app.py

# app_FP.py — ST_GeoMech_FP (Fracture Pressure)
# Mirrors the SHmin app's specs & workflow (password gate, preview panel, train/validate/predict, Excel export).
# Self-contained: trains a fixed, optimized RF pipeline in-app. No external model files.

import io
import os
import base64
import math
from pathlib import Path
from datetime import datetime

import streamlit as st
import pandas as pd
import numpy as np

# Matplotlib (static previews & cross-plot)
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
from matplotlib.ticker import FuncFormatter

import plotly.graph_objects as go
from sklearn.metrics import mean_squared_error
from sklearn.ensemble import RandomForestRegressor
from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer

# =========================
# App constants / defaults
# =========================
APP_NAME = "ST_GeoMech_FP"
TAGLINE  = "Real-Time Fracture Pressure Prediction"

# Canonical features (match SHmin app)
FEATURES = ["Q (gpm)", "SPP (psi)", "T (kft.lbf)", "WOB (klbf)", "ROP (ft/h)"]

# Canonical prediction/target labels
TARGET_CANON = "FracPress_Actual"
PRED_COL     = "FracPress_Pred"
UNITS        = "Psi"

# Target aliases accepted in input workbooks
TARGET_ALIASES = [
    "FracPress_Actual", "FracturePressure_Actual", "Fracture Pressure (psi)",
    "Frac Pressure (psi)", "FracPressure", "Frac_Pressure", "FracturePressure",
    "FP_Actual", "FP (psi)"
]

# Optional transform (kept for parity; RF is used on raw scale)
TRANSFORM = "none"  # "none" | "log10" | "ln"

# Fixed "best" RandomForest params
BEST_PARAMS = dict(
    n_estimators=400,
    max_depth=None,
    min_samples_split=2,
    min_samples_leaf=1,
    max_features=0.6,
    bootstrap=True,
    random_state=42,
    n_jobs=-1,
)

# Color / layout
COLORS   = {"pred": "#1f77b4", "actual": "#f2b702", "ref": "#5a5a5a"}
CROSS_W, CROSS_H = 350, 350
TRACK_H, TRACK_W = 1000, 500
FONT_SZ  = 13
BOLD_FONT = "Arial Black, Arial, sans-serif"

# =========================
# Page / CSS
# =========================
st.set_page_config(page_title=APP_NAME, page_icon="logo.png", layout="wide")
st.markdown("""
<style>
 .brand-logo { width: 200px; height: auto; object-fit: contain; }
 .centered-container { display:flex; flex-direction:column; align-items:center; text-align:center; }
 .st-message-box { background:#f0f2f6; color:#333; padding:10px; border-radius:10px; border:1px solid #e6e9ef; }
 .st-message-box.st-success { background:#d4edda; color:#155724; border-color:#c3e6cb; }
 .st-message-box.st-warning { background:#fff3cd; color:#856404; border-color:#ffeeba; }
 .st-message-box.st-error   { background:#f8d7da; color:#721c24; border-color:#f5c6cb; }
 .main .block-container { overflow: unset !important; }
 div[data-testid="stVerticalBlock"] { overflow: unset !important; }
 div[data-testid="stExpander"] > details > summary {
   position: sticky; top: 0; z-index: 10; background: #fff; border-bottom: 1px solid #eee;
 }
 div[data-testid="stExpander"] div[data-baseweb="tab-list"] {
   position: sticky; top: 42px; z-index: 9; background: #fff; padding-top: 6px;
 }
</style>
""", unsafe_allow_html=True)

TABLE_CENTER_CSS = [
    dict(selector="th", props=[("text-align", "center")]),
    dict(selector="td", props=[("text-align", "center")]),
]

# =========================
# Password gate (optional)
# =========================
def inline_logo(path: str = "logo.png") -> str:
    try:
        p = Path(path)
        if not p.exists():
            return ""
        return f"data:image/png;base64,{base64.b64encode(p.read_bytes()).decode('ascii')}"
    except Exception:
        return ""

def add_password_gate() -> None:
    try:
        required = st.secrets.get("APP_PASSWORD", "")
    except Exception:
        required = os.environ.get("APP_PASSWORD", "")
    if not required:
        return
    if st.session_state.get("auth_ok", False):
        return
    st.sidebar.markdown(f"""
        <div class="centered-container">
            <img src="{inline_logo('logo.png')}" class="brand-logo">
            <div style='font-weight:800;font-size:1.2rem; margin-top: 10px;'>{APP_NAME}</div>
            <div style='color:#667085;'>Smart Thinking • Secure Access</div>
        </div>
    """, unsafe_allow_html=True)
    pwd = st.sidebar.text_input("Access key", type="password", placeholder="••••••••")
    if st.sidebar.button("Unlock", type="primary"):
        if pwd == required:
            st.session_state.auth_ok = True
            st.rerun()
        else:
            st.error("Incorrect key.")
    st.stop()

add_password_gate()

# =========================
# Utilities
# =========================
def rmse(y_true, y_pred) -> float:
    return float(np.sqrt(mean_squared_error(y_true, y_pred)))

def mape(y_true, y_pred, eps: float = 1e-9) -> float:
    a = np.asarray(y_true, dtype=float)
    p = np.asarray(y_pred, dtype=float)
    den = np.maximum(np.abs(a), eps)
    return float(np.mean(np.abs((a - p) / den)) * 100.0)

def pearson_r(y_true, y_pred) -> float:
    a = np.asarray(y_true, dtype=float)
    p = np.asarray(y_pred, dtype=float)
    if a.size < 2:
        return float("nan")
    if np.all(a == a[0]) or np.all(p == p[0]):
        return float("nan")
    return float(np.corrcoef(a, p)[0, 1])

@st.cache_data(show_spinner=False)
def parse_excel(data_bytes: bytes) -> dict[str, pd.DataFrame]:
    bio = io.BytesIO(data_bytes)
    xl = pd.ExcelFile(bio)
    return {sh: xl.parse(sh) for sh in xl.sheet_names}

def read_book_bytes(b: bytes) -> dict[str, pd.DataFrame]:
    return parse_excel(b) if b else {}

def _excel_engine() -> str:
    try:
        import xlsxwriter  # noqa: F401
        return "xlsxwriter"
    except Exception:
        return "openpyxl"

def _excel_safe_name(name: str) -> str:
    bad = '[]:*?/\\'
    safe = ''.join('_' if ch in bad else ch for ch in str(name))
    return safe[:31]

def _round_numeric(df: pd.DataFrame, ndigits: int = 3) -> pd.DataFrame:
    out = df.copy()
    for c in out.columns:
        if pd.api.types.is_float_dtype(out[c]) or pd.api.types.is_integer_dtype(out[c]):
            out[c] = pd.to_numeric(out[c], errors="coerce").round(ndigits)
    return out

def df_centered_rounded(df: pd.DataFrame, hide_index: bool = True) -> None:
    out = df.copy()
    numcols = out.select_dtypes(include=[np.number]).columns
    styler = (
        out.style
           .format({c: "{:.3f}" for c in numcols})
           .set_properties(**{"text-align": "center"})
           .set_table_styles(TABLE_CENTER_CSS)
    )
    st.dataframe(styler, use_container_width=True, hide_index=hide_index)

def ensure_cols(df: pd.DataFrame, cols: list[str]) -> bool:
    miss = [c for c in cols if c not in df.columns]
    if miss:
        st.error(f"Missing columns: {miss}\nFound: {list(df.columns)}")
        return False
    return True

def _nice_tick0(xmin: float, step: float = 0.1) -> float:
    return step * math.floor(xmin / step) if np.isfinite(xmin) else xmin

# ---------- Transform helpers ----------
def _inv_transform(x: np.ndarray, transform: str) -> np.ndarray:
    t = (transform or "none").lower()
    if t in ("log10", "log_10", "log10()"):
        return np.power(10.0, x)
    if t in ("ln", "log", "loge", "log_e", "natural"):
        return np.exp(x)
    return x

# ---------- Build X exactly as trained ----------
def _make_X(df: pd.DataFrame, features: list[str]) -> pd.DataFrame:
    X = df.reindex(columns=features, copy=False)
    for c in X.columns:
        X[c] = pd.to_numeric(X[c], errors="coerce")
    return X

# ---------- Target resolver (use aliases) ----------
def _resolve_target_col(df: pd.DataFrame) -> str | None:
    cols_lower = {c.lower(): c for c in df.columns}
    for cand in TARGET_ALIASES:
        if cand.lower() in cols_lower:
            return cols_lower[cand.lower()]
    return None

# =========================
# Export helpers
# =========================
def _summary_table(df: pd.DataFrame, cols: list[str]) -> pd.DataFrame:
    cols = [c for c in cols if c in df.columns]
    if not cols:
        return pd.DataFrame()
    tbl = (
        df[cols]
        .agg(["min", "max", "mean", "std"])
        .T.rename(columns={"min": "Min", "max": "Max", "mean": "Mean", "std": "Std"})
        .reset_index(names="Field")
    )
    return _round_numeric(tbl, 3)

def _train_ranges_df(ranges: dict[str, tuple[float, float]]) -> pd.DataFrame:
    if not ranges:
        return pd.DataFrame()
    df = pd.DataFrame(ranges).T.reset_index()
    df.columns = ["Feature", "Min", "Max"]
    return _round_numeric(df, 3)

def _excel_autofit(writer, sheet_name: str, df: pd.DataFrame, min_w: int = 8, max_w: int = 40) -> None:
    try:
        import xlsxwriter  # noqa: F401
    except Exception:
        return
    ws = writer.sheets[sheet_name]
    for i, col in enumerate(df.columns):
        series = df[col].astype(str)
        max_len = max([len(str(col))] + series.map(len).tolist())
        ws.set_column(i, i, max(min_w, min(max_len + 2, max_w)))
    ws.freeze_panes(1, 0)

def _available_sections() -> list[str]:
    res = st.session_state.get("results", {})
    sections: list[str] = []
    if "Train" in res:
        sections += ["Training", "Training_Metrics", "Training_Summary"]
    if "Test" in res:
        sections += ["Testing", "Testing_Metrics", "Testing_Summary"]
    if "Validate" in res:
        sections += ["Validation", "Validation_Metrics", "Validation_Summary", "Validation_OOR"]
    if "PredictOnly" in res:
        sections += ["Prediction", "Prediction_Summary"]
    if st.session_state.get("train_ranges"):
        sections += ["Training_Ranges"]
    sections += ["Info"]
    return sections

def build_export_workbook(selected: list[str], ndigits: int = 3, do_autofit: bool = True) -> tuple[bytes | None, str | None, list[str]]:
    res = st.session_state.get("results", {})
    if not res:
        return None, None, []
    sheets: dict[str, pd.DataFrame] = {}
    order: list[str] = []

    def _add(name: str, df: pd.DataFrame) -> None:
        if df is None or (isinstance(df, pd.DataFrame) and df.empty):
            return
        sheets[name] = _round_numeric(df, ndigits)
        order.append(name)

    # Training / Testing
    if "Training" in selected and "Train" in res:
        _add("Training", res["Train"])
    if "Training_Metrics" in selected and res.get("m_train"):
        _add("Training_Metrics", pd.DataFrame([res["m_train"]]))
    if "Training_Summary" in selected and "Train" in res:
        tr_cols = FEATURES + [c for c in [st.session_state.get("tcol_train", TARGET_CANON), PRED_COL] if c in res["Train"].columns]
        _add("Training_Summary", _summary_table(res["Train"], tr_cols))

    if "Testing" in selected and "Test" in res:
        _add("Testing", res["Test"])
    if "Testing_Metrics" in selected and res.get("m_test"):
        _add("Testing_Metrics", pd.DataFrame([res["m_test"]]))
    if "Testing_Summary" in selected and "Test" in res:
        te_cols = FEATURES + [c for c in [st.session_state.get("tcol_test", TARGET_CANON), PRED_COL] if c in res["Test"].columns]
        _add("Testing_Summary", _summary_table(res["Test"], te_cols))

    # Validation / Prediction
    if "Validation" in selected and "Validate" in res:
        _add("Validation", res["Validate"])
    if "Validation_Metrics" in selected and res.get("m_val"):
        _add("Validation_Metrics", pd.DataFrame([res["m_val"]]))
    if "Validation_Summary" in selected and res.get("sv_val"):
        _add("Validation_Summary", pd.DataFrame([res["sv_val"]]))
    if "Validation_OOR" in selected and isinstance(res.get("oor_tbl"), pd.DataFrame) and not res["oor_tbl"].empty:
        _add("Validation_OOR", res["oor_tbl"].reset_index(drop=True))

    if "Prediction" in selected and "PredictOnly" in res:
        _add("Prediction", res["PredictOnly"])
    if "Prediction_Summary" in selected and res.get("sv_pred"):
        _add("Prediction_Summary", pd.DataFrame([res["sv_pred"]]))

    if "Training_Ranges" in selected and st.session_state.get("train_ranges"):
        _add("Training_Ranges", _train_ranges_df(st.session_state["train_ranges"]))

    if "Info" in selected:
        info = pd.DataFrame([
            {"Key": "AppName",    "Value": APP_NAME},
            {"Key": "Tagline",    "Value": TAGLINE},
            {"Key": "Target",     "Value": st.session_state.get("tcol_train", TARGET_CANON)},
            {"Key": "PredColumn", "Value": PRED_COL},
            {"Key": "Features",   "Value": ", ".join(FEATURES)},
            {"Key": "ExportedAt", "Value": datetime.now().strftime("%Y-%m-%d %H:%M:%S")},
        ])
        _add("Info", info)

    if not order:
        return None, None, []

    bio = io.BytesIO()
    engine = _excel_engine()
    with pd.ExcelWriter(bio, engine=engine) as writer:
        for name in order:
            df = sheets[name]
            sheet = _excel_safe_name(name)
            df.to_excel(writer, sheet_name=sheet, index=False)
            if do_autofit:
                _excel_autofit(writer, sheet, df)
    bio.seek(0)
    fname = f"FracPressure_Export_{datetime.now().strftime('%Y%m%d_%H%M%S')}.xlsx"
    return bio.getvalue(), fname, order

def render_export_button(phase_key: str) -> None:
    res = st.session_state.get("results", {})
    if not res:
        return
    st.divider()
    st.markdown("### Export to Excel")
    options = _available_sections()
    selected_sheets = st.multiselect(
        "Sheets to include",
        options=options,
        default=[],
        placeholder="Choose option(s)",
        help="Pick the sheets you want in the Excel export.",
        key=f"sheets_{phase_key}",
    )
    if not selected_sheets:
        st.caption("Select one or more sheets above to enable export.")
        st.download_button(
            "⬇️ Export Excel", data=b"", file_name="FracPressure_Export.xlsx",
            mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
            disabled=True, key=f"download_{phase_key}",
        )
        return
    data, fname, names = build_export_workbook(selected=selected_sheets, ndigits=3, do_autofit=True)
    if names:
        st.caption("Will include: " + ", ".join(names))
    st.download_button(
        "⬇️ Export Excel", data=(data or b""), file_name=(fname or "FracPressure_Export.xlsx"),
        mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
        disabled=(data is None), key=f"download_{phase_key}",
    )

# =========================
# Plots (integer ticks)
# =========================
def cross_plot_static(actual, pred, label: str = "Fracture Pressure"):
    a = pd.Series(actual, dtype=float)
    p = pd.Series(pred,   dtype=float)
    lo = float(min(a.min(), p.min()))
    hi = float(max(a.max(), p.max()))
    pad = 0.03 * (hi - lo if hi > lo else 1.0)
    lo2, hi2 = lo - pad, hi + pad
    ticks = np.linspace(lo2, hi2, 5)

    dpi = 110
    fig, ax = plt.subplots(figsize=(CROSS_W / dpi, CROSS_H / dpi), dpi=dpi, constrained_layout=False)
    ax.scatter(a, p, s=14, c=COLORS["pred"], alpha=0.9, linewidths=0)
    ax.plot([lo2, hi2], [lo2, hi2], linestyle="--", linewidth=1.2, color=COLORS["ref"])

    ax.set_xlim(lo2, hi2); ax.set_ylim(lo2, hi2)
    ax.set_xticks(ticks);  ax.set_yticks(ticks)
    ax.set_aspect("equal", adjustable="box")

    fmt = FuncFormatter(lambda x, _: f"{x:.0f}")
    ax.xaxis.set_major_formatter(fmt); ax.yaxis.set_major_formatter(fmt)

    ax.set_xlabel(f"Actual {label} ({UNITS})",  fontweight="bold", fontsize=10, color="black")
    ax.set_ylabel(f"Predicted {label} ({UNITS})", fontweight="bold", fontsize=10, color="black")
    ax.tick_params(labelsize=6, colors="black")
    ax.grid(True, linestyle=":", alpha=0.3)
    for spine in ax.spines.values():
        spine.set_linewidth(1.1); spine.set_color("#444")

    fig.subplots_adjust(left=0.16, bottom=0.16, right=0.98, top=0.98)
    return fig

def track_plot(df: pd.DataFrame, actual_col: str | None, include_actual: bool = True):
    depth_col = next((c for c in df.columns if ("depth" in str(c).lower()) or ("tvd" in str(c).lower())), None)
    if depth_col is not None:
        y = pd.to_numeric(df[depth_col], errors="coerce")
        ylab = depth_col
        y_range = [float(np.nanmax(y)), float(np.nanmin(y))]  # reversed
    else:
        y = pd.Series(np.arange(1, len(df) + 1))
        ylab = "Point Index"
        y_range = [float(y.max()), float(y.min())]

    x_series = pd.Series(df.get(PRED_COL, pd.Series(dtype=float))).astype(float)
    if include_actual and actual_col and actual_col in df.columns:
        x_series = pd.concat([x_series, pd.Series(df[actual_col]).astype(float)], ignore_index=True)

    x_lo, x_hi = float(x_series.min()), float(x_series.max())
    x_pad = 0.03 * (x_hi - x_lo if x_hi > x_lo else 1.0)
    xmin, xmax = x_lo - x_pad, x_hi + x_pad
    tick0 = _nice_tick0(xmin, step=max((xmax - xmin) / 10.0, 0.1))

    fig = go.Figure()
    if PRED_COL in df.columns:
        fig.add_trace(go.Scatter(
            x=df[PRED_COL], y=y, mode="lines",
            line=dict(color=COLORS["pred"], width=1.8),
            name=PRED_COL,
            hovertemplate=f"{PRED_COL}: "+"%{x:.0f}<br>"+ylab+": %{y}<extra></extra>"
        ))
    if include_actual and actual_col and actual_col in df.columns:
        fig.add_trace(go.Scatter(
            x=df[actual_col], y=y, mode="lines",
            line=dict(color=COLORS["actual"], width=2.0, dash="dot"),
            name=f"{actual_col} (actual)",
            hovertemplate=f"{actual_col}: "+"%{x:.0f}<br>"+ylab+": %{y}<extra></extra>"
        ))

    fig.update_layout(
        height=TRACK_H, width=TRACK_W, autosize=False,
        paper_bgcolor="#fff", plot_bgcolor="#fff",
        margin=dict(l=64, r=16, t=36, b=48), hovermode="closest",
        font=dict(size=FONT_SZ, color="#000"),
        legend=dict(x=0.98, y=0.05, xanchor="right", yanchor="bottom",
                    bgcolor="rgba(255,255,255,0.75)", bordercolor="#ccc", borderwidth=1),
        legend_title_text=""
    )
    fig.update_xaxes(
        title_text=f"Fracture Pressure ({UNITS})",
        title_font=dict(size=20, family=BOLD_FONT, color="#000"),
        tickfont=dict(size=15, family=BOLD_FONT, color="#000"),
        side="top", range=[xmin, xmax],
        ticks="outside", tickformat=",.0f", tickmode="auto", tick0=tick0,
        showline=True, linewidth=1.2, linecolor="#444", mirror=True,
        showgrid=True, gridcolor="rgba(0,0,0,0.12)", automargin=True
    )
    fig.update_yaxes(
        title_text=ylab,
        title_font=dict(size=20, family=BOLD_FONT, color="#000"),
        tickfont=dict(size=15, family=BOLD_FONT, color="#000"),
        range=y_range, ticks="outside",
        showline=True, linewidth=1.2, linecolor="#444", mirror=True,
        showgrid=True, gridcolor="rgba(0,0,0,0.12)", automargin=True
    )
    return fig

def preview_tracks(df: pd.DataFrame, cols: list[str]):
    cols = [c for c in cols if c in df.columns]
    n = len(cols)
    if n == 0:
        fig, ax = plt.subplots(figsize=(4, 2))
        ax.text(0.5, 0.5, "No selected columns", ha="center", va="center")
        ax.axis("off")
        return fig

    depth_col = next((c for c in df.columns if ("depth" in str(c).lower()) or ("tvd" in str(c).lower())), None)
    if depth_col is not None:
        idx = pd.to_numeric(df[depth_col], errors="coerce")
        y_label = depth_col
        y_min, y_max = float(np.nanmin(idx)), float(np.nanmax(idx))
    else:
        idx = pd.Series(np.arange(1, len(df) + 1))
        y_label = "Point Index"
        y_min, y_max = float(idx.min()), float(idx.max())

    cmap = plt.get_cmap("tab20")
    col_colors = {col: cmap(i % cmap.N) for i, col in enumerate(cols)}

    fig, axes = plt.subplots(1, n, figsize=(2.4 * n, 7.0), sharey=True, dpi=100)
    if n == 1:
        axes = [axes]

    for i, (ax, col) in enumerate(zip(axes, cols)):
        x = pd.to_numeric(df[col], errors="coerce")
        ax.plot(x, idx, '-', lw=1.6, color=col_colors[col])
        ax.set_xlabel(col); ax.xaxis.set_label_position('top'); ax.xaxis.tick_top()
        ax.set_ylim(y_max, y_min)  # reversed depth down
        ax.grid(True, linestyle=":", alpha=0.3)
        if i == 0:
            ax.set_ylabel(y_label)
        else:
            ax.tick_params(labelleft=False); ax.set_ylabel("")

    fig.tight_layout()
    return fig

# =========================
# Fixed training pipeline
# =========================
def build_pipeline() -> Pipeline:
    model = RandomForestRegressor(**BEST_PARAMS)
    pipe = Pipeline(steps=[
        ("imputer", SimpleImputer(strategy="median")),
        ("model", model),
    ])
    return pipe

# =========================
# Session state (mirrors SHmin)
# =========================
st.session_state.setdefault("app_step", "intro")
st.session_state.setdefault("results", {})
st.session_state.setdefault("train_ranges", None)
st.session_state.setdefault("dev_file_name", "")
st.session_state.setdefault("dev_file_bytes", b"")
st.session_state.setdefault("dev_file_loaded", False)
st.session_state.setdefault("fitted_model", None)

# Persistent top-of-page preview panel
st.session_state.setdefault("show_preview_panel", False)
st.session_state.setdefault("preview_book", {})

# =========================
# Sidebar branding
# =========================
st.sidebar.markdown(f"""
    <div class="centered-container">
        <img src="{inline_logo('logo.png')}" class="brand-logo">
        <div style='font-weight:800;font-size:1.2rem;'>{APP_NAME}</div>
        <div style='color:#667085;'>{TAGLINE}</div>
    </div>
""", unsafe_allow_html=True)

def sticky_header(title: str, message: str) -> None:
    st.markdown(
        f"""
        <style>
        .sticky-container {{
            position: sticky; top: 0; background-color: white; z-index: 100;
            padding-top: 10px; padding-bottom: 10px; border-bottom: 1px solid #eee;
        }}
        </style>
        <div class="sticky-container">
            <h3>{title}</h3>
            <p>{message}</p>
        </div>
        """,
        unsafe_allow_html=True
    )

def render_preview_panel() -> None:
    """Top-of-page preview panel (same behavior as SHmin)."""
    if not st.session_state.get("show_preview_panel"):
        return
    st.markdown("## 🔎 Data preview")
    book = st.session_state.get("preview_book", {}) or {}
    if not book:
        st.info("No data loaded yet.")
        col = st.columns(2)[1]
        with col:
            if st.button("Hide preview"):
                st.session_state.show_preview_panel = False
                st.session_state.preview_book = {}
                st.rerun()
        return

    names = list(book.keys())
    tabs = st.tabs(names + ["✖ Hide preview"])
    for i, name in enumerate(names):
        with tabs[i]:
            df = book[name]
            t1, t2 = st.tabs(["Tracks", "Summary"])
            with t1:
                st.pyplot(preview_tracks(df, FEATURES), use_container_width=True)
            with t2:
                feat_present = [c for c in FEATURES if c in df.columns]
                if not feat_present:
                    st.info("No feature columns found to summarize.")
                else:
                    tbl = (
                        df[feat_present]
                        .agg(["min", "max", "mean", "std"])
                        .T.rename(columns={"min": "Min", "max": "Max", "mean": "Mean", "std": "Std"})
                        .reset_index(names="Feature")
                    )
                    df_centered_rounded(tbl)
    with tabs[-1]:
        if st.button("Hide preview", use_container_width=True):
            st.session_state.show_preview_panel = False
            st.session_state.preview_book = {}
            st.rerun()

# =========================
# INTRO
# =========================
if st.session_state.app_step == "intro":
    st.header("Welcome!")
    st.markdown(
        f"This software is developed by *Smart Thinking AI-Solutions Team* to estimate **Fracture Pressure** ({UNITS}) from drilling/offset data."
    )
    st.subheader("How It Works")
    st.markdown(
        "1) **Upload your data file** and click **Run Model** to fit the baked-in pipeline.  \n"
        "2) **Validate** on held-out wells (with actual).  \n"
        "3) **Predict** on wells without actual."
    )
    if st.button("Start Showcase", type="primary"):
        st.session_state.app_step = "dev"
        st.rerun()

# =========================
# CASE BUILDING (Train/Test)
# =========================
def _find_sheet(book: dict[str, pd.DataFrame], names: list[str]) -> str | None:
    low2orig = {k.lower(): k for k in book.keys()}
    for nm in names:
        if nm.lower() in low2orig:
            return low2orig[nm.lower()]
    return None

if st.session_state.app_step == "dev":
    st.sidebar.header("Case Building")
    up = st.sidebar.file_uploader("Upload Your Data File", type=["xlsx", "xls"])
    if up is not None:
        st.session_state.dev_file_bytes = up.getvalue()
        st.session_state.dev_file_name = up.name
        st.session_state.dev_file_loaded = True
        st.session_state.fitted_model = None
        # show preview panel immediately
        st.session_state.preview_book = read_book_bytes(st.session_state.dev_file_bytes) if st.session_state.dev_file_bytes else {}
        st.session_state.show_preview_panel = True
        st.rerun()

    if st.session_state.dev_file_loaded:
        tmp = read_book_bytes(st.session_state.dev_file_bytes)
        if tmp:
            df0 = next(iter(tmp.values()))
            st.sidebar.caption(f"**Data loaded:** {st.session_state.dev_file_name} • {df0.shape[0]} rows × {df0.shape[1]} cols")

    run = st.sidebar.button("Run Model", type="primary", use_container_width=True)
    if st.sidebar.button("Proceed to Validation ▶", use_container_width=True):
        st.session_state.app_step = "validate"; st.rerun()
    if st.sidebar.button("Proceed to Prediction ▶", use_container_width=True):
        st.session_state.app_step = "predict"; st.rerun()

    if st.session_state.dev_file_loaded and st.session_state.show_preview_panel:
        sticky_header("Case Building", "Previewed ✓ — now click **Run Model**.")
    elif st.session_state.dev_file_loaded:
        sticky_header("Case Building", "📄 **Preview uploaded data** using the sidebar button, then click **Run Model**.")
    else:
        sticky_header("Case Building", "**Upload your data to build a case, then run the model to review performance.**")

    render_preview_panel()

    if run and st.session_state.dev_file_bytes:
        book = read_book_bytes(st.session_state.dev_file_bytes)
        sh_train = _find_sheet(book, ["Train", "Training", "training2", "train", "training"])
        sh_test  = _find_sheet(book, ["Test", "Testing", "testing2", "test", "testing"])
        if sh_train is None or sh_test is None:
            st.markdown('<div class="st-message-box st-error">Workbook must include Train/Training and Test/Testing sheets.</div>', unsafe_allow_html=True)
            st.stop()

        tr0 = book[sh_train].copy()
        te0 = book[sh_test].copy()

        # Resolve target name per-sheet from aliases
        tcol_tr = _resolve_target_col(tr0)
        tcol_te = _resolve_target_col(te0)
        if tcol_tr is None or tcol_te is None:
            st.error(f"Missing target column. Expected one of: {TARGET_ALIASES}")
            st.stop()

        # Ensure feature columns exist
        if not (ensure_cols(tr0, FEATURES) and ensure_cols(te0, FEATURES)):
            st.markdown('<div class="st-message-box st-error">Missing required feature columns.</div>', unsafe_allow_html=True)
            st.stop()

        # Prepare X,y
        X_tr = _make_X(tr0, FEATURES)
        y_tr = pd.to_numeric(tr0[tcol_tr], errors="coerce")
        X_te = _make_X(te0, FEATURES)
        y_te = pd.to_numeric(te0[tcol_te], errors="coerce")

        # Drop rows with NA in y
        mask_tr = np.isfinite(y_tr); X_tr, y_tr = X_tr.loc[mask_tr], y_tr.loc[mask_tr]
        mask_te = np.isfinite(y_te); X_te, y_te = X_te.loc[mask_te], y_te.loc[mask_te]

        pipe = build_pipeline()
        pipe.fit(X_tr, y_tr)
        st.session_state.fitted_model = pipe

        # Predictions
        tr = tr0.copy(); te = te0.copy()
        tr[PRED_COL] = _inv_transform(pipe.predict(_make_X(tr0, FEATURES)), TRANSFORM)
        te[PRED_COL] = _inv_transform(pipe.predict(_make_X(te0, FEATURES)), TRANSFORM)

        # Save results & metrics
        st.session_state.results["Train"] = tr
        st.session_state.results["Test"]  = te
        st.session_state.results["m_train"] = {
            "R":     pearson_r(tr[tcol_tr], tr[PRED_COL]),
            "RMSE":  rmse(tr[tcol_tr], tr[PRED_COL]),
            "MAPE%": mape(tr[tcol_tr], tr[PRED_COL]),
        }
        st.session_state.results["m_test"] = {
            "R":     pearson_r(te[tcol_te], te[PRED_COL]),
            "RMSE":  rmse(te[tcol_te], te[PRED_COL]),
            "MAPE%": mape(te[tcol_te], te[PRED_COL]),
        }

        # Persist used target names (for export/plots)
        st.session_state["tcol_train"] = tcol_tr
        st.session_state["tcol_test"]  = tcol_te

        # Training min–max ranges
        tr_min = tr[FEATURES].min().to_dict()
        tr_max = tr[FEATURES].max().to_dict()
        st.session_state.train_ranges = {f: (float(tr_min[f]), float(tr_max[f])) for f in FEATURES}

        st.markdown('<div class="st-message-box st-success">Case has been built and results are displayed below.</div>', unsafe_allow_html=True)

    def _dev_block(df: pd.DataFrame, actual_col: str, m: dict):
        c1, c2, c3 = st.columns(3)
        c1.metric("R",     f"{m['R']:.3f}")
        c2.metric("RMSE",  f"{m['RMSE']:.2f}")
        c3.metric("MAPE%", f"{m['MAPE%']:.2f}")
        st.markdown("""
            <div style='text-align: left; font-size: 0.8em; color: #6b7280; margin-top: -16px; margin-bottom: 8px;'>
                <strong>R:</strong> Pearson Correlation Coefficient<br>
                <strong>RMSE:</strong> Root Mean Square Error<br>
                <strong>MAPE:</strong> Mean Absolute Percentage Error
            </div>
        """, unsafe_allow_html=True)
        col_track, col_cross = st.columns([2, 3], gap="large")
        with col_track:
            st.plotly_chart(
                track_plot(df, actual_col, include_actual=True),
                use_container_width=False,
                config={"displayModeBar": False, "scrollZoom": True}
            )
        with col_cross:
            st.pyplot(
                cross_plot_static(df[actual_col], df[PRED_COL], label="Fracture Pressure"),
                use_container_width=False
            )

    if "Train" in st.session_state.results or "Test" in st.session_state.results:
        tab1, tab2 = st.tabs(["Training", "Testing"])
        if "Train" in st.session_state.results:
            with tab1:
                _dev_block(st.session_state.results["Train"], st.session_state.get("tcol_train", TARGET_CANON), st.session_state.results["m_train"])
        if "Test" in st.session_state.results:
            with tab2:
                _dev_block(st.session_state.results["Test"], st.session_state.get("tcol_test", TARGET_CANON),  st.session_state.results["m_test"])
        render_export_button(phase_key="dev")

# =========================
# VALIDATION (with actual)
# =========================
if st.session_state.app_step == "validate":
    st.sidebar.header("Validate the Model")
    up = st.sidebar.file_uploader("Upload Validation Excel", type=["xlsx", "xls"])
    if up is not None:
        book = read_book_bytes(up.getvalue())
        if book:
            df0 = next(iter(book.values()))
            st.sidebar.caption(f"**Data loaded:** {up.name} • {df0.shape[0]} rows × {df0.shape[1]} cols")

    # Preview button
    if st.sidebar.button("Preview data", use_container_width=True, disabled=(up is None)):
        st.session_state.preview_book = read_book_bytes(up.getvalue()) if up is not None else {}
        st.session_state.show_preview_panel = True
        st.rerun()

    go_btn = st.sidebar.button("Predict & Validate", type="primary", use_container_width=True)
    if st.sidebar.button("⬅ Back to Case Building", use_container_width=True):
        st.session_state.app_step = "dev"; st.rerun()
    if st.sidebar.button("Proceed to Prediction ▶", use_container_width=True):
        st.session_state.app_step = "predict"; st.rerun()

    sticky_header("Validate the Model", "Upload a dataset with the same **features** and an **actual fracture pressure** column.")
    render_preview_panel()

    if go_btn and up is not None:
        if st.session_state.fitted_model is None:
            st.error("Please train the model first in Case Building.")
            st.stop()

        book = read_book_bytes(up.getvalue())
        names = list(book.keys())
        name = next((s for s in names if s.lower() in ("validation", "validate", "validation2", "val", "val2")), names[0])
        df0 = book[name].copy()

        tcol = _resolve_target_col(df0)
        if tcol is None:
            st.error(f"Missing target column. Expected one of: {TARGET_ALIASES}")
            st.stop()
        if not ensure_cols(df0, FEATURES):
            st.markdown('<div class="st-message-box st-error">Missing required feature columns.</div>', unsafe_allow_html=True)
            st.stop()

        df = df0.copy()
        df[PRED_COL] = _inv_transform(st.session_state.fitted_model.predict(_make_X(df0, FEATURES)), TRANSFORM)
        st.session_state.results["Validate"] = df

        # Range checks
        ranges = st.session_state.train_ranges
        oor_pct = 0.0
        tbl = None
        if ranges:
            any_viol = pd.DataFrame({f: (df[f] < ranges[f][0]) | (df[f] > ranges[f][1]) for f in FEATURES}).any(axis=1)
            oor_pct = float(any_viol.mean() * 100.0)
            if any_viol.any():
                tbl = df.loc[any_viol, FEATURES].copy()
                for c in FEATURES:
                    if pd.api.types.is_numeric_dtype(tbl[c]):
                        tbl[c] = tbl[c].round(3)
                tbl["Violations"] = pd.DataFrame({f: (df[f] < ranges[f][0]) | (df[f] > ranges[f][1]) for f in FEATURES}).loc[any_viol].apply(
                    lambda r: ", ".join([c for c, v in r.items() if v]), axis=1
                )

        st.session_state.results["m_val"] = {
            "R":     pearson_r(df[tcol], df[PRED_COL]),
            "RMSE":  rmse(df[tcol], df[PRED_COL]),
            "MAPE%": mape(df[tcol], df[PRED_COL]),
        }
        st.session_state.results["sv_val"] = {"n": len(df), "pred_min": float(df[PRED_COL].min()), "pred_max": float(df[PRED_COL].max()), "oor": oor_pct}
        st.session_state.results["oor_tbl"] = tbl
        st.session_state["tcol_val"] = tcol

    if "Validate" in st.session_state.results:
        m = st.session_state.results["m_val"]
        tcol = st.session_state.get("tcol_val", TARGET_CANON)
        c1, c2, c3 = st.columns(3)
        c1.metric("R", f"{m['R']:.3f}"); c2.metric("RMSE", f"{m['RMSE']:.2f}"); c3.metric("MAPE%", f"{m['MAPE%']:.2f}")
        st.markdown("""
            <div style='text-align: left; font-size: 0.8em; color: #6b7280; margin-top: -16px; margin-bottom: 8px;'>
                <strong>R:</strong> Pearson Correlation Coefficient<br>
                <strong>RMSE:</strong> Root Mean Square Error<br>
                <strong>MAPE:</strong> Mean Absolute Percentage Error
            </div>
        """, unsafe_allow_html=True)

        col_track, col_cross = st.columns([2, 3], gap="large")
        with col_track:
            st.plotly_chart(
                track_plot(st.session_state.results["Validate"], tcol, include_actual=True),
                use_container_width=False, config={"displayModeBar": False, "scrollZoom": True}
            )
        with col_cross:
            st.pyplot(
                cross_plot_static(st.session_state.results["Validate"][tcol], st.session_state.results["Validate"][PRED_COL], label="Fracture Pressure"),
                use_container_width=False
            )

        render_export_button(phase_key="validate")

        sv = st.session_state.results["sv_val"]
        if sv["oor"] > 0:
            st.markdown('<div class="st-message-box st-warning">Some inputs fall outside **training min–max** ranges.</div>', unsafe_allow_html=True)
        if st.session_state.results["oor_tbl"] is not None:
            st.write("*Out-of-range rows (vs. Training min–max):*")
            df_centered_rounded(st.session_state.results["oor_tbl"])

# =========================
# PREDICTION (no actual)
# =========================
if st.session_state.app_step == "predict":
    st.sidebar.header("Prediction (No Actual)")
    up = st.sidebar.file_uploader("Upload Prediction Excel", type=["xlsx", "xls"])
    if up is not None:
        book = read_book_bytes(up.getvalue())
        if book:
            df0 = next(iter(book.values()))
            st.sidebar.caption(f"**Data loaded:** {up.name} • {df0.shape[0]} rows × {df0.shape[1]} cols")

    # Preview button
    if st.sidebar.button("Preview data", use_container_width=True, disabled=(up is None)):
        st.session_state.preview_book = read_book_bytes(up.getvalue()) if up is not None else {}
        st.session_state.show_preview_panel = True
        st.rerun()

    go_btn = st.sidebar.button("Predict", type="primary", use_container_width=True)
    if st.sidebar.button("⬅ Back to Case Building", use_container_width=True):
        st.session_state.app_step = "dev"; st.rerun()

    sticky_header("Prediction", "Upload a dataset with the 5 feature columns (no actual column).")
    render_preview_panel()

    if go_btn and up is not None:
        if st.session_state.fitted_model is None:
            st.error("Please train the model first in Case Building.")
            st.stop()

        book = read_book_bytes(up.getvalue())
        name = list(book.keys())[0]
        df0 = book[name].copy()
        if not ensure_cols(df0, FEATURES):
            st.markdown('<div class="st-message-box st-error">Missing required columns.</div>', unsafe_allow_html=True)
            st.stop()

        df = df0.copy()
        df[PRED_COL] = _inv_transform(st.session_state.fitted_model.predict(_make_X(df0, FEATURES)), TRANSFORM)
        st.session_state.results["PredictOnly"] = df

        ranges = st.session_state.train_ranges
        oor_pct = 0.0
        if ranges:
            any_viol = pd.DataFrame({f: (df[f] < ranges[f][0]) | (df[f] > ranges[f][1]) for f in FEATURES}).any(axis=1)
            oor_pct = float(any_viol.mean() * 100.0)
        st.session_state.results["sv_pred"] = {
            "n": len(df),
            "pred_min": float(df[PRED_COL].min()),
            "pred_max": float(df[PRED_COL].max()),
            "pred_mean": float(df[PRED_COL].mean()),
            "pred_std": float(df[PRED_COL].std(ddof=0)),
            "oor": oor_pct
        }

    if "PredictOnly" in st.session_state.results:
        df = st.session_state.results["PredictOnly"]
        sv = st.session_state.results["sv_pred"]
        col_left, col_right = st.columns([2, 3], gap="large")
        with col_left:
            table = pd.DataFrame({
                "Metric": ["# points", "Pred min", "Pred max", "Pred mean", "Pred std", "OOR %"],
                "Value": [sv["n"], round(sv["pred_min"], 3), round(sv["pred_max"], 3),
                          round(sv["pred_mean"], 3), round(sv["pred_std"], 3), f'{sv["oor"]:.1f}%']
            })
            st.markdown('<div class="st-message-box st-success">Predictions ready ✓</div>', unsafe_allow_html=True)
            df_centered_rounded(table, hide_index=True)
            st.caption("**★ OOR** = % of rows with input features outside the training min–max range.")
        with col_right:
            st.plotly_chart(
                track_plot(df, actual_col=None, include_actual=False),
                use_container_width=False, config={"displayModeBar": False, "scrollZoom": True}
            )
        render_export_button(phase_key="predict")

# =========================
# Footer
# =========================
st.markdown("""
<br><br><br>
<hr>
<div style='text-align:center;color:#6b7280;font-size:1.0em;'>
    © 2025 Smart Thinking AI-Solutions Team. All rights reserved.<br>
    Website: <a href="https://smartthinking.com.sa" target="_blank" rel="noopener noreferrer">smartthinking.com.sa</a>
</div>
""", unsafe_allow_html=True)