diff --git a/test1234.ipynb b/test1234.ipynb
new file mode 100644
index 0000000..67f539b
--- /dev/null
+++ b/test1234.ipynb
@@ -0,0 +1,516 @@
+{
+ "cells": [
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "id": "fa5f8ef5-ada5-4316-9344-f38bddcc40c2",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import numpy as np\n",
+ "import pandas as pd\n",
+ "import matplotlib.pyplot as plt\n",
+ "import torch\n",
+ "import torch.nn as nn\n",
+ "from sklearn.preprocessing import StandardScaler\n",
+ "\n",
+ "torch.manual_seed(42)\n",
+ "np.random.seed(42)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "id": "a94d4039-fc00-4153-8070-4e1b4a7d2899",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ " Date Upstream Depth (ft) Downstream Depth (ft) \\\n",
+ "0 2016-01-01 1.624416 7.003261 \n",
+ "1 2016-01-02 1.605676 7.059753 \n",
+ "2 2016-01-03 1.611514 7.043261 \n",
+ "3 2016-01-04 1.608600 7.000180 \n",
+ "4 2016-01-05 1.605676 6.966737 \n",
+ "\n",
+ " Upstream Mean Discharge (cfs) Downstream Mean Discharge (cfs) \\\n",
+ "0 137.0 146.0 \n",
+ "1 134.0 147.0 \n",
+ "2 136.0 146.0 \n",
+ "3 135.0 144.0 \n",
+ "4 134.0 145.0 \n",
+ "\n",
+ " Upstream Width (ft) Downstream Width (ft) \n",
+ "0 58.532544 56.496681 \n",
+ "1 58.455077 56.519038 \n",
+ "2 58.506900 56.496681 \n",
+ "3 58.481079 56.451532 \n",
+ "4 58.455077 56.474180 \n",
+ "shape: (3653, 7)\n",
+ "missing values by column:\n",
+ "Date 0\n",
+ "Upstream Depth (ft) 4\n",
+ "Downstream Depth (ft) 62\n",
+ "Upstream Mean Discharge (cfs) 4\n",
+ "Downstream Mean Discharge (cfs) 4\n",
+ "Upstream Width (ft) 4\n",
+ "Downstream Width (ft) 4\n",
+ "dtype: int64\n",
+ "missing calendar dates: 0\n"
+ ]
+ }
+ ],
+ "source": [
+ "# Load the daily data\n",
+ "df = pd.read_csv(\"Data.csv\")\n",
+ "df[\"Date\"] = pd.to_datetime(df[\"Date\"])\n",
+ "df = df.sort_values(\"Date\").reset_index(drop=True)\n",
+ "\n",
+ "print(df.head())\n",
+ "print(\"shape:\", df.shape)\n",
+ "print(\"missing values by column:\")\n",
+ "print(df.isna().sum())\n",
+ "\n",
+ "# Check if any calendar dates are missing\n",
+ "all_days = pd.date_range(df[\"Date\"].min(), df[\"Date\"].max(), freq=\"D\")\n",
+ "missing_dates = all_days.difference(df[\"Date\"])\n",
+ "print(\"missing calendar dates:\", len(missing_dates))"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 5,
+ "id": "772e058c-fee7-4f74-81d0-0f74f0b35b0b",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ " start end days year\n",
+ "0 2017-05-11 2017-05-17 7 2017\n",
+ "1 2017-08-15 2017-08-20 6 2017\n",
+ "2 2019-06-10 2019-06-23 14 2019\n",
+ "3 2019-08-15 2019-08-21 7 2019\n",
+ "4 2023-01-21 2023-01-24 4 2023\n",
+ "5 2023-05-04 2023-05-04 1 2023\n",
+ "6 2023-07-26 2023-07-30 5 2023\n",
+ "7 2024-08-03 2024-08-06 4 2024\n",
+ "8 2024-08-08 2024-08-15 8 2024\n",
+ "9 2024-08-17 2024-08-18 2 2024\n",
+ "10 2025-12-19 2025-12-22 4 2025\n"
+ ]
+ }
+ ],
+ "source": [
+ "# Quick look at missing downstream depth dates\n",
+ "depth_missing = df[\"Downstream Depth (ft)\"].isna()\n",
+ "\n",
+ "# Find consecutive spans\n",
+ "spans = []\n",
+ "start = None\n",
+ "prev = None\n",
+ "for date, is_missing in zip(df[\"Date\"], depth_missing):\n",
+ " if is_missing and start is None:\n",
+ " start = date\n",
+ " if (not is_missing) and (start is not None):\n",
+ " spans.append((start, prev, (prev - start).days + 1))\n",
+ " start = None\n",
+ " prev = date\n",
+ "if start is not None:\n",
+ " spans.append((start, prev, (prev - start).days + 1))\n",
+ "\n",
+ "spans_df = pd.DataFrame(spans, columns=[\"start\", \"end\", \"days\"])\n",
+ "spans_df[\"year\"] = spans_df[\"start\"].dt.year\n",
+ "print(spans_df)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 7,
+ "id": "607b5f3a-51cb-4d4e-8bf2-43a6ad2cb6e3",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "rows kept for modeling: 3649\n",
+ "rows with observed downstream depth: 3591\n",
+ "rows with missing downstream depth but still usable: 58\n"
+ ]
+ }
+ ],
+ "source": [
+ "# Keep rows with usable inputs\n",
+ "input_cols = [\n",
+ " \"Upstream Depth (ft)\",\n",
+ " \"Upstream Mean Discharge (cfs)\",\n",
+ " \"Upstream Width (ft)\",\n",
+ " \"Downstream Width (ft)\",\n",
+ "]\n",
+ "\n",
+ "# We also require downstream discharge for the current version\n",
+ "required_now = input_cols + [\"Downstream Mean Discharge (cfs)\"]\n",
+ "\n",
+ "df_model = df.dropna(subset=required_now).copy()\n",
+ "df_model[\"depth_mask\"] = (~df_model[\"Downstream Depth (ft)\"].isna()).astype(int)\n",
+ "\n",
+ "print(\"rows kept for modeling:\", len(df_model))\n",
+ "print(\"rows with observed downstream depth:\", int(df_model[\"depth_mask\"].sum()))\n",
+ "print(\"rows with missing downstream depth but still usable:\", int((df_model[\"depth_mask\"] == 0).sum()))"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 39,
+ "id": "cf10f8cd-676b-41bc-8503-82cfcabf0dad",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# Train/test split by time\n",
+ "split_idx = int(len(df_model) * 0.8)\n",
+ "\n",
+ "train_df = df_model.iloc[:split_idx].copy()\n",
+ "test_df = df_model.iloc[split_idx:].copy()\n",
+ "\n",
+ "feature_cols = [\n",
+ " \"Upstream Depth (ft)\",\n",
+ " \"Upstream Mean Discharge (cfs)\",\n",
+ " \"Upstream Width (ft)\",\n",
+ " \"Downstream Width (ft)\",\n",
+ "]\n",
+ "\n",
+ "# Add a simple time feature\n",
+ "train_df[\"time_index\"] = np.arange(len(train_df))\n",
+ "test_df[\"time_index\"] = np.arange(len(train_df), len(train_df) + len(test_df))\n",
+ "\n",
+ "feature_cols_with_time = feature_cols + [\"time_index\"]\n",
+ "\n",
+ "# Scale inputs only\n",
+ "x_scaler = StandardScaler()\n",
+ "X_train = x_scaler.fit_transform(train_df[feature_cols_with_time])\n",
+ "X_test = x_scaler.transform(test_df[feature_cols_with_time])\n",
+ "\n",
+ "# Targets stay in physical units for easier interpretation\n",
+ "y_train_h = train_df[\"Downstream Depth (ft)\"].to_numpy().reshape(-1, 1)\n",
+ "y_train_q = train_df[\"Downstream Mean Discharge (cfs)\"].to_numpy().reshape(-1, 1)\n",
+ "\n",
+ "y_test_h = test_df[\"Downstream Depth (ft)\"].to_numpy().reshape(-1, 1)\n",
+ "y_test_q = test_df[\"Downstream Mean Discharge (cfs)\"].to_numpy().reshape(-1, 1)\n",
+ "\n",
+ "# Masks for missing downstream depth\n",
+ "mask_train_h = train_df[\"depth_mask\"].to_numpy().reshape(-1, 1)\n",
+ "mask_test_h = test_df[\"depth_mask\"].to_numpy().reshape(-1, 1)\n",
+ "\n",
+ "# Raw physical values needed in the PDE loss\n",
+ "q_u_train = train_df[\"Upstream Mean Discharge (cfs)\"].to_numpy().reshape(-1, 1)\n",
+ "h_u_train = train_df[\"Upstream Depth (ft)\"].to_numpy().reshape(-1, 1)\n",
+ "b_d_train = train_df[\"Downstream Width (ft)\"].to_numpy().reshape(-1, 1)\n",
+ "\n",
+ "q_u_test = test_df[\"Upstream Mean Discharge (cfs)\"].to_numpy().reshape(-1, 1)\n",
+ "h_u_test = test_df[\"Upstream Depth (ft)\"].to_numpy().reshape(-1, 1)\n",
+ "b_d_test = test_df[\"Downstream Width (ft)\"].to_numpy().reshape(-1, 1)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 40,
+ "id": "705d017e-f1ea-4206-a76c-7f8a02d7d08d",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# Convert to torch tensors\n",
+ "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
+ "\n",
+ "X_train_t = torch.tensor(X_train, dtype=torch.float32, device=device)\n",
+ "X_test_t = torch.tensor(X_test, dtype=torch.float32, device=device)\n",
+ "\n",
+ "y_train_h_t = torch.tensor(np.nan_to_num(y_train_h, nan=0.0), dtype=torch.float32, device=device)\n",
+ "y_train_q_t = torch.tensor(y_train_q, dtype=torch.float32, device=device)\n",
+ "\n",
+ "y_test_h_t = torch.tensor(np.nan_to_num(y_test_h, nan=0.0), dtype=torch.float32, device=device)\n",
+ "y_test_q_t = torch.tensor(y_test_q, dtype=torch.float32, device=device)\n",
+ "\n",
+ "mask_train_h_t = torch.tensor(mask_train_h, dtype=torch.float32, device=device)\n",
+ "mask_test_h_t = torch.tensor(mask_test_h, dtype=torch.float32, device=device)\n",
+ "\n",
+ "q_u_train_t = torch.tensor(q_u_train, dtype=torch.float32, device=device)\n",
+ "h_u_train_t = torch.tensor(h_u_train, dtype=torch.float32, device=device)\n",
+ "b_d_train_t = torch.tensor(b_d_train, dtype=torch.float32, device=device)\n",
+ "\n",
+ "q_u_test_t = torch.tensor(q_u_test, dtype=torch.float32, device=device)\n",
+ "h_u_test_t = torch.tensor(h_u_test, dtype=torch.float32, device=device)\n",
+ "b_d_test_t = torch.tensor(b_d_test, dtype=torch.float32, device=device)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 41,
+ "id": "0e0edcf0-411e-46f4-9c27-ae93e187464d",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# Test setup\n",
+ "def finite_difference(y, dt):\n",
+ " dy = torch.zeros_like(y)\n",
+ " dy[1:-1] = (y[2:] - y[:-2]) / (2.0 * dt)\n",
+ " dy[0] = (y[1] - y[0]) / dt\n",
+ " dy[-1] = (y[-1] - y[-2]) / dt\n",
+ " return dy\n",
+ "\n",
+ "def continuity_residual(h_d, q_d, b_d, q_u):\n",
+ " A_d = b_d * h_d\n",
+ " dA_dt = finite_difference(A_d, DT_SEC)\n",
+ " dQ_dx = (q_d - q_u) / DX_FT\n",
+ " return dA_dt + dQ_dx\n",
+ "\n",
+ "def momentum_residual(h_u, h_d, q_d, b_d, n_hat):\n",
+ " A = b_d * h_d\n",
+ " P = b_d + 2.0 * h_d\n",
+ " R = A / (P + 1e-6)\n",
+ "\n",
+ " dQ_dt = finite_difference(q_d, DT_SEC)\n",
+ " dh_dx = (h_d - h_u) / DX_FT\n",
+ "\n",
+ " # Simplified Manning friction slope\n",
+ " Sf = (n_hat**2 * q_d**2) / ((A**2) * (R**(4.0 / 3.0)) + 1e-6)\n",
+ "\n",
+ " return dQ_dt + G * A * (dh_dx + Sf - S0)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 42,
+ "id": "1b3b5f5e-e478-4510-979c-a19b1e167a2c",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# Training setup\n",
+ "optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)\n",
+ "mse = nn.MSELoss()\n",
+ "\n",
+ "lambda_cont = 1e-4\n",
+ "lambda_mom = 1e-6\n",
+ "epochs = 1500\n",
+ "\n",
+ "history = []"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 43,
+ "id": "20cf81fa-c195-4e56-8a2c-9a4820666ff2",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "
\n",
+ " \n",
+ " \n",
+ " | \n",
+ " Date | \n",
+ " Observed Downstream Depth (ft) | \n",
+ " Predicted Downstream Depth (ft) | \n",
+ " Observed Downstream Discharge (cfs) | \n",
+ " Predicted Downstream Discharge (cfs) | \n",
+ " Predicted Manning_n | \n",
+ "
\n",
+ " \n",
+ " \n",
+ " \n",
+ " | 2919 | \n",
+ " 2023-12-29 | \n",
+ " 6.997467 | \n",
+ " 0.676136 | \n",
+ " 161.0 | \n",
+ " 0.720054 | \n",
+ " 0.056595 | \n",
+ "
\n",
+ " \n",
+ " | 2920 | \n",
+ " 2023-12-30 | \n",
+ " 7.003519 | \n",
+ " 0.676441 | \n",
+ " 162.0 | \n",
+ " 0.720123 | \n",
+ " 0.056578 | \n",
+ "
\n",
+ " \n",
+ " | 2921 | \n",
+ " 2023-12-31 | \n",
+ " 7.003519 | \n",
+ " 0.676462 | \n",
+ " 162.0 | \n",
+ " 0.720091 | \n",
+ " 0.056577 | \n",
+ "
\n",
+ " \n",
+ " | 2922 | \n",
+ " 2024-01-01 | \n",
+ " 7.003519 | \n",
+ " 0.676483 | \n",
+ " 162.0 | \n",
+ " 0.720058 | \n",
+ " 0.056576 | \n",
+ "
\n",
+ " \n",
+ " | 2923 | \n",
+ " 2024-01-02 | \n",
+ " 7.037459 | \n",
+ " 0.677449 | \n",
+ " 166.0 | \n",
+ " 0.720354 | \n",
+ " 0.056576 | \n",
+ "
\n",
+ " \n",
+ "
\n",
+ "