base_perceptron.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Implementação de rede perceptron em python"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Objetivo"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Esse notebook tem como objetivo mostrar uma implementação detalhada de uma rede neural perceptron utilizando python e a biblioteca numpy."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Como exemplo, utilizaremos o problema clássico de classificação de flores utilizando o famoso dataset iris.csv, com simplificação para duas classes de flores."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Carregando as bibliotecas que iremos utilizar\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Carregando funções auxiliares\n",
    "from funcoes_utils import carregar_dados"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Abrindo o nosso dataset utilizando a função carregar_dados\n",
    "header, features, saida_y = carregar_dados('iris.csv')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plotando os valores das features teremos um gráfico que claramente pode ser separado linearmente, ou seja, podemos aplicar a rede perceptron para solucionar nosso problema de classificação de flores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0,0.5,'Sepal Length')"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.scatter(features[:,1],features[:,0], c=saida_y)\n",
    "plt.title(\"Setosa x versicolor\" )\n",
    "plt.xlabel('Sepal Width')\n",
    "plt.ylabel('Sepal Length')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Criaremos a classe perceptron\n",
    "\n",
    "class Perceptron(object):\n",
    "    def __init__(self, alpha=0.01, n_features=4, n_iter = 2000):\n",
    "        self.pesos = np.random.uniform(-1,1, n_features+1) # adicionamos +1 por causa do bias\n",
    "        self.alpha = 0.01\n",
    "        self.n_epocas = n_iter\n",
    "        self.bias = 1\n",
    "    \n",
    "    def funcao_ativacao(self, y):\n",
    "        if y >= 0.0:\n",
    "            return 1\n",
    "        else:\n",
    "            return 0\n",
    "    \n",
    "    def treinamento_perceptron(self, features, resultados):\n",
    "        #inserindo o bias de 1 nas nossas features\n",
    "        features = np.insert(features[:,], len(features[0]), self.bias, axis=1)\n",
    "        for i in range(self.n_epocas):\n",
    "            #print(\"Estamos na época: {}\".format(i))\n",
    "            erro = 0\n",
    "            for j, linha in enumerate(features):\n",
    "                y = self.pesos.dot(linha)\n",
    "                if self.funcao_ativacao(y) != resultados[j]:\n",
    "                    erro +=1\n",
    "                    self.pesos += self.alpha*(resultados[j] - y)*linha\n",
    "            if erro == 0:\n",
    "                print(\"Script encerrado em época: {}\".format(i))\n",
    "                break\n",
    "     \n",
    "    def predicao(self, dados):\n",
    "        # podemos passar um conjunto de valores ou apenas um vetor de uma dimensao\n",
    "        if np.ndim(dados) == 1:\n",
    "            dados = np.insert(dados,len(dados),self.bias) #inserindo o bias\n",
    "            return self.funcao_ativacao(self.pesos.dot(dados)) # realizando a predição\n",
    "        else: #se tivermos um conjuntno de dados com mais de uma dimensão\n",
    "            dados = np.insert(dados[:,], len(dados[0]), self.bias, axis=1) #inserindo o bias\n",
    "            resultado_predicao = [self.funcao_ativacao(self.pesos.dot(linha)) for linha in dados]\n",
    "            return resultado_predicao"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "perceptron = Perceptron(n_features=4)\n",
    "perceptron.treinamento_perceptron(features, saida_y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "predicoes = perceptron.predicao(features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "# podemos criar uma função para calcular a acurácia\n",
    "def acuracia(saida_y, predictions):\n",
    "    correct = 0\n",
    "    for x in range(len(saida_y)):\n",
    "        if saida_y[x] == predictions[x]:#Compara os respectivos valores\n",
    "            correct += 1\n",
    "    return float(correct)/len(saida_y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.99"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "acuracia(saida_y, predicoes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}