src/streamlit_app.py

# proportion_inference_app.py
# Streamlit ≥1.32 — Accessible, minimal color design

import streamlit as st
import pandas as pd
import numpy as np
from scipy.stats import norm
import io

# ---------- Page Config ----------
st.set_page_config(
    page_title="Inference for Proportions",
    page_icon="📊",
    layout="centered",
    initial_sidebar_state="collapsed"
)

# ---------- Accessible CSS - Compact for embedding ----------
st.markdown("""
<style>
    /* Maximize vertical space usage */
    .block-container { 
        padding-top: 0.5rem !important; 
        padding-bottom: 0.5rem !important;
        max-width: 100% !important;
    }
    
    /* Tighter element spacing */
    .element-container { margin-bottom: 0.3rem !important; }
    .stRadio > div { margin-bottom: 0 !important; }
    .stSelectSlider { padding-top: 0 !important; padding-bottom: 0 !important; }
    h2 { margin-top: 0 !important; margin-bottom: 0.3rem !important; font-size: 1.4rem !important; }
    h3 { margin-top: 0.3rem !important; margin-bottom: 0.2rem !important; font-size: 1.1rem !important; }
    p { margin-bottom: 0.3rem !important; }
    .stNumberInput { margin-bottom: 0 !important; }
    
    /* Compact info boxes */
    .stAlert { padding: 0.4rem 0.7rem !important; margin: 0.2rem 0 !important; }
    
    /* Clean result box - compact */
    .result-box {
        background: #f8f9fa;
        border: 2px solid #dee2e6;
        padding: 0.6rem;
        border-radius: 8px;
        text-align: center;
        margin: 0.2rem 0;
    }
    .result-box .label {
        font-size: 0.75rem;
        color: #6c757d;
        margin-bottom: 0.1rem;
    }
    .result-box .value {
        font-size: 1.2rem;
        font-weight: 600;
        color: #212529;
    }
    
    /* Decision boxes - compact */
    .decision-box {
        padding: 0.6rem;
        border-radius: 8px;
        text-align: center;
        margin: 0.3rem 0;
        font-weight: 600;
        font-size: 1rem;
    }
    .decision-reject {
        background: #fff;
        border: 3px solid #212529;
    }
    .decision-accept {
        background: #fff;
        border: 3px dashed #6c757d;
    }
    
    /* Compact dividers */
    hr { margin: 0.5rem 0 !important; border: none; border-top: 1px solid #dee2e6; }
    
    /* Compact metrics */
    [data-testid="stMetricValue"] { font-size: 1.1rem !important; }
    [data-testid="stMetricLabel"] { font-size: 0.75rem !important; }
    
    /* Compact expander */
    .streamlit-expanderHeader { padding: 0.3rem 0 !important; font-size: 0.9rem !important; }
    
    /* Hide Streamlit branding for cleaner embed */
    #MainMenu {visibility: hidden;}
    footer {visibility: hidden;}
    header {visibility: hidden;}
</style>
""", unsafe_allow_html=True)

# ---------- Title ----------
st.markdown("## 📊 Inference for Proportions")

# ---------- INPUTS ----------
col1, col2 = st.columns(2)
with col1:
    inf_type = st.radio("Inference type:", ["One-Sample Proportion", "Two-Sample Proportions (independent)"],
                       key="inf_type")
with col2:
    analysis_type = st.radio("Analysis:", ["Confidence Interval", "Hypothesis Test"],
                            key="analysis_type")

st.markdown("---")

# ---------- Sample Data Inputs ----------
if inf_type == "One-Sample Proportion":
    input_method = st.radio(
        "Input method:",
        ["Successes and sample size", "Sample proportion and sample size"],
        key="input_method",
        horizontal=True
    )
    
    col1, col2 = st.columns(2)
    with col1:
        n = st.number_input("Sample size (n)", min_value=2, step=1, value=100, key="n")
    with col2:
        if input_method == "Successes and sample size":
            x = st.number_input("Number of successes (x)", min_value=0, max_value=int(n), 
                               step=1, value=50, key="x")
            phat = x / n if n > 0 else 0
        else:
            phat = st.number_input("Sample proportion (p̂)", min_value=0.0, max_value=1.0,
                                   format="%.4f", value=0.5, key="phat")
            x = round(phat * n)
    
    st.info(f"Sample proportion: **p̂ = {phat:.4f}**")

else:  # Two-Sample
    col1, col2 = st.columns(2)
    
    with col1:
        st.subheader("Group 1")
        input_method1 = st.radio(
            "Input method:",
            ["Successes and sample size", "Sample proportion and sample size"],
            key="input_method1"
        )
        n1 = st.number_input("Sample size n₁", min_value=2, step=1, value=100, key="n1")
        
        if input_method1 == "Successes and sample size":
            x1 = st.number_input("Number of successes x₁", min_value=0, max_value=int(n1),
                                step=1, value=50, key="x1")
            phat1 = x1 / n1 if n1 > 0 else 0
        else:
            phat1 = st.number_input("Sample proportion p̂₁", min_value=0.0, max_value=1.0,
                                    format="%.4f", value=0.5, key="phat1")
            x1 = round(phat1 * n1)
        st.info(f"p̂₁ = {phat1:.4f}")
            
    with col2:
        st.subheader("Group 2")
        input_method2 = st.radio(
            "Input method:",
            ["Successes and sample size", "Sample proportion and sample size"],
            key="input_method2"
        )
        n2 = st.number_input("Sample size n₂", min_value=2, step=1, value=100, key="n2")
        
        if input_method2 == "Successes and sample size":
            x2 = st.number_input("Number of successes x₂", min_value=0, max_value=int(n2),
                                step=1, value=50, key="x2")
            phat2 = x2 / n2 if n2 > 0 else 0
        else:
            phat2 = st.number_input("Sample proportion p̂₂", min_value=0.0, max_value=1.0,
                                    format="%.4f", value=0.5, key="phat2")
            x2 = round(phat2 * n2)
        st.info(f"p̂₂ = {phat2:.4f}")

st.markdown("---")

# ---------- CI / HT specific controls ----------
if analysis_type == "Confidence Interval":
    conf_level = st.select_slider("Confidence level:",
                                  options=[0.90, 0.95, 0.99], value=0.95,
                                  format_func=lambda x: f"{x*100:.0f}%", key="conf_level")
else:
    col1, col2 = st.columns(2)
    with col1:
        alpha = st.select_slider("Significance level (α):", 
                                options=[0.01, 0.05, 0.10], value=0.05, key="alpha")
    with col2:
        if inf_type == "One-Sample Proportion":
            p0 = st.number_input("Null proportion (p₀):", min_value=0.0001, max_value=0.9999,
                                 format="%.4f", value=0.5, key="p0")
    
    if inf_type == "One-Sample Proportion":
        alt = st.radio("Alternative hypothesis (H₁):",
                       ["p ≠ p₀ (Two-sided)", "p > p₀ (Right-sided)", "p < p₀ (Left-sided)"],
                       horizontal=True, key="alt_one")
    else:
        alt = st.radio("Alternative hypothesis (H₁):",
                       ["p₁ ≠ p₂ (Two-sided)", "p₁ > p₂ (Right-sided)", "p₁ < p₂ (Left-sided)"],
                       horizontal=True, key="alt_two")

# ---------- RUN ----------
run = st.button("▶ Run Analysis", type="primary", use_container_width=True)

# ---------- Helper function ----------
def result_box(label, value):
    return f'<div class="result-box"><div class="label">{label}</div><div class="value">{value}</div></div>'

# ===== RESULTS =====
if run:
    st.markdown("---")
    
    # ===== ONE-SAMPLE =====
    if inf_type == "One-Sample Proportion" and n >= 2:
        
        if analysis_type == "Confidence Interval":
            se = np.sqrt(phat * (1 - phat) / n)
            crit = norm.ppf(1 - (1 - conf_level) / 2)
            margin = crit * se
            lower, upper = max(0, phat - margin), min(1, phat + margin)
            check1, check2 = n * phat >= 10, n * (1 - phat) >= 10
            
            cols = st.columns(2)
            with cols[0]:
                st.markdown(result_box("Sample Proportion (p̂)", f"{phat:.4f}"), unsafe_allow_html=True)
            with cols[1]:
                st.markdown(result_box(f"{conf_level*100:.0f}% Confidence Interval", 
                                       f"[{lower:.4f}, {upper:.4f}]"), unsafe_allow_html=True)
            
            st.markdown("---")
            st.subheader("Calculation Details")
            col1, col2, col3 = st.columns(3)
            with col1:
                st.metric("Standard Error", f"{se:.4f}")
            with col2:
                st.metric("Critical Value (z)", f"±{crit:.4f}")
            with col3:
                st.metric("Margin of Error", f"{margin:.4f}")
            
            st.markdown("**Normal Approximation Check:**")
            st.write(f"• np̂ = {n*phat:.1f} {'✓ ≥ 10' if check1 else '✗ < 10'}")
            st.write(f"• n(1−p̂) = {n*(1-phat):.1f} {'✓ ≥ 10' if check2 else '✗ < 10'}")
            
            results = {"Analysis": ["CI"], "n": [n], "p̂": [phat], "SE": [se],
                      "Conf": [conf_level], "Lower": [lower], "Upper": [upper]}

        else:  # Hypothesis Test
            se = np.sqrt(p0 * (1 - p0) / n)
            z = (phat - p0) / se
            
            if "≠" in alt:
                p_val = 2 * (1 - norm.cdf(abs(z)))
                crit = norm.ppf(1 - alpha/2)
                crit_display = f"±{crit:.4f}"
            elif ">" in alt:
                p_val = 1 - norm.cdf(z)
                crit = norm.ppf(1 - alpha)
                crit_display = f"{crit:.4f}"
            else:
                p_val = norm.cdf(z)
                crit = -norm.ppf(1 - alpha)
                crit_display = f"{crit:.4f}"

            reject = p_val <= alpha
            
            cols = st.columns(2)
            with cols[0]:
                st.markdown(result_box("z-statistic", f"{z:.4f}"), unsafe_allow_html=True)
            with cols[1]:
                st.markdown(result_box("p-value", f"{p_val:.4g}"), unsafe_allow_html=True)
            
            # Decision with pattern, not color
            if reject:
                st.markdown(f'<div class="decision-box decision-reject">✗ REJECT H₀ at α = {alpha}</div>', 
                           unsafe_allow_html=True)
            else:
                st.markdown(f'<div class="decision-box decision-accept">— Fail to reject H₀ at α = {alpha}</div>', 
                           unsafe_allow_html=True)
            
            st.markdown("---")
            st.subheader("Calculation Details")
            st.write(f"**Hypotheses:** H₀: p = {p0:.4f} vs H₁: {alt}")
            st.write(f"**Sample proportion:** p̂ = {phat:.4f}")
            st.write(f"**Standard Error (under H₀):** {se:.4f}")
            st.write(f"**Critical value(s):** {crit_display}")
            
            results = {"Analysis": ["HT"], "n": [n], "p̂": [phat], "p₀": [p0],
                      "z": [z], "p-value": [p_val], "Reject": [reject]}

    # ===== TWO-SAMPLE =====
    elif inf_type.startswith("Two-Sample") and n1 >= 2 and n2 >= 2:
        diff = phat1 - phat2
        
        if analysis_type == "Confidence Interval":
            se = np.sqrt(phat1*(1-phat1)/n1 + phat2*(1-phat2)/n2)
            crit = norm.ppf(1 - (1 - conf_level) / 2)
            margin = crit * se
            lower, upper = diff - margin, diff + margin
            
            cols = st.columns(2)
            with cols[0]:
                st.markdown(result_box("Difference (p̂₁ − p̂₂)", f"{diff:.4f}"), unsafe_allow_html=True)
            with cols[1]:
                st.markdown(result_box(f"{conf_level*100:.0f}% Confidence Interval", 
                                       f"[{lower:.4f}, {upper:.4f}]"), unsafe_allow_html=True)
            
            st.markdown("---")
            st.subheader("Calculation Details")
            st.write(f"**p̂₁ = {phat1:.4f}**, **p̂₂ = {phat2:.4f}**")
            col1, col2, col3 = st.columns(3)
            with col1:
                st.metric("SE (unpooled)", f"{se:.4f}")
            with col2:
                st.metric("Critical Value (z)", f"±{crit:.4f}")
            with col3:
                st.metric("Margin of Error", f"{margin:.4f}")
            
            results = {"Analysis": ["CI-2"], "p̂₁": [phat1], "p̂₂": [phat2],
                      "Diff": [diff], "Lower": [lower], "Upper": [upper]}

        else:  # Hypothesis Test
            p_pooled = (x1 + x2) / (n1 + n2)
            se = np.sqrt(p_pooled * (1 - p_pooled) * (1/n1 + 1/n2))
            z = diff / se
            
            if "≠" in alt:
                p_val = 2 * (1 - norm.cdf(abs(z)))
                crit = norm.ppf(1 - alpha/2)
                crit_display = f"±{crit:.4f}"
            elif ">" in alt:
                p_val = 1 - norm.cdf(z)
                crit = norm.ppf(1 - alpha)
                crit_display = f"{crit:.4f}"
            else:
                p_val = norm.cdf(z)
                crit = -norm.ppf(1 - alpha)
                crit_display = f"{crit:.4f}"

            reject = p_val <= alpha
            
            cols = st.columns(2)
            with cols[0]:
                st.markdown(result_box("z-statistic", f"{z:.4f}"), unsafe_allow_html=True)
            with cols[1]:
                st.markdown(result_box("p-value", f"{p_val:.4g}"), unsafe_allow_html=True)
            
            if reject:
                st.markdown(f'<div class="decision-box decision-reject">✗ REJECT H₀ at α = {alpha}</div>', 
                           unsafe_allow_html=True)
            else:
                st.markdown(f'<div class="decision-box decision-accept">— Fail to reject H₀ at α = {alpha}</div>', 
                           unsafe_allow_html=True)
            
            st.markdown("---")
            st.subheader("Calculation Details")
            st.write(f"**Hypotheses:** H₀: p₁ = p₂ vs H₁: {alt}")
            st.write(f"**p̂₁ = {phat1:.4f}**, **p̂₂ = {phat2:.4f}**, **Difference = {diff:.4f}**")
            st.write(f"**Pooled proportion:** p̂ = {p_pooled:.4f}")
            st.write(f"**Standard Error (pooled):** {se:.4f}")
            st.write(f"**Critical value(s):** {crit_display}")
            
            results = {"Analysis": ["HT-2"], "p̂₁": [phat1], "p̂₂": [phat2],
                      "z": [z], "p-value": [p_val], "Reject": [reject]}

    # Download
    if 'results' in locals():
        df_out = pd.DataFrame(results)
        buff = io.BytesIO()
        with pd.ExcelWriter(buff, engine="xlsxwriter") as writer:
            df_out.to_excel(writer, index=False)
        st.download_button("📥 Download Results", data=buff.getvalue(),
                          file_name="proportion_inference.xlsx",
                          mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet")

# ---------- Formulas (collapsed) ----------
with st.expander("📚 Formulas & Theory"):
    st.markdown(r"""
**One-Sample Proportion**
- Sample proportion: $\hat p = x/n$
- SE for CI: $\sqrt{\hat p(1-\hat p)/n}$
- SE for HT: $\sqrt{p_0(1-p_0)/n}$
- CI: $\hat p \pm z_{\alpha/2} \cdot SE$
- Test stat: $z = (\hat p - p_0)/SE$

**Two-Sample Proportions**
- Pooled: $\hat p = (x_1+x_2)/(n_1+n_2)$
- SE for CI: $\sqrt{\frac{\hat p_1(1-\hat p_1)}{n_1} + \frac{\hat p_2(1-\hat p_2)}{n_2}}$
- SE for HT: $\sqrt{\hat p(1-\hat p)(1/n_1 + 1/n_2)}$

**Conditions:** $np \geq 10$ and $n(1-p) \geq 10$
""")