20161216_Predicting Abalone Rings Part 3.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "The third blog post in this series will model the Abalone dataset using a multilayer perceptron neural network. As before, the following code will load and preprocess the Abalone data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import seaborn as sns\n",
    "from scipy import stats\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.neural_network import MLPRegressor\n",
    "from sklearn.preprocessing import StandardScaler \n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import r2_score, mean_absolute_error\n",
    "%matplotlib inline\n",
    "abaloneDF = pd.read_csv('abalone.csv', names=['Sex', 'Length', 'Diameter', 'Height',\n",
    "                                              'Whole Weight', 'Shucked Weight',\n",
    "                                              'Viscera Weight', 'Shell Weight',\n",
    "                                              'Rings'])\n",
    "abaloneDF['Male'] = (abaloneDF['Sex'] == 'M').astype(int)\n",
    "abaloneDF['Female'] = (abaloneDF['Sex'] == 'F').astype(int)\n",
    "abaloneDF['Infant'] = (abaloneDF['Sex'] == 'I').astype(int)\n",
    "abaloneDF = abaloneDF[abaloneDF['Height'] > 0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Perceptron Model Construction\n",
    "\n",
    "Multilayer perceptrons can be used for a variety of purposes, including classification or function approximation. A perceptron accepts a number of inputs (variables), which usually include all the available information about the problem to be solved. The inputs are given weights as a linear combination within a hidden layer or layers that contains multiple nodes. The hidden layer(s) are passed through an activation function (e.g. a hard limit or sigmoid function). Finally, the output(s) are calculated from the activation function. \n",
    "\n",
    "This may be represented by the following diagram:\n",
    "\n",
    "Perceptrons are trained by successively updating the weights until the outputs of the perceptron closely match the \"ground truth\" (known outputs). Weights are updated by the error of the outputs relative to the ground truth values, based on the learning rate, called η. It is important that η be chosen carefully when training a perceptron. If η is too large, the weight updating might simply \"bounce\" back and forth from two different points on an error surface, never reaching the minimum required accuracy. However, if η is too small, the weights will update very slowly, and the perceptron will take a long time to reach the required accuracy, although this is usually a more certain method than selecting a large η. \n",
    "\n",
    "Multilayer perceptrons (as a regression model) are implemented in scikit-learn as the class \"MLPRegressor\". Initially, I'll select some default model hyperparameters, and in the following section we will investigate how hyperparameter selection changes the final results. Note that the data was rescaled to give it zero mean and unit variance, typical preprocessing for a perceptron to ensure that each variable is given equal importance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAhMAAAGJCAYAAAAwtrGcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIySYEARBEAQhIySYEARBEAQhI/4/Uj4oQkC+LLYAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x281241debe0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Y-Yhat plot, as before\n",
    "def plot_yyhat(ytest,ypred):\n",
    "    r2 = r2_score(ytest, ypred )\n",
    "    mae = mean_absolute_error(ytest, ypred)\n",
    "    absmin = min([ytest.min(),ypred.min()])\n",
    "    absmax = max([ytest.max(),ypred.max()])\n",
    "    ax = plt.axes()\n",
    "    ax.scatter(ytest,ypred)\n",
    "    ax.set_title('Y vs. YHat')\n",
    "    ax.axis([absmin, absmax, absmin, absmax])\n",
    "    ax.plot([absmin, absmax], [absmin, absmax],c=\"k\")\n",
    "    ax.set_ylabel('Predicted Rings') \n",
    "    ax.set_xlabel('Actual Rings')\n",
    "    props = dict(boxstyle='round', facecolor='wheat', alpha=0.5)\n",
    "    textStr = '$MAE=%.3f$\\n$R2=%.3f$' % (mae, r2)\n",
    "    ax.text(0.05, 0.95, textStr, transform=ax.transAxes, fontsize=14,\n",
    "               verticalalignment='top', bbox=props);\n",
    "#Cross-validation, as before\n",
    "train, test = train_test_split(abaloneDF,train_size=0.7)\n",
    "xtrain = train.drop(['Rings','Sex'],axis=1)\n",
    "ytrain = train['Rings']\n",
    "xtest = test.drop(['Rings','Sex'],axis=1)\n",
    "ytest = test['Rings']\n",
    "#Scale the data\n",
    "stScaler = StandardScaler()  \n",
    "stScaler.fit(xtrain)  \n",
    "xtrain2 = stScaler.transform(xtrain)  \n",
    "xtest2 = stScaler.transform(xtest)  \n",
    "#Construct a multi-layer perceptron \n",
    "mlp = MLPRegressor(hidden_layer_sizes = [10,10],\n",
    "                   alpha=0.00001,\n",
    "                   activation = 'logistic',\n",
    "                   learning_rate_init = 0.001,\n",
    "                   max_iter = 1000)\n",
    "mlp.fit(xtrain2,ytrain)\n",
    "ypred = mlp.predict(xtest2)\n",
    "yresid = ytest - ypred\n",
    "plot_yyhat(ytest,ypred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Without any hyper-parameter optimization, and really without any thought at all, we already achieved a MAE better than the multiple regression model in Part 1! However, note that the MAE was unstable based on selection of training and testing data, and since the training and testing dataset is different between Part 1 and Part 3, the interpretability of this result is questionable. Your results may vary, if you are re-running this notebook.\n",
    "\n",
    "### Hyperparameter Selection Overview\n",
    "\n",
    "The hyperparameters that we could investigate for our multilayer perceptron might include:\n",
    "\n",
    "* Number of input nodes\n",
    "* Number of output nodes\n",
    "* Hidden layer architecture (number of layers and nodes)\n",
    "* Regularization parameter (alpha)\n",
    "* Activation function\n",
    "* Learning rate\n",
    "\n",
    "The number of inputs nodes are determined by the information we have available (variables). We might consider reducing the number of input variables, as we did in the previous two posts, but to simplify this blog post, we will just use all 10 variables as the inputs. The number of output nodes is determined by the scope of the problem; for this situation, we are only interested in the number of rings.\n",
    "\n",
    "Much of the \"art\" of building neural networks has to do with the selection of the hidden layer architecture. Any number of hidden layers may be used, and any number of hidden nodes may be within the layer. In general, though, the minimum number of layers and nodes required to achieve the desired accuracy is preferred. Selecting too many layers or nodes will usually cause overfitting.\n",
    "\n",
    "The regularization parameter (alpha) is L2 regularization, the same hyperparameter that was used for the Lasso model in Part 1. For the Lasso model, we chose alpha using cross-validation; the same method is applicable here, as well.\n",
    "\n",
    "The activation function is roughly analogous to the binary operation of actual biological neurons; either the neuron fires, or it doesn't. However, for artificial neuron networks, it is often useful to provide a little bit of \"leeway\" within the algorithm near the on/off switching point. Activation functions are continuously differentiable, and a common choice is a logistic function.\n",
    "\n",
    "The learning rate η is a measure of the step size that the neural network will use to find a solution to the problem at hand. After each training iteration, the weights and bias will be updated according to a general rule:\n",
    "\n",
    "w(n+1)=w(n) +/- η(n)p(n)\n",
    "\n",
    "Where w refers to the weights, η is the learning rate, and p is the input parameters. \n",
    "\n",
    "Usually, low-valued η will take longer to train but will be more likely to recognize a minimum in the error space. Higher-valued η might improve the training speed but will be more likely to \"jump\" back and forth across a local minimum. Furthermore, a lower η might merely find a local minimum, when another nearby local minimum might achieve much better accuracy. Choosing η is heavily dependent on the particular dataset being examined, so it is important to investigate the effects of different combinations.\n",
    "\n",
    "In the following sections, we will investigate the behavior of the perceptron as the hyperparameters are modified. Note that this process would typically be done simultaneously (i.e. we would change the alpha, activation function, learning rate, and hidden layer architecture at the same time), but for clarity, we will split it into 4 steps.\n",
    "\n",
    "### Activation Function Selection\n",
    "\n",
    "The initially chosen activation function was a logistic function. Let's examine how the model behaves when other types of activation functions are selected:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'identity': 1.5676114287084544,\n",
       " 'logistic': 1.4468648164410229,\n",
       " 'relu': 1.5212894186491739,\n",
       " 'tanh': 1.4430986184978607}"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Obtain MAE from modifying the activation function\n",
    "def activationResults(func):\n",
    "    mlp = MLPRegressor(hidden_layer_sizes = [10,10],\n",
    "                       alpha=0.00001,\n",
    "                       activation = func,\n",
    "                       learning_rate_init = 0.001,\n",
    "                       max_iter = 2000)\n",
    "    mlp.fit(xtrain2,ytrain)\n",
    "    ypred = mlp.predict(xtest2)\n",
    "    return mean_absolute_error(ytest, ypred)\n",
    "#Run the possibilities in a loop\n",
    "dictResults = {}\n",
    "for func in ['identity','logistic','tanh','relu']:\n",
    "    dictResults[func] = activationResults(func)\n",
    "#Print the results\n",
    "dictResults"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Based on the above results, the logistic and tanh functions achieved the lowest MAE, with a fairly close difference. Let's select the logistic function. Note that model architecture may have an effect on the ideal activation function for a given dataset, so this step should ideally be performed in conjunction with other hyperparameter selection.\n",
    "\n",
    "### Regularization Parameter Selection\n",
    "\n",
    "Next, we will select the ideal regularization parameter (alpha). 35 log-spaced values between 1e-6 and 1 will be tried (I did some quick manual investigation to restrict the hyperparameter space to alpha < 1).\n",
    "\n",
    "Note- this code will run much more slowly, as the number of models is 35 instead of 4, as in the previous section. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAioAAAGBCAYAAABIA5oDAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nBEuFioiIiATr/wGf0Od9v1XmDgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x281255b9630>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Obtain MAE from modifying alpha\n",
    "def alphaResults(alpha):\n",
    "    mlp = MLPRegressor(hidden_layer_sizes = [10,10],\n",
    "                       alpha=alpha,\n",
    "                       activation = 'logistic',\n",
    "                       learning_rate_init = 0.001,\n",
    "                       max_iter = 2000)\n",
    "    mlp.fit(xtrain2,ytrain)\n",
    "    ypred = mlp.predict(xtest2)\n",
    "    return mean_absolute_error(ytest, ypred)\n",
    "#Run the possibilities in a loop\n",
    "results2 = []\n",
    "alphas = np.logspace(-6, 0, num=35)\n",
    "for alpha in alphas:\n",
    "    results2.append(alphaResults(alpha))\n",
    "#Plot the results\n",
    "plt.plot(alphas,results2)\n",
    "plt.xscale('log')\n",
    "plt.xlabel('Alpha')\n",
    "plt.ylabel('MAE');"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Results are fairly stable for alpha < 0.1. Remember that model architecture will also affect these results, so the MAE noise between 1e-6 and 1e-1 is probably insignificant. Let's choose alpha = 0.01.\n",
    "\n",
    "### Hidden Layer Architecture Selection\n",
    "\n",
    "Next, we will examine varying the hidden layer architecture. Initially, two hidden layers with 10 nodes each was chosen."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>[20, 5]</th>\n",
       "      <td>1.423391</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[20, 2, 20]</th>\n",
       "      <td>1.429209</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[10, 5]</th>\n",
       "      <td>1.429290</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[20, 10]</th>\n",
       "      <td>1.429421</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[20, 15]</th>\n",
       "      <td>1.437638</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[10, 2, 10]</th>\n",
       "      <td>1.437960</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[15, 5]</th>\n",
       "      <td>1.438687</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[15, 10]</th>\n",
       "      <td>1.438940</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[5, 15]</th>\n",
       "      <td>1.441288</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[10, 5, 10]</th>\n",
       "      <td>1.442504</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[15, 15]</th>\n",
       "      <td>1.442513</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[5, 5]</th>\n",
       "      <td>1.442841</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[10, 15]</th>\n",
       "      <td>1.444806</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[20, 5, 20]</th>\n",
       "      <td>1.446865</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[5, 10]</th>\n",
       "      <td>1.447412</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[20, 20]</th>\n",
       "      <td>1.447489</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[15, 20]</th>\n",
       "      <td>1.451034</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[10, 20]</th>\n",
       "      <td>1.452357</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[10, 1]</th>\n",
       "      <td>1.456077</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[10, 10]</th>\n",
       "      <td>1.462106</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[5, 20]</th>\n",
       "      <td>1.463455</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[20]</th>\n",
       "      <td>1.467650</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[10]</th>\n",
       "      <td>1.474535</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[2, 2]</th>\n",
       "      <td>1.480549</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>[1, 10]</th>\n",
       "      <td>1.580852</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                  MAE\n",
       "[20, 5]      1.423391\n",
       "[20, 2, 20]  1.429209\n",
       "[10, 5]      1.429290\n",
       "[20, 10]     1.429421\n",
       "[20, 15]     1.437638\n",
       "[10, 2, 10]  1.437960\n",
       "[15, 5]      1.438687\n",
       "[15, 10]     1.438940\n",
       "[5, 15]      1.441288\n",
       "[10, 5, 10]  1.442504\n",
       "[15, 15]     1.442513\n",
       "[5, 5]       1.442841\n",
       "[10, 15]     1.444806\n",
       "[20, 5, 20]  1.446865\n",
       "[5, 10]      1.447412\n",
       "[20, 20]     1.447489\n",
       "[15, 20]     1.451034\n",
       "[10, 20]     1.452357\n",
       "[10, 1]      1.456077\n",
       "[10, 10]     1.462106\n",
       "[5, 20]      1.463455\n",
       "[20]         1.467650\n",
       "[10]         1.474535\n",
       "[2, 2]       1.480549\n",
       "[1, 10]      1.580852"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Obtain MAE from modifying the hidden layers\n",
    "def hlayerResults(hlayer):\n",
    "    mlp = MLPRegressor(hidden_layer_sizes = hlayer,\n",
    "                       alpha= 0.01,\n",
    "                       activation = 'logistic',\n",
    "                       learning_rate_init = 0.001,\n",
    "                       max_iter = 2000)\n",
    "    mlp.fit(xtrain2,ytrain)\n",
    "    ypred = mlp.predict(xtest2)\n",
    "    return mean_absolute_error(ytest, ypred)\n",
    "#Let's try some \"typical\" values, with 2 layers, all combinations\n",
    "#of 5-spaced intervals\n",
    "hlayers = [[x,y] for x in range(5,21,5) for y in range(5,21,5)]\n",
    "#Now let's add a few \"weirder\" ones to see what happens:\n",
    "hlayers.extend([[1,10],[10,1],[2,2],[10,2,10],[10,5,10],\n",
    "                [20,5,20],[20,2,20],[10],[20]])\n",
    "dfResults = pd.DataFrame(columns=['MAE'])\n",
    "for hlayer in hlayers:\n",
    "    dfResults.loc[str(hlayer)] = {'MAE' : hlayerResults(hlayer)}\n",
    "#Print the results\n",
    "dfResults.sort_values('MAE')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The best results were [20, 5] by a small margin. A hidden layer architecture of [20, 5] is not excessively large, considering there are 10 input variables, so the selection seems reasonable. However, [10,5] might have been a good choice, as well. (Determining a reasonable hidden layer architecture is a skill that comes with practice using many different types of datasets.)\n",
    "\n",
    "### Learning Rate Selection\n",
    "\n",
    "Finally, let's select an appropriate learning rate. As the learning rate is most dependent on the hidden layer architecture, it should be performed last in this analysis, although of course as mentioned before, all hyperparameter selection should ideally be performed simultaneously."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAiMAAAGBCAYAAAC0FrHJAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nJCXKMCJJkhJlGJEkSYn6/wHG306S0QvNTAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x281252d4898>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Obtain MAE from modifying learning rate\n",
    "def lrResults(lr):\n",
    "    mlp = MLPRegressor(hidden_layer_sizes = [20,5],\n",
    "                       alpha=0.01,\n",
    "                       activation = 'logistic',\n",
    "                       learning_rate_init = lr,\n",
    "                       max_iter = 4000)\n",
    "    mlp.fit(xtrain2,ytrain)\n",
    "    ypred = mlp.predict(xtest2)\n",
    "    return mean_absolute_error(ytest, ypred)\n",
    "#Run the possibilities in a loop\n",
    "results4 = []\n",
    "lrs = np.logspace(-4, -1, num=20)\n",
    "for lr in lrs:\n",
    "    results4.append(lrResults(lr))\n",
    "#Plot the results\n",
    "plt.plot(lrs,results4)\n",
    "plt.xscale('log')\n",
    "plt.xlabel('Learning Rate')\n",
    "plt.ylabel('MAE');"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Based on these results, any learning rate less than 10^-2 is appropriate.\n",
    "\n",
    "### Final Results\n",
    "\n",
    "The best achieved MAE error was 1.42, using a perceptron architecture with 2 hidden layers ([20,5]), an alpha of 0.01, a learning rate of 0.01, and a logistic activation function. Compare these results to the ones obtained during Part 1, which achieved a MAE of 1.568. \n",
    "\n",
    "(But again, these results are not directly comparable due to the different training and testing dataset divisions. If you are interested in comparable results, please keep your training and testing datasets consistent across all models. I did not do so because I did not want to personally host a copy of the Abalone dataset, which I do not own)\n",
    "\n",
    "Despite the high accuracy of the neural network, there are several disadvantages to neural network modeling, including the difficulty of hyperparameter tuning. Notice that we simply used a trial-and-error process to select the hyperparameters. There are very few hard-and-fast rules for selecting appropriate hyperparameters, and it can often be more of an art than a science. \n",
    "\n",
    "Although cross-validation is meant to minimize over-generalization, sometimes it is necessary to manually examine models for potential over-fitting issues. Being able to quickly and easily interpret models can be a significant advantage in many cases. Of course, if all you need is accuracy, neural networks might be a great choice.\n",
    "\n",
    "And after all this work, did we actually solve the core problem? Personally, I would say no. An MAE of ~1.4 rings is not good enough for the researchers to no longer need to measure the rings manually. I would recommend the manual ring measuring process in this case, unfortunately."
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda root]",
   "language": "python",
   "name": "conda-root-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}