{
 "cells": [
  {
   "cell_type": "markdown",
   "source": [
    "### Solution for Assignment 5 of the course \"Introduction to Machine Learning\" at the University of Leoben.\n",
    "##### Author: Fotios Lygerakis\n",
    "##### Semester: SS 2022/2023"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# Perceptron Algorithm for Classification of Iris Dataset"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(150, 4)\n",
      "(150,)\n"
     ]
    }
   ],
   "source": [
    "# load the iris dataset\n",
    "from sklearn.datasets import load_iris\n",
    "from sklearn.metrics import accuracy_score\n",
    "import numpy as np\n",
    "\n",
    "iris = load_iris()\n",
    "X = iris.data\n",
    "y = iris.target\n",
    "print(X.shape)\n",
    "print(y.shape)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Preprocess the data"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [],
   "source": [
    "# Preprocess the data\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "# split the scaled data into train and test sets\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Define the perceptron algorithm"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [],
   "source": [
    "# Define the perceptron algorithm\n",
    "class MultiClassPerceptron:\n",
    "    def __init__(self, input_dim, output_dim, lr=0.01, epochs=1000):\n",
    "        self.W = np.random.randn(input_dim, output_dim)\n",
    "        self.b = np.zeros((1, output_dim))\n",
    "        self.lr = lr\n",
    "        self.epochs = epochs\n",
    "\n",
    "    def forward(self, X):\n",
    "        self.z = np.dot(X, self.W) + self.b\n",
    "        self.y_hat = np.exp(self.z) / np.sum(np.exp(self.z), axis=1, keepdims=True)\n",
    "\n",
    "    def backward(self, X, y):\n",
    "        m = X.shape[0]  # number of samples\n",
    "        # Calculate the gradients\n",
    "        grad_z = self.y_hat # shape (m, C)\n",
    "        # Subtract 1 from the predicted class for each sample\n",
    "        grad_z[range(m), y] -= 1 # shape (m, C)\n",
    "        # Calculate the gradients with respect to the parameters\n",
    "        grad_W = np.dot(X.T, grad_z) # shape (n, C)\n",
    "        # Reshape the gradients into a 2-D array\n",
    "        grad_b = np.sum(grad_z, axis=0, keepdims=True) # shape (1, C)\n",
    "        # Update the parameters\n",
    "        self.W -= self.lr * grad_W # shape (n, C)\n",
    "        self.b -= self.lr * grad_b # shape (1, C)\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        for epoch in range(self.epochs):\n",
    "            self.forward(X)\n",
    "            self.backward(X, y)\n",
    "\n",
    "    def predict(self, X):\n",
    "        self.forward(X)\n",
    "        return np.argmax(self.y_hat, axis=1)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Train the model"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [],
   "source": [
    "# Train the model\n",
    "p = MultiClassPerceptron(input_dim=X_train.shape[1], output_dim=3, lr=0.01, epochs=1000)\n",
    "p.fit(X_train, y_train)\n",
    "predictions_train = p.predict(X_train)\n",
    "predictions = p.predict(X_test)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Evaluate the model"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Perceptron classification train accuracy 0.975\n",
      "Perceptron classification accuracy 1.0\n"
     ]
    }
   ],
   "source": [
    "# evaluate train accuracy\n",
    "print(\"Perceptron classification train accuracy\", accuracy_score(y_train, predictions_train))\n",
    "print(\"Perceptron classification accuracy\", accuracy_score(y_test, predictions))"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Non-linear feature transformation on the concrete compressive strength dataset"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "outputs": [],
   "source": [
    "def polynomial_features(X, degree):\n",
    "    \"\"\"\n",
    "    Creates a new feature matrix consisting of all polynomial combinations of the features with degree less than or equal to the specified degree.\n",
    "    For example, if an input sample is two dimensional and of the form [a, b], the degree-2 polynomial features are [1, a, b, a^2, ab, b^2].\n",
    "    Parameters\n",
    "    ----------\n",
    "    X : array-like, shape (n_samples, n_features)\n",
    "        The input samples.\n",
    "    degree : int\n",
    "        The degree of the polynomial features.\n",
    "    Returns\n",
    "    -------\n",
    "    X_new : array-like, shape (n_samples, 1 + n_features + n_features*(n_features+1)/2)\n",
    "        The polynomial features with degree `degree`.\n",
    "    \"\"\"\n",
    "    n_samples, n_features = np.shape(X)\n",
    "    new_features = np.ones(shape=(n_samples, 1))\n",
    "\n",
    "    for i in range(n_features):\n",
    "        for j in range(1, degree+1):\n",
    "            # create a new column for each feature, with values raised to the power of j\n",
    "            new_col = np.power(X[:, i], j) # shape (n_samples, 1)\n",
    "            # reshape the new column to a 2-D array\n",
    "            new_col = new_col.reshape(n_samples, 1) # shape (n_samples, 1)\n",
    "            # append the new column to the new_features array\n",
    "            new_features = np.hstack((new_features, new_col)) # shape (n_samples, j+1)\n",
    "\n",
    "    return new_features"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "outputs": [],
   "source": [
    "# Non-linear feature transformation\n",
    "import pandas as pd\n",
    "from sklearn.preprocessing import PolynomialFeatures\n",
    "from sklearn.linear_model import LinearRegression\n",
    "from sklearn.metrics import mean_squared_error, r2_score\n",
    "\n",
    "# load the concrete compressive strength dataset\n",
    "df = pd.read_excel('Concrete_Data.xls')\n",
    "\n",
    "# split the data into train and test sets\n",
    "X = df.drop(['Concrete compressive strength(MPa, megapascals) '], axis=1)\n",
    "y = df['Concrete compressive strength(MPa, megapascals) ']\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)\n",
    "\n",
    "# transform the features into second degree polynomial features\n",
    "poly = PolynomialFeatures(degree=2)\n",
    "X_train_poly = poly.fit_transform(X_train)\n",
    "X_test_poly = poly.transform(X_test)\n",
    "\n",
    "X_train_poly_custom = polynomial_features(X_train.values, degree=2)\n",
    "X_test_poly_custom = polynomial_features(X_test.values, degree=2)\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Train the linear regression model"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean squared error (train poly custom): 64.55\n",
      "Mean squared error (test poly custom): 58.28\n",
      "Mean squared error (train): 110.66\n",
      "Mean squared error (test): 95.98\n",
      "R^2 (train poly custom): 0.77\n",
      "R^2 (test poly custom): 0.77\n",
      "R^2 (train): 0.61\n",
      "R^2 (test): 0.63\n"
     ]
    }
   ],
   "source": [
    "# Train the model\n",
    "lr_poly_custom = LinearRegression()\n",
    "lr = LinearRegression()\n",
    "# fit the model\n",
    "lr_poly_custom.fit(X_train_poly_custom, y_train)\n",
    "lr.fit(X_train, y_train)\n",
    "# predict values from the polynomial transformed features\n",
    "predictions_poly_custom_train = lr_poly_custom.predict(X_train_poly_custom)\n",
    "predictions_poly_custom = lr_poly_custom.predict(X_test_poly_custom)\n",
    "# predict values from the original features\n",
    "predictions_train = lr.predict(X_train)\n",
    "predictions = lr.predict(X_test)\n",
    "\n",
    "# mean squared error\n",
    "print(\"Mean squared error (train poly custom): {:.2f}\".format(mean_squared_error(y_train, predictions_poly_custom_train)))\n",
    "print(\"Mean squared error (test poly custom): {:.2f}\".format(mean_squared_error(y_test, predictions_poly_custom)))\n",
    "print(\"Mean squared error (train): {:.2f}\".format(mean_squared_error(y_train, predictions_train)))\n",
    "print(\"Mean squared error (test): {:.2f}\".format(mean_squared_error(y_test, predictions)))\n",
    "\n",
    "# coefficient of determination (R^2)\n",
    "print(\"R^2 (train poly custom): {:.2f}\".format(r2_score(y_train, predictions_poly_custom_train)))\n",
    "print(\"R^2 (test poly custom): {:.2f}\".format(r2_score(y_test, predictions_poly_custom)))\n",
    "print(\"R^2 (train): {:.2f}\".format(r2_score(y_train, predictions_train)))\n",
    "print(\"R^2 (test): {:.2f}\".format(r2_score(y_test, predictions)))\n",
    "\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "RBFs on the California Housing Prices dataset"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "outputs": [],
   "source": [
    "def rbf_kernel(X, centers, gamma):\n",
    "    # Pairwise Euclidean distances calculation:\n",
    "    # Compute the squared Euclidean distances between each sample and each center using broadcasting:\n",
    "    # - Subtract each center from each sample to get a difference matrix of shape (n_samples, n_centers, n_features)\n",
    "    # - Square each element in the difference matrix\n",
    "    # - Sum the squared differences along the feature axis to get the squared distances matrix of shape (n_samples, n_centers)\n",
    "    # - Take the square root of each element in the squared distances matrix to obtain the pairwise Euclidean distances matrix of shape (n_samples, n_centers)\n",
    "    dists = np.sqrt(((X[:, np.newaxis] - centers)**2).sum(axis=2)) # shape (n_samples, n_centers)\n",
    "    # Compute the RBF values for each distance using the Gaussian kernel\n",
    "    rbf_vals = np.exp(-gamma * dists**2) # shape (n_samples, n_centers)\n",
    "    return rbf_vals"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Linear regression on original data:\n",
      "MSE: 0.5558915986952443\n",
      "R^2: 0.5757877060324508\n",
      "\n",
      "Linear regression on RBF-transformed data:\n",
      "MSE: 0.37106446913117447\n",
      "R^2: 0.7168330839511696\n"
     ]
    }
   ],
   "source": [
    "from sklearn.datasets import fetch_california_housing\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.linear_model import LinearRegression\n",
    "from sklearn.metrics import mean_squared_error, r2_score\n",
    "\n",
    "# Load the California Housing Prices dataset\n",
    "data = fetch_california_housing()\n",
    "X = data['data']\n",
    "y = data['target']\n",
    "\n",
    "# Split the data into training and testing sets\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)\n",
    "\n",
    "# Standardize the features\n",
    "scaler = StandardScaler()\n",
    "X_train_std = scaler.fit_transform(X_train)\n",
    "X_test_std = scaler.transform(X_test)\n",
    "\n",
    "# Choose the number of centroids and the RBF kernel width\n",
    "num_centroids = 100\n",
    "gamma = 0.1\n",
    "\n",
    "# Randomly select the centroids from the training set\n",
    "np.random.seed(42)\n",
    "idx = np.random.choice(X_train_std.shape[0], num_centroids, replace=False)\n",
    "centroids = X_train_std[idx] # (100, 8)\n",
    "\n",
    "# Compute the RBF features for the training and testing sets\n",
    "rbf_train = rbf_kernel(X_train_std, centroids, gamma)   # (16512, 100)\n",
    "rbf_test = rbf_kernel(X_test_std, centroids, gamma)    # (4128, 100)\n",
    "\n",
    "# Fit a linear regression model on the original and RBF-transformed data\n",
    "linreg_orig = LinearRegression().fit(X_train_std, y_train)\n",
    "linreg_rbf = LinearRegression().fit(rbf_train, y_train)\n",
    "\n",
    "# Evaluate the models on the testing set\n",
    "y_pred_orig = linreg_orig.predict(X_test_std)\n",
    "mse_orig = mean_squared_error(y_test, y_pred_orig)\n",
    "r2_orig = r2_score(y_test, y_pred_orig)\n",
    "\n",
    "y_pred_rbf = linreg_rbf.predict(rbf_test)\n",
    "mse_rbf = mean_squared_error(y_test, y_pred_rbf)\n",
    "r2_rbf = r2_score(y_test, y_pred_rbf)\n",
    "\n",
    "# Print the results\n",
    "print(\"Linear regression on original data:\")\n",
    "print(\"MSE:\", mse_orig)\n",
    "print(\"R^2:\", r2_orig)\n",
    "\n",
    "print(\"\\nLinear regression on RBF-transformed data:\")\n",
    "print(\"MSE:\", mse_rbf)\n",
    "print(\"R^2:\", r2_rbf)\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# **(Bonus)** Multilayer Perceptron Algorithm for Regression of Concrete Compressive Strength Dataset"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Download the Concrete Compressive Strength Dataset from the UCI Machine Learning Repository."
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1030, 9)\n"
     ]
    }
   ],
   "source": [
    "# Download the Concrete Compressive Strength Dataset from the UCI Machine Learning Repository.\n",
    "import pandas as pd\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "import numpy as np\n",
    "\n",
    "df = pd.read_excel('Concrete_Data.xls')\n",
    "print(df.shape)\n",
    "# df.head()"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Preprocess the data"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "outputs": [],
   "source": [
    "# Preprocess the data\n",
    "X = df.iloc[:, :-1].values\n",
    "y = df.iloc[:, -1].values.reshape(-1, 1)\n",
    "\n",
    "# Normalize the features\n",
    "X_norm = StandardScaler().fit_transform(X)\n",
    "\n",
    "# Split the data into training and testing sets\n",
    "X_train, X_test, y_train, y_test = train_test_split(X_norm, y, test_size=0.2, random_state=42)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Define the multilayer perceptron algorithm"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "outputs": [],
   "source": [
    "# a multilayer perceptron algorithm class for regression problems\n",
    "class MLP:\n",
    "    def __init__(self, input_dim, hidden_dim, output_dim, lr=0.01, epochs=1000):\n",
    "        self.W1 = np.random.randn(input_dim, hidden_dim)\n",
    "        self.b1 = np.zeros((1, hidden_dim))\n",
    "        self.W2 = np.random.randn(hidden_dim, output_dim)\n",
    "        self.b2 = np.zeros((1, output_dim))\n",
    "        self.lr = lr\n",
    "        self.epochs = epochs\n",
    "\n",
    "    def forward(self, X):\n",
    "        # forward propagation through our network\n",
    "        self.z1 = np.dot(X, self.W1) + self.b1\n",
    "        # activation function\n",
    "        self.a1 = np.tanh(self.z1)\n",
    "        # output layer\n",
    "        self.z2 = np.dot(self.a1, self.W2) + self.b2\n",
    "        # final activation function\n",
    "        self.y_hat = self.z2\n",
    "\n",
    "    def backward(self, X, y):\n",
    "        # number of samples\n",
    "        m = X.shape[0]\n",
    "        # output layer gradient\n",
    "        self.loss = np.mean((self.y_hat - y) ** 2) # MSE loss. shape (n_samples, output_dim)\n",
    "        # output layer gradient\n",
    "        delta2 = (self.y_hat - y) # shape (n_samples, output_dim)\n",
    "        # hidden layer gradient\n",
    "        dW2 = np.dot(self.a1.T, delta2) # shape (hidden_dim, output_dim)\n",
    "        # bias gradient\n",
    "        db2 = np.sum(delta2, axis=0, keepdims=True) # shape (1, output_dim)\n",
    "        # hidden layer gradient\n",
    "        delta1 = np.dot(delta2, self.W2.T) * (1 - np.power(self.a1, 2)) # shape (n_samples, hidden_dim)\n",
    "        # input layer gradient\n",
    "        dW1 = np.dot(X.T, delta1) # shape (input_dim, hidden_dim)\n",
    "        # bias gradient\n",
    "        db1 = np.sum(delta1, axis=0) # shape (1, hidden_dim)\n",
    "        # update parameters\n",
    "        self.W2 -= self.lr * dW2 / m\n",
    "        self.b2 -= self.lr * db2 / m\n",
    "        self.W1 -= self.lr * dW1 / m\n",
    "        self.b1 -= self.lr * db1 / m\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        for epoch in range(self.epochs):\n",
    "            self.forward(X)\n",
    "            self.backward(X, y)\n",
    "\n",
    "    def predict(self, X):\n",
    "        self.forward(X)\n",
    "        return self.y_hat\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Train the model"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "outputs": [],
   "source": [
    "# Create an instance of the MLP class\n",
    "mlp = MLP(input_dim=X_train.shape[1], hidden_dim=10, output_dim=1, lr=0.01, epochs=1000)\n",
    "# Train the model\n",
    "mlp.fit(X_train, y_train)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Evaluate the model"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean Squared Error: 36.8911071801165\n"
     ]
    }
   ],
   "source": [
    "# Evaluate the model\n",
    "from sklearn.metrics import mean_squared_error\n",
    "\n",
    "y_pred = mlp.predict(X_test)\n",
    "print(\"Mean Squared Error:\", mean_squared_error(y_test, y_pred))"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Compare the results with the linear regression model"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean Squared Error: 95.97548435337708\n"
     ]
    }
   ],
   "source": [
    "# Compare the results with the linear regression model\n",
    "from sklearn.linear_model import LinearRegression\n",
    "\n",
    "lr = LinearRegression()\n",
    "lr.fit(X_train, y_train)\n",
    "y_pred = lr.predict(X_test)\n",
    "print(\"Mean Squared Error:\", mean_squared_error(y_test, y_pred))"
   ],
   "metadata": {
    "collapsed": false
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}