# app.py — ST_Log_GR (Gamma Ray) — UI aligned with TS/Tc apps

import io, json, os, base64, math
from pathlib import Path
import streamlit as st
import pandas as pd
import numpy as np
import joblib
from datetime import datetime

# Matplotlib (static plots: preview, 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, mean_absolute_error

# =========================
# Constants (GR)
# =========================
APP_NAME = "ST_Log_GR"
TAGLINE  = "Real-Time Gamma Ray Prediction"

FEATURES = ["GPM", "SPP", "RPM", "WOB", "T", "ROP"]

# Target used during training
TARGET = "log(GR)"           # set to your training target column name if different

# Inverse transform used to map predictions/target back to API
TARGET_TRANSFORM = "log10"   # "log10" for log10(GR); "ln" for ln(GR); "none" for raw

# Column with actual GR in API units (if present)
ACTUAL_COL = "GR"

MODELS_DIR = Path("models")
DEFAULT_MODEL = MODELS_DIR / "gr_rf.joblib"
MODEL_FALLBACKS = [MODELS_DIR / "model.joblib", MODELS_DIR / "model.pkl"]

COLORS = {"pred": "#1f77b4", "actual": "#f2b702", "ref": "#5a5a5a"}

# ---- Plot sizing controls ----
CROSS_W = 350
CROSS_H = 350
TRACK_H = 1000
TRACK_W = 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; }
 .main .block-container { overflow: unset !important; }
 div[data-testid="stVerticalBlock"] { overflow: unset !important; }

 /* Sticky preview expander + its tabs */
 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;
 }

 /* Hide uploader helper text */
 section[data-testid="stFileUploader"] div[data-testid="stMarkdownContainer"]{display:none !important;}
 section[data-testid="stFileUploader"] [data-testid="stFileUploaderDropzone"] > div:first-child{display:none !important;}
 section[data-testid="stFileUploader"] [data-testid="stFileUploaderInstructions"]{display:none !important;}
 section[data-testid="stFileUploader"] p, section[data-testid="stFileUploader"] small{display:none !important;}

 /* Message boxes */
 .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; }
</style>
""", unsafe_allow_html=True)

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

# =========================
# Password gate
# =========================
def inline_logo(path="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:
        st.warning("Set APP_PASSWORD in Secrets (or environment) and restart.")
        st.stop()

    if st.session_state.get("auth_ok", False):
        return

    st.sidebar.markdown(f"""
        <div class="centered-container">
            <img src="{inline_logo('logo.png')}" style="width: 200px; height: auto; object-fit: contain;">
            <div style='font-weight:800;font-size:1.2rem;'>{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 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 or np.all(a == a[0]) or np.all(p == p[0]): return float("nan")
    return float(np.corrcoef(a, p)[0, 1])

@st.cache_resource(show_spinner=False)
def load_model(model_path: str):
    return joblib.load(model_path)

@st.cache_data(show_spinner=False)
def parse_excel(data_bytes: bytes):
    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):
    return parse_excel(b) if b else {}

def normalize_df(df: pd.DataFrame) -> pd.DataFrame:
    out = df.copy()
    out.columns = [str(c).strip().replace("  ", " ") for c in out.columns]
    return out

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 find_sheet(book, names):
    low2orig = {k.lower(): k for k in book.keys()}
    for nm in names:
        if nm.lower() in low2orig: return low2orig[nm.lower()]
    return None

def _nice_tick0(xmin: float, step: int = 5) -> float:
    return step * math.floor(xmin / step) if np.isfinite(xmin) else xmin

def df_centered_rounded(df: pd.DataFrame, hide_index=True):
    out = df.copy()
    numcols = out.select_dtypes(include=[np.number]).columns
    styler = (
        out.style
          .format({c: "{:.2f}" 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)

# --- target transform helpers (to support models trained on log(GR)) ---
def inverse_target(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", "log_e", "natural"]:
        return np.exp(x)
    return x  # "none"

def to_actual_series(df: pd.DataFrame, target_col: str, actual_col_hint: str, transform: str) -> pd.Series:
    # Prefer explicit GR column if available; else invert target
    if actual_col_hint and actual_col_hint in df.columns:
        return pd.Series(df[actual_col_hint], dtype=float)
    if target_col in df.columns:
        return pd.Series(inverse_target(np.asarray(df[target_col], dtype=float), transform), dtype=float)
    if "GR" in df.columns:
        return pd.Series(df["GR"], dtype=float)
    raise ValueError("Cannot find actual GR column or target to invert.")

# === Excel export helpers (TS/Tc-style multiselect) =======================
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 = 2) -> 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 _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, 2)

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, 2)

def _excel_autofit(writer, sheet_name: str, df: pd.DataFrame, min_w: int = 8, max_w: int = 40):
    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 = []
    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","Prediction_OOR"]
    if st.session_state.get("train_ranges"): sections += ["Training_Ranges"]
    sections += ["Info"]
    return sections

def build_export_workbook(selected: list[str], ndigits: int = 2, 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] = []

    # Training
    if "Training" in selected and "Train" in res:
        sheets["Training"] = _round_numeric(res["Train"], ndigits); order.append("Training")
    if "Training_Metrics" in selected and res.get("m_train"):
        sheets["Training_Metrics"] = _round_numeric(pd.DataFrame([res["m_train"]]), ndigits); order.append("Training_Metrics")
    if "Training_Summary" in selected and "Train" in res:
        tr_cols = FEATURES + [c for c in ["GR_Actual","GR_Pred"] if c in res["Train"].columns]
        s = _summary_table(res["Train"], tr_cols)
        if not s.empty:
            sheets["Training_Summary"] = s; order.append("Training_Summary")

    # Testing
    if "Testing" in selected and "Test" in res:
        sheets["Testing"] = _round_numeric(res["Test"], ndigits); order.append("Testing")
    if "Testing_Metrics" in selected and res.get("m_test"):
        sheets["Testing_Metrics"] = _round_numeric(pd.DataFrame([res["m_test"]]), ndigits); order.append("Testing_Metrics")
    if "Testing_Summary" in selected and "Test" in res:
        te_cols = FEATURES + [c for c in ["GR_Actual","GR_Pred"] if c in res["Test"].columns]
        s = _summary_table(res["Test"], te_cols)
        if not s.empty:
            sheets["Testing_Summary"] = s; order.append("Testing_Summary")

    # Validation
    if "Validation" in selected and "Validate" in res:
        sheets["Validation"] = _round_numeric(res["Validate"], ndigits); order.append("Validation")
    if "Validation_Metrics" in selected and res.get("m_val"):
        sheets["Validation_Metrics"] = _round_numeric(pd.DataFrame([res["m_val"]]), ndigits); order.append("Validation_Metrics")
    if "Validation_Summary" in selected and res.get("sv_val"):
        sheets["Validation_Summary"] = _round_numeric(pd.DataFrame([res["sv_val"]]), ndigits); order.append("Validation_Summary")
    if "Validation_OOR" in selected and isinstance(res.get("oor_tbl"), pd.DataFrame) and not res["oor_tbl"].empty:
        sheets["Validation_OOR"] = _round_numeric(res["oor_tbl"].reset_index(drop=True), ndigits); order.append("Validation_OOR")

    # Prediction
    if "Prediction" in selected and "PredictOnly" in res:
        sheets["Prediction"] = _round_numeric(res["PredictOnly"], ndigits); order.append("Prediction")
    if "Prediction_Summary" in selected and res.get("sv_pred"):
        sheets["Prediction_Summary"] = _round_numeric(pd.DataFrame([res["sv_pred"]]), ndigits); order.append("Prediction_Summary")
    if "Prediction_OOR" in selected and isinstance(res.get("oor_tbl_pred"), pd.DataFrame) and not res["oor_tbl_pred"].empty:
        sheets["Prediction_OOR"] = _round_numeric(res["oor_tbl_pred"].reset_index(drop=True), ndigits); order.append("Prediction_OOR")

    # Training ranges
    if "Training_Ranges" in selected and st.session_state.get("train_ranges"):
        sheets["Training_Ranges"] = _train_ranges_df(st.session_state["train_ranges"]); order.append("Training_Ranges")

    # Info
    if "Info" in selected:
        info = pd.DataFrame([
            {"Key": "AppName",          "Value": APP_NAME},
            {"Key": "Tagline",          "Value": TAGLINE},
            {"Key": "Target",           "Value": TARGET},
            {"Key": "TargetTransform",  "Value": TARGET_TRANSFORM},
            {"Key": "ActualColumn",     "Value": ACTUAL_COL},
            {"Key": "Features",         "Value": ", ".join(FEATURES)},
            {"Key": "ExportedAt",       "Value": datetime.now().strftime("%Y-%m-%d %H:%M:%S")},
        ])
        sheets["Info"] = info; order.append("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)
            _excel_autofit(writer, sheet, df)
    bio.seek(0)
    fname = f"GR_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 to include in the Excel export.",
        key=f"sheets_{phase_key}",
    )

    if not selected_sheets:
        st.caption("Select one or more sheets above to enable the export.")
        st.download_button("⬇️ Export Excel", data=b"", file_name="GR_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=2, do_autofit=True)
    if names:
        st.caption("Will include: " + ", ".join(names))
    st.download_button(
        "⬇️ Export Excel",
        data=(data or b""),
        file_name=(fname or "GR_Export.xlsx"),
        mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
        disabled=(data is None),
        key=f"download_{phase_key}",
    )

# =========================
# Cross plot (Matplotlib)
# =========================
def _nice_bounds(arr_min, arr_max, n_ticks=6):
    if not np.isfinite(arr_min) or not np.isfinite(arr_max):
        return 0.0, 100.0, 20.0
    span = arr_max - arr_min
    if span <= 0:
        return max(arr_min-5, 0), arr_max+5, 5.0
    raw_step = span / max(n_ticks, 1)
    mag = 10 ** math.floor(math.log10(raw_step))
    steps = np.array([1, 2, 2.5, 5, 10]) * mag
    step = steps[np.argmin(np.abs(steps - raw_step))]
    lo = step * math.floor(arr_min / step)
    hi = step * math.ceil(arr_max / step)
    return float(lo), float(hi), float(step)

def cross_plot_static(actual, pred):
    a = pd.Series(actual, dtype=float)
    p = pd.Series(pred,   dtype=float)

    lo = min(a.min(), p.min())
    hi = max(a.max(), p.max())
    fixed_min, fixed_max, step = _nice_bounds(lo, hi, n_ticks=6)
    ticks = np.arange(fixed_min, fixed_max + step, step)

    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([fixed_min, fixed_max], [fixed_min, fixed_max],
            linestyle="--", linewidth=1.2, color=COLORS["ref"])

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

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

    ax.set_xlabel("Actual GR (API)",    fontweight="bold", fontsize=10, color="black")
    ax.set_ylabel("Predicted GR (API)", fontweight="bold", fontsize=10, color="black")
    ax.tick_params(labelsize=8, 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

# =========================
# Track plot (Plotly) — y-axis reversed
# =========================
def track_plot(df, include_actual=True, pred_col="GR_Pred", actual_col="GR"):
    def _col_1d(frame: pd.DataFrame, col: str) -> pd.Series:
        if col not in frame.columns: return pd.Series(dtype=float)
        v = frame[col]
        if isinstance(v, pd.DataFrame): v = v.iloc[:, 0]
        return pd.Series(v, dtype=float)

    depth_col = next((c for c in df.columns if 'depth' in str(c).lower()), None)
    if depth_col is not None:
        y = pd.Series(df[depth_col]).astype(float); ylab = depth_col
    else:
        y = pd.Series(np.arange(1, len(df) + 1), dtype=float); ylab = "Point Index"

    x_pred = _col_1d(df, pred_col)
    if include_actual and actual_col in df.columns:
        x_act = _col_1d(df, actual_col)
        x_series = pd.concat([x_pred, x_act], ignore_index=True)
    else:
        x_act = None
        x_series = x_pred

    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=5)

    fig = go.Figure()
    fig.add_trace(go.Scatter(
        x=x_pred, y=y, mode="lines",
        line=dict(color=COLORS["pred"], width=1.8),
        name=pred_col,
        hovertemplate="GR_Pred: %{x:.0f}<br>"+ylab+": %{y}<extra></extra>"
    ))
    if include_actual and x_act is not None:
        fig.add_trace(go.Scatter(
            x=x_act, y=y, mode="lines",
            line=dict(color=COLORS["actual"], width=2.0, dash="dot"),
            name="GR (actual)",
            hovertemplate="GR (actual): %{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="GR (API)",
        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"),
        autorange="reversed",
        ticks="outside",
        showline=True, linewidth=1.2, linecolor="#444", mirror=True,
        showgrid=True, gridcolor="rgba(0,0,0,0.12)", automargin=True,
    )
    return fig

# ---------- Preview (Matplotlib) — colorful tracks; shared Y; ticks only left ----------
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()), None)
    if depth_col is not None:
        y = pd.to_numeric(df[depth_col], errors="coerce")
        ylab = depth_col
    else:
        y = pd.Series(np.arange(1, len(df) + 1), dtype=float)
        ylab = "Point Index"

    # Stable qualitative colors
    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=(2.3 * n, 7.0), sharey=True, dpi=100)
    if n == 1:
        axes = [axes]

    y_min, y_max = float(np.nanmin(y)), float(np.nanmax(y))

    for i, (ax, col) in enumerate(zip(axes, cols)):
        x = pd.to_numeric(df[col], errors="coerce")
        ax.plot(x, y, '-', lw=1.8, 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)  # reverse Y (Depth down)
        ax.grid(True, linestyle=":", alpha=0.3)
        if i == 0:
            ax.set_ylabel(ylab)
        else:
            # Hide Y ticks and labels for non-left tracks
            ax.tick_params(axis='y', left=False, labelleft=False)

    fig.tight_layout()
    return fig

# =========================
# Load model + meta
# =========================
def ensure_model() -> Path|None:
    for p in [DEFAULT_MODEL, *MODEL_FALLBACKS]:
        if p.exists() and p.stat().st_size > 0:
            return p
    url = os.environ.get("MODEL_URL", "")
    if not url:
        return None
    try:
        import requests
        DEFAULT_MODEL.parent.mkdir(parents=True, exist_ok=True)
        with requests.get(url, stream=True, timeout=30) as r:
            r.raise_for_status()
            with open(DEFAULT_MODEL, "wb") as f:
                for chunk in r.iter_content(1<<20):
                    if chunk: f.write(chunk)
        return DEFAULT_MODEL
    except Exception:
        return None

mpath = ensure_model()
if not mpath:
    st.error("Model not found. Upload models/gr_rf.joblib (or set MODEL_URL).")
    st.stop()
try:
    model = load_model(str(mpath))
except Exception as e:
    st.error(f"Failed to load model: {e}")
    st.stop()

meta_path = MODELS_DIR / "meta.json"
if meta_path.exists():
    try:
        meta = json.loads(meta_path.read_text(encoding="utf-8"))
        FEATURES         = meta.get("features", FEATURES)
        TARGET           = meta.get("target", TARGET)
        TARGET_TRANSFORM = meta.get("target_transform", TARGET_TRANSFORM)
        ACTUAL_COL       = meta.get("actual_col", ACTUAL_COL)
    except Exception:
        pass

# =========================
# Session state
# =========================
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("dev_preview",False)
st.session_state.setdefault("show_preview_modal", False)

# =========================
# Sidebar branding
# =========================
st.sidebar.markdown(f"""
    <div class="centered-container">
        <img src="{inline_logo('logo.png')}" style="width: 200px; height: auto; object-fit: contain;">
        <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, message):
    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
    )

# =========================
# INTRO
# =========================
if st.session_state.app_step == "intro":
    st.header("Welcome!")
    st.markdown("This software is developed by *Smart Thinking AI-Solutions Team* to estimate **Gamma Ray (GR)** from drilling data.")
    st.subheader("How It Works")
    st.markdown(
        "1) **Upload your data to build the case and preview model performance.**  \n"
        "2) Click **Run Model** to compute metrics and plots.  \n"
        "3) **Proceed to Validation** (with actual GR) or **Proceed to Prediction** (no GR)."
    )
    if st.button("Start Showcase", type="primary"):
        st.session_state.app_step = "dev"; st.rerun()

# =========================
# CASE BUILDING
# =========================
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.dev_preview = False
    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")

    if st.sidebar.button("Preview data", use_container_width=True, disabled=not st.session_state.dev_file_loaded):
        st.session_state.show_preview_modal = True
        st.session_state.dev_preview = True

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

    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()

        tr = normalize_df(book[sh_train].copy())
        te = normalize_df(book[sh_test].copy())
        if not (ensure_cols(tr, FEATURES) and ensure_cols(te, FEATURES)):
            st.markdown('<div class="st-message-box st-error">Missing required feature columns.</div>', unsafe_allow_html=True)
            st.stop()

        # Predict (model trained on transformed target)
        tr_pred_raw = model.predict(tr[FEATURES])
        te_pred_raw = model.predict(te[FEATURES])
        tr["GR_Pred"] = inverse_target(np.asarray(tr_pred_raw, dtype=float), TARGET_TRANSFORM)
        te["GR_Pred"] = inverse_target(np.asarray(te_pred_raw, dtype=float), TARGET_TRANSFORM)

        # Actual GR for metrics/plots
        tr["GR_Actual"] = to_actual_series(tr, TARGET, ACTUAL_COL, TARGET_TRANSFORM)
        te["GR_Actual"] = to_actual_series(te, TARGET, ACTUAL_COL, TARGET_TRANSFORM)

        st.session_state.results["Train"]=tr; st.session_state.results["Test"]=te
        st.session_state.results["m_train"]={
            "R": pearson_r(tr["GR_Actual"], tr["GR_Pred"]),
            "RMSE": rmse(tr["GR_Actual"], tr["GR_Pred"]),
            "MAE": mean_absolute_error(tr["GR_Actual"], tr["GR_Pred"])
        }
        st.session_state.results["m_test"]={
            "R": pearson_r(te["GR_Actual"], te["GR_Pred"]),
            "RMSE": rmse(te["GR_Actual"], te["GR_Pred"]),
            "MAE": mean_absolute_error(te["GR_Actual"], te["GR_Pred"])
        }

        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, m):
        c1, c2, c3 = st.columns(3)
        c1.metric("R",    f"{m['R']:.3f}")
        c2.metric("RMSE", f"{m['RMSE']:.3f}")
        c3.metric("MAE",  f"{m['MAE']:.3f}")
        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>MAE:</strong> Mean Absolute 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, include_actual=True, pred_col="GR_Pred", actual_col="GR_Actual"),
                use_container_width=False,
                config={"displayModeBar": False, "scrollZoom": True},
            )
        with col_cross:
            st.pyplot(cross_plot_static(df["GR_Actual"], df["GR_Pred"]), 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.results["m_train"])
        if "Test" in st.session_state.results:
            with tab2: _dev_block(st.session_state.results["Test"],  st.session_state.results["m_test"])
        render_export_button(phase_key="dev")

# =========================
# VALIDATION (with actual GR)
# =========================
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")
    if st.sidebar.button("Preview data", use_container_width=True, disabled=(up is None)):
        st.session_state.show_preview_modal = True
    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 **GR** to evaluate performance.")

    if go_btn and up is not None:
        book = read_book_bytes(up.getvalue())
        name = find_sheet(book, ["Validation","Validate","validation2","Val","val"]) or list(book.keys())[0]
        df = normalize_df(book[name].copy())
        if not ensure_cols(df, FEATURES):
            st.markdown('<div class="st-message-box st-error">Missing required feature columns.</div>', unsafe_allow_html=True); st.stop()

        pred_raw = model.predict(df[FEATURES])
        df["GR_Pred"] = inverse_target(np.asarray(pred_raw, dtype=float), TARGET_TRANSFORM)
        try:
            df["GR_Actual"] = to_actual_series(df, TARGET, ACTUAL_COL, TARGET_TRANSFORM)
        except Exception:
            st.markdown('<div class="st-message-box st-error">Validation sheet must include actual GR (or a target column that can be inverse-transformed).</div>', unsafe_allow_html=True)
            st.stop()

        st.session_state.results["Validate"]=df

        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(2)
                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["GR_Actual"], df["GR_Pred"]),
            "RMSE": rmse(df["GR_Actual"], df["GR_Pred"]),
            "MAE": mean_absolute_error(df["GR_Actual"], df["GR_Pred"])
        }
        st.session_state.results["sv_val"]={"n":len(df),"pred_min":float(df["GR_Pred"].min()),"pred_max":float(df["GR_Pred"].max()),"oor":oor_pct}
        st.session_state.results["oor_tbl"]=tbl

    if "Validate" in st.session_state.results:
        m = st.session_state.results["m_val"]
        c1,c2,c3 = st.columns(3)
        c1.metric("R", f"{m['R']:.3f}"); c2.metric("RMSE", f"{m['RMSE']:.2f}"); c3.metric("MAE", f"{m['MAE']:.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>MAE:</strong> Mean Absolute 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"], include_actual=True, pred_col="GR_Pred", actual_col="GR_Actual"),
                use_container_width=False, config={"displayModeBar": False, "scrollZoom": True}
            )
        with col_cross:
            st.pyplot(cross_plot_static(st.session_state.results["Validate"]["GR_Actual"],
                                        st.session_state.results["Validate"]["GR_Pred"]),
                      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 GR)
# =========================
if st.session_state.app_step == "predict":
    st.sidebar.header("Prediction (No Actual GR)")
    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")
    if st.sidebar.button("Preview data", use_container_width=True, disabled=(up is None)):
        st.session_state.show_preview_modal = True

    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 feature columns (no **GR**).")

    if go_btn and up is not None:
        book = read_book_bytes(up.getvalue()); name = list(book.keys())[0]
        df = normalize_df(book[name].copy())
        if not ensure_cols(df, FEATURES):
            st.markdown('<div class="st-message-box st-error">Missing required feature columns.</div>', unsafe_allow_html=True); st.stop()

        pred_raw = model.predict(df[FEATURES])
        df["GR_Pred"] = inverse_target(np.asarray(pred_raw, dtype=float), TARGET_TRANSFORM)
        st.session_state.results["PredictOnly"]=df

        ranges = st.session_state.train_ranges; oor_pct = 0.0; oor_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():
                oor_tbl = df.loc[any_viol, FEATURES].copy()
                for c in FEATURES:
                    if pd.api.types.is_numeric_dtype(oor_tbl[c]): oor_tbl[c] = oor_tbl[c].round(2)
                oor_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["sv_pred"]={
            "n":len(df),
            "pred_min":float(df["GR_Pred"].min()),
            "pred_max":float(df["GR_Pred"].max()),
            "pred_mean":float(df["GR_Pred"].mean()),
            "pred_std":float(df["GR_Pred"].std(ddof=0)),
            "oor":oor_pct
        }
        st.session_state.results["oor_tbl_pred"] = oor_tbl

    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"],2), round(sv["pred_max"],2),
                           round(sv["pred_mean"],2), round(sv["pred_std"],2), 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 whose input features fall outside the training min–max range.")

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

        with col_right:
            st.plotly_chart(
                track_plot(df, include_actual=False, pred_col="GR_Pred", actual_col="GR"),
                use_container_width=False, config={"displayModeBar": False, "scrollZoom": True}
            )

        render_export_button(phase_key="predict")

# =========================
# Preview modal (re-usable)
# =========================
if st.session_state.show_preview_modal:
    book_to_preview = {}
    if st.session_state.app_step == "dev":
        book_to_preview = read_book_bytes(st.session_state.dev_file_bytes)
    elif st.session_state.app_step in ["validate", "predict"] and 'up' in locals() and up is not None:
        book_to_preview = read_book_bytes(up.getvalue())

    with st.expander("Preview data", expanded=True):
        if not book_to_preview:
            st.markdown('<div class="st-message-box">No data loaded yet.</div>', unsafe_allow_html=True)
        else:
            names = list(book_to_preview.keys())
            tabs = st.tabs(names)
            for t, name in zip(tabs, names):
                with t:
                    df = normalize_df(book_to_preview[name])
                    t1, t2 = st.tabs(["Tracks", "Summary"])
                    with t1:
                        present = [c for c in FEATURES if c in df.columns]
                        if present:
                            st.pyplot(preview_tracks(df, present), use_container_width=True)
                        else:
                            st.info(f"No expected feature columns found. Expected any of: {FEATURES}. Found: {list(df.columns)}")
                    with t2:
                        present = [c for c in FEATURES if c in df.columns]
                        if present:
                            tbl = (df[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)
                        else:
                            st.info("No expected feature columns found to summarize.")
    st.session_state.show_preview_modal = False

# =========================
# 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)