
"Stress prediction will appear here...
",
elem_classes='nr-html-out'
)
gr.HTML('💡 Health Recommendations
')
rec_out = gr.HTML(
value=""
"Personalized recommendations will appear here...
",
elem_classes='nr-html-out'
)
gr.HTML('🤖 AI Sleep Coaching
')
coaching_out = gr.HTML(
value=""
"AI coaching will appear after analysis
"
)
gr.HTML("""
""")
analyze_btn.click(
fn=predict,
inputs=[
age, gender, occupation,
sleep_duration, quality_of_sleep,
physical_activity, bmi_category,
bp_systolic, bp_diastolic,
heart_rate, daily_steps
],
outputs=[apnea_out, stress_out, rec_out, coaching_out]
)
demo.launch()