{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "d24d3c2e",
   "metadata": {},
   "source": [
    "## Import Necessary Libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "5d7cbeb2",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import mean_squared_error\n",
    "from sklearn.neural_network import MLPRegressor\n",
    "from sklearn.model_selection import RandomizedSearchCV\n",
    "from sklearn.datasets import make_regression\n",
    "from sklearn.model_selection import train_test_split\n",
    "from scipy.stats import randint as sp_randint\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7e2197d2",
   "metadata": {},
   "source": [
    "## Pull in Texture and Mesri Estimated Friction Angles from Excel Sheet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "2c20df52",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\dmfr224.AD\\Anaconda3\\lib\\site-packages\\openpyxl\\worksheet\\header_footer.py:48: UserWarning: Cannot parse header or footer so it will be ignored\n",
      "  warn(\"\"\"Cannot parse header or footer so it will be ignored\"\"\")\n"
     ]
    }
   ],
   "source": [
    "Regress = pd.read_excel(r'F:\\Research\\Journal Paper #2\\Machine Learning Friction Angles\\Site Texture and Mesri Friction Angles at KYTC Boreholes.xlsx')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b56f7abb",
   "metadata": {},
   "source": [
    "## Assign Training and Response Variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "a6f2b699",
   "metadata": {},
   "outputs": [],
   "source": [
    "X = Regress[['Percent Sand', 'Percent Silt', 'Percent Clay']] #Removed PI and LL\n",
    "Response = Regress[['Friction Angle']] #Terzaghi Friction Angle from Plasticity Index\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, Response, test_size=0.2, random_state=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4c2ae198",
   "metadata": {},
   "source": [
    "## Use RandomizedSearchCV to Find Best Tuning for Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2ba394d0",
   "metadata": {},
   "outputs": [],
   "source": [
    "# specify parameters and distributions to sample from\n",
    "param_dist = {\"random_state\": (0,1),\n",
    "              \"activation\": ['identity'],            \n",
    "              \"solver\": ['adam'],\n",
    "              \"max_iter\": sp_randint(200, 100000),\n",
    "              \"hidden_layer_sizes\": sp_randint(1100, 1150),\n",
    "              \"learning_rate\":['adaptive']}\n",
    "random_search = RandomizedSearchCV(estimator = regr, param_distributions=param_dist,\n",
    "                                   n_iter=10000, cv=5, verbose = 3, random_state=42, n_jobs = -1)\n",
    "random_search.fit(X_train, y_train.values.ravel());\n",
    "\n",
    "print(random_search.best_score_)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "96c15837",
   "metadata": {},
   "source": [
    "## Show Best Parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "aa0433db",
   "metadata": {},
   "outputs": [],
   "source": [
    "print(random_search.best_params_)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a7024283",
   "metadata": {},
   "source": [
    "## Construct MLP Regressive Model Based on RandomizedSearchCV Parameters\n",
    "May differ due to randomization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "7db0386d",
   "metadata": {},
   "outputs": [],
   "source": [
    "regr = MLPRegressor(random_state=1, activation='identity', hidden_layer_sizes=1104, learning_rate='adaptive', max_iter=95635, solver='adam', ).fit(X_train, y_train.values.ravel())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "491ad994",
   "metadata": {},
   "source": [
    "## Predict Friction Angle using Trained Model and Test Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "c957712c",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([29.08695336, 28.63646272, 33.34170334, 31.56994005, 32.30337337,\n",
       "       28.39146397, 28.91747906, 32.30337337, 31.09844273, 28.3024598 ,\n",
       "       28.66782231, 31.77309641, 28.83174369, 33.96006796, 31.93156479,\n",
       "       34.44198829, 27.52909642, 31.80464005, 29.31804079, 32.47641638])"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Predicted = regr.predict(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cd9b55b1",
   "metadata": {},
   "source": [
    "## Retrieve RMSE of Predicted vs True Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "e4992980",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8604795387351668\n"
     ]
    }
   ],
   "source": [
    "rms = mean_squared_error(y_test, Predicted, squared=False)\n",
    "print(rms)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8e1200a5",
   "metadata": {},
   "source": [
    "The trained model can be seen to perform well (RMSE = 0.86) and is considered valid for use in predicting friction angles based on soil texture data over unseen data"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}