04_Conceptos_básicos_de_estadísticas.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Fundamentos de estadística\n",
    "\n",
    "Como buen cientista de datos, vamos a repasar la teoría estadística que necesitamos como base de apoyo en nuestros análisis y construcción de modelos.\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"Estadisticas_bases.PNG\">"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Utilizamos las librerías para ello\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Medidas de centralización\n",
    "\n",
    "Nos sirve para ver cómo se sitúan los datos. Son la media, mediana, percentiles y moda.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Generamos una lista de números aleatorios y revisaremos sus medidas de centralización\n",
    "\n",
    "import random # Módulo para generar números aleatorios\n",
    "n = 10000 # Muestras\n",
    "m = 1000000 # Límite superior del conjunto\n",
    "rango = range(m) # Rango de números para hacer la selección, en este caso del 0 a m\n",
    "A = random.sample(rango, n) # 'A' es el conjunto de datos numéricos aleatorios, de tamaño n\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "496113.7417"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Promedio (media)\n",
    "np.mean(A)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "494419.0"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Mediana\n",
    "np.median(A)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ModeResult(mode=array([222]), count=array([1]))"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Para la moda, utilizaremos otra librería\n",
    "\n",
    "from scipy import stats\n",
    "\n",
    "stats.mode(A)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "239168.75"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Percentil 25\n",
    "np.percentile(A, 25)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Medidas de dispersión\n",
    "\n",
    "La varianza y desviación típica, nos indica si los valores se desplazan mucho o poco con respecto de la media.\n",
    "\n",
    "La *varianza* es como se aleja cada valor de la media\n",
    " * La varianza eleva los valores al cuadradonos introduce en una nueva dimensión.\n",
    " * Puede no tener sentido.\n",
    " \n",
    "La *desviación típica* es la raíz cuadrada de la varianza. \n",
    " * Con la desviación típica volvemos a la dimensión original.\n",
    "\n",
    "*Coeficiente de variación*: nos mide la variabilidad relativa entre la desviación típica entre la media."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "84260856060.67038"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Varianza\n",
    "\n",
    "np.var(A)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "290277.2055478528"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Desviación típica (estándar)\n",
    "\n",
    "np.std(A)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "58.51021270912173"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# coeficiente de variacion std/mean*100\n",
    "# variabilidad relativa entre la media y la std, si hay mucha variabilidad será grande el coeficiente.\n",
    "\n",
    "np.std(A)/np.mean(A)*100\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Medidas de asimetría\n",
    "\n",
    "Momento de orden r, respecto a la media. El momento de orden r. son los momentos de distribución respecto a la media.\n",
    "\n",
    "3.1 Asimetría de Fisher\n",
    "* Si el coeficiente es = 0; Significa que vuestra función es perfectamente simetríca, se distribuye igual, por ejemplo la distribución normal. Raro es que salga cero\n",
    "\n",
    "* Si el coeficiente es >0; Significa que cuánto más positivo es este valor más desplazada está la distribución hacía la izquierda, de modo que tenemos una asimetría positiva, nos queda la media muy por encima de la distribución.\n",
    "\n",
    "* Si el el coeficiente es <0; Significa que cuánto más negativo es este valor más desplazado está la distribución hacía la derecha, de modo que tenemos una asimetría negativa, nos queda la media muy por debajo de la distribución.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.01676462523284354"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import scipy.stats as sp # Para modelos de asimetría\n",
    "\n",
    "asimetria = sp.skew(A)\n",
    "asimetria"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "3.2 Curtosis\n",
    "\n",
    "* igual 0 Mesocúrtica: Distribución perfecta, asemejada a la distribución normal en forma, no en valores. Está compensado tanto el centro como las colas.\n",
    "\n",
    "* mayor a 0 Leptocúrtica: Distribución donde se le concentran mucho los datos en el valor central, y apenas tiene cola.\n",
    "\n",
    "* menor a 0 Platicúrtica: Distribución donde hay pocos valores que se concentren respecto al valor central (media) y hay muchos que aparecen hacia las colas, se concentran más en los laterales. Existe valor central, pero también hay mucha presencia de colas directamente en la distribución de nuestros datos.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "-1.2201063205820903"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "curtosis=sp.kurtosis(A)\n",
    "curtosis\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Muestreo aleatorio\n",
    "\n",
    "<img src=\"muestreo.PNG\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 5. Contrastes de hipótesis\n",
    "\n",
    "<img src=\"Cnotraste_hipotesis.PNG\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "$H_0$ es lo que defenderemos a morir! En caso de que no quede más remedio, nos quedaremos con $H_1$\n",
    "\n",
    "TCL = Teorema central del Límite\n",
    "\n",
    "<img src=\"Casos_hipotesis.PNG\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "¿Y cómo aceptamos/rechazamos la hipótesis nula ($H_0$)?\n",
    "\n",
    "<img src=\"Rechazo_hipotesis.PNG\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "En este caso tenemos el gráfico de una distribución de una variable aleatoria $X$, con $Z_1$ y $Z_2$ dos estadísticos correspondientes a dos valores de la variable aleatoria en cuestión y llamemos $P_1$ y $P_2$ a las probabilidades encerradas por debajo de la curva justo a la izquierda. Entonces la probabilidad de que la variable aleatoria tome un valor menor a $Z_1$ es $P_1$, en notación matemática sería:\n",
    "\n",
    "$$P(X < Z_1) = P_1 $$\n",
    "\n",
    "Y de forma complementaria\n",
    "\n",
    "$$P(X > Z_1) = 1- P_1 $$\n",
    "\n",
    "Con esto podemos definir niveles de confianza, pero espera, ¿que son los niveles de confianza? Son los valores límites que estamos dispuestos a considerar para defender una hipótesis nula $H_0$ como verdadera, dicho de otra forma, *es la probabilidad con la que nosotros estamos seguros de que la variable aleatoria va a caer ahí dentro*\n",
    "\n",
    "<img src=\"Intervalos_confianza.PNG\">\n",
    "\n",
    "Oook, pero y ¿qué es eso del nivel de significación? Es básicamente la representación de la probabilidad de que la hipótesis nula $H_0$ **no** sea cierta, y se representa con $\\alpha$. Simplemente es el opuesto al nivel de confianza.\n",
    "\n",
    "¿Y eso del $p_{valor}$? Es la probabilidad de que la función de distribución supere el valor del estadístico de contraste (que viene de un z-test o t-test), es la probabilidad máxima de caer fuera del estadístico de contraste.    \n",
    "\n",
    "<img src=\"p_valor.PNG\">\n",
    "\n",
    "\n",
    "\n",
    "Y entonces, ¿cuando aceptamos la hipótesis nula $H_0$?\n",
    "\n",
    "\n",
    "\n",
    "<img src=\"Conclusion_hipotesis.PNG\">\n",
    "\n",
    "Entonces, el resumen del contraste de hipótesis es:\n",
    "\n",
    "1. Definir hipótesis nula ($\\mu_0$) y alternativa uni o bilateral.\n",
    "2. Tomar una muestra aletoria de tamaño $n$ y calcular el valor del estimador (promedio, proporción, etc..)\n",
    "3. Calcular el estadístico de contraste $Z-valor$ o $t-valor$.\n",
    "4. Calcular el $p_{valor}$ asociado\n",
    "5. Comparar $p_{valor}$ y nivel de significacion y decidir.\n",
    "\n",
    "¿Y un ejemplo práctico de esto? Vamos por ello!\n",
    "\n",
    "Ejemplo: Just Eat\n",
    "\n",
    "El pizzero de Just-Eat **afirma** que trae la comida en un **tiempo promedio** inferior a **20 minutos** con una **desviación típica de 3**.\n",
    "\n",
    "Como sospechamos que es falso, tomamos 64 de las entregas de la última semana y obtenemos una media de 21.2 minutos.\n",
    "\n",
    "¿Podemos aceptar su afirmación a un nivel de confianza del 95%?\n",
    "\n",
    "Resolvamos:\n",
    "\n",
    "**1. Identificar hipótesis nula y alternativa**\n",
    "\n",
    "En este caso sería: \n",
    "\n",
    "$H_0$: tiempo promedio inferior a 20 minutos -> $\\mu$ <= 20\n",
    "\n",
    "La hipótesis alternativa sería el complemento:\n",
    "\n",
    "$H_1$: tiempo promedio superior a 20 minutos -> $\\mu$ > 20\n",
    "\n",
    "Además el enunciado me entrega el valor de la desviación típica (estándar): $\\sigma$ = 3\n",
    "\n",
    "**2. Tomar una muestra aletoria de tamaño $n$ y calcular el valor del estimador**\n",
    "\n",
    "En este caso queremos estimar el promedio, y para ello se utilizó una muestra de 64 entregas, obteniendo un valor promedio de 21.2 minutos, entonces:\n",
    "\n",
    "$\\bar{X} = 21.2$ y $n = 64$\n",
    "\n",
    "\n",
    "**3. Calcular el estadístico de contraste $Z-valor$ o $t-valor$**\n",
    "\n",
    "Como el valor de desviación estándar es conocido, utilizamos el estadístico de prueba $Z_{valor}$. Sabemos que:\n",
    "\n",
    "$$ Z = \\frac{\\bar{X} - \\mu}{\\frac{\\sigma}{\\sqrt{n}}}$$\n",
    "\n",
    "Reemplazamos:\n",
    "\n",
    "$$ Z = \\frac{21.2 - 20}{\\frac{3}{\\sqrt{64}}} = 3.2$$\n",
    "\n",
    "**4. Calcular el $p_{valor}$ asociado**\n",
    "\n",
    "Sabemos que: $$p = P(Z>3.2) = 1 - P(Z<3.2)$$\n",
    "\n",
    "Aquí podemos utilizar alguna tabla de la distribución normal estandarizada, o bien generar nosotros el valor con la librería **scipy**\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9993128620620841"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from scipy.stats import norm\n",
    "\n",
    "norm.cdf(3.2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Entonces el valor p sería:\n",
    "\n",
    "$$p = 1 - 0.999 = 0.001$$\n",
    "$$p = 0.001 $$\n",
    "\n",
    "**5. Comparar $p_{valor}$ y nivel de significacion y decidir.**\n",
    "\n",
    "En este caso el nivel de confianza es del 95%, por lo tanto nuestreo nivel de significancia es el complemento $\\alpha = 0.05$\n",
    "\n",
    "Y como se cumple que: $$0.001 < 0.05 \\Rightarrow p_{valor} < \\alpha$$\n",
    "\n",
    "Por lo tanto, podemos llamar al pizzero de Just-Eat y comentarle que tenemos evidencia suficiente para rechazar su afirmación (rechazar la hipotesis nula) y que en definitiva se demora más de 20 minutos en entregar sus pizzas.\n",
    "\n",
    "¿Genial no?\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 6. Correlación\n",
    "\n",
    "De forma teórica, presentamos la correlación de Pearson, luego lo haremos de forma práctica\n",
    "\n",
    "<img src=\"Corr_Pearson.PNG\">"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>TV</th>\n",
       "      <th>Radio</th>\n",
       "      <th>Newspaper</th>\n",
       "      <th>Sales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>230.1</td>\n",
       "      <td>37.8</td>\n",
       "      <td>69.2</td>\n",
       "      <td>22.1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>44.5</td>\n",
       "      <td>39.3</td>\n",
       "      <td>45.1</td>\n",
       "      <td>10.4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>17.2</td>\n",
       "      <td>45.9</td>\n",
       "      <td>69.3</td>\n",
       "      <td>9.3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>151.5</td>\n",
       "      <td>41.3</td>\n",
       "      <td>58.5</td>\n",
       "      <td>18.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>180.8</td>\n",
       "      <td>10.8</td>\n",
       "      <td>58.4</td>\n",
       "      <td>12.9</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      TV  Radio  Newspaper  Sales\n",
       "0  230.1   37.8       69.2   22.1\n",
       "1   44.5   39.3       45.1   10.4\n",
       "2   17.2   45.9       69.3    9.3\n",
       "3  151.5   41.3       58.5   18.5\n",
       "4  180.8   10.8       58.4   12.9"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Hagámoslo de forma práctica, con un data set\n",
    "\n",
    "data_ads = pd.read_csv('/Users/fsanmartin/python-ml-course-master/datasets/ads/Advertising.csv')\n",
    "data_ads.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>TV</th>\n",
       "      <th>Radio</th>\n",
       "      <th>Newspaper</th>\n",
       "      <th>Sales</th>\n",
       "      <th>corrn</th>\n",
       "      <th>corrl</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>230.1</td>\n",
       "      <td>37.8</td>\n",
       "      <td>69.2</td>\n",
       "      <td>22.1</td>\n",
       "      <td>670.896956</td>\n",
       "      <td>6898.548306</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>44.5</td>\n",
       "      <td>39.3</td>\n",
       "      <td>45.1</td>\n",
       "      <td>10.4</td>\n",
       "      <td>371.460206</td>\n",
       "      <td>10514.964306</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>17.2</td>\n",
       "      <td>45.9</td>\n",
       "      <td>69.3</td>\n",
       "      <td>9.3</td>\n",
       "      <td>613.181206</td>\n",
       "      <td>16859.074806</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>151.5</td>\n",
       "      <td>41.3</td>\n",
       "      <td>58.5</td>\n",
       "      <td>18.5</td>\n",
       "      <td>19.958456</td>\n",
       "      <td>19.869306</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>180.8</td>\n",
       "      <td>10.8</td>\n",
       "      <td>58.4</td>\n",
       "      <td>12.9</td>\n",
       "      <td>-37.892794</td>\n",
       "      <td>1139.568806</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      TV  Radio  Newspaper  Sales       corrn         corrl\n",
       "0  230.1   37.8       69.2   22.1  670.896956   6898.548306\n",
       "1   44.5   39.3       45.1   10.4  371.460206  10514.964306\n",
       "2   17.2   45.9       69.3    9.3  613.181206  16859.074806\n",
       "3  151.5   41.3       58.5   18.5   19.958456     19.869306\n",
       "4  180.8   10.8       58.4   12.9  -37.892794   1139.568806"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Agregaremos una columna comenzando con un coeficiente de correlación entre \n",
    "# Los gastos por publicidad en TV con respecto a las ventas\n",
    "\n",
    "data_ads['corrn'] = (data_ads['TV'] - np.mean(data_ads['TV'])) * (data_ads['Sales'] - np.mean(data_ads['Sales']))\n",
    "data_ads['corrl'] = (data_ads['TV'] - np.mean(data_ads['TV']))**2\n",
    "\n",
    "data_ads.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>TV</th>\n",
       "      <th>Radio</th>\n",
       "      <th>Newspaper</th>\n",
       "      <th>Sales</th>\n",
       "      <th>corrn</th>\n",
       "      <th>corrl</th>\n",
       "      <th>corr2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>230.1</td>\n",
       "      <td>37.8</td>\n",
       "      <td>69.2</td>\n",
       "      <td>22.1</td>\n",
       "      <td>670.896956</td>\n",
       "      <td>6898.548306</td>\n",
       "      <td>65.246006</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>44.5</td>\n",
       "      <td>39.3</td>\n",
       "      <td>45.1</td>\n",
       "      <td>10.4</td>\n",
       "      <td>371.460206</td>\n",
       "      <td>10514.964306</td>\n",
       "      <td>13.122506</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>17.2</td>\n",
       "      <td>45.9</td>\n",
       "      <td>69.3</td>\n",
       "      <td>9.3</td>\n",
       "      <td>613.181206</td>\n",
       "      <td>16859.074806</td>\n",
       "      <td>22.302006</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>151.5</td>\n",
       "      <td>41.3</td>\n",
       "      <td>58.5</td>\n",
       "      <td>18.5</td>\n",
       "      <td>19.958456</td>\n",
       "      <td>19.869306</td>\n",
       "      <td>20.048006</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>180.8</td>\n",
       "      <td>10.8</td>\n",
       "      <td>58.4</td>\n",
       "      <td>12.9</td>\n",
       "      <td>-37.892794</td>\n",
       "      <td>1139.568806</td>\n",
       "      <td>1.260006</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      TV  Radio  Newspaper  Sales       corrn         corrl      corr2\n",
       "0  230.1   37.8       69.2   22.1  670.896956   6898.548306  65.246006\n",
       "1   44.5   39.3       45.1   10.4  371.460206  10514.964306  13.122506\n",
       "2   17.2   45.9       69.3    9.3  613.181206  16859.074806  22.302006\n",
       "3  151.5   41.3       58.5   18.5   19.958456     19.869306  20.048006\n",
       "4  180.8   10.8       58.4   12.9  -37.892794   1139.568806   1.260006"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_ads['corr2'] = (data_ads['Sales'] - np.mean(data_ads['Sales']))**2\n",
    "data_ads.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.782224424861606"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "corr_pearson = sum(data_ads['corrn']) / np.sqrt(sum(data_ads['corrl'])*sum(data_ads['corr2']))\n",
    "corr_pearson"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Y cómo ya saben, si esto hay que hacerlo muchas veces, mejor tener una función que lo realice"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "def corr_coeff(df, var1, var2):\n",
    "    \n",
    "    df['corrn'] = (df[var1] - np.mean(df[var1])) * (df[var2] - np.mean(df[var2]))\n",
    "    \n",
    "    df['corrl'] = (df[var1] - np.mean(df[var1]))**2\n",
    "    \n",
    "    df['corr2'] = (df[var2] - np.mean(df[var2]))**2\n",
    "    \n",
    "    corr_p = sum(df['corrn']) / np.sqrt(sum(df['corrl'])*sum(df['corr2']))\n",
    "    \n",
    "    return corr_p"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TV, TV : 1.0\n",
      "TV, Radio : 0.05480866446583009\n",
      "TV, Newspaper : 0.056647874965056993\n",
      "TV, Sales : 0.782224424861606\n",
      "Radio, TV : 0.05480866446583009\n",
      "Radio, Radio : 1.0\n",
      "Radio, Newspaper : 0.3541037507611752\n",
      "Radio, Sales : 0.5762225745710553\n",
      "Newspaper, TV : 0.056647874965056993\n",
      "Newspaper, Radio : 0.3541037507611752\n",
      "Newspaper, Newspaper : 1.0\n",
      "Newspaper, Sales : 0.22829902637616525\n",
      "Sales, TV : 0.782224424861606\n",
      "Sales, Radio : 0.5762225745710553\n",
      "Sales, Newspaper : 0.22829902637616525\n",
      "Sales, Sales : 1.0\n"
     ]
    }
   ],
   "source": [
    "df = pd.read_csv('/Users/fsanmartin/python-ml-course-master/datasets/ads/Advertising.csv')\n",
    "\n",
    "# Calculemos las correlaciones entre todas las variables\n",
    "\n",
    "cols = df.columns.values\n",
    "\n",
    "for x in cols:\n",
    "    for y in cols:\n",
    "        print(x + \", \"+ y + \" : \" + str(corr_coeff(df, x, y)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Hagamos una gráfica de nube de puntos entre las ventas y los gastos en anuncios en los medios\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "plt.plot(df['TV'], df['Sales'], 'ro')\n",
    "plt.plot(df['Radio'], df['Sales'], 'go')\n",
    "plt.plot(df['Newspaper'], df['Sales'], 'bo')\n",
    "plt.title('Gastos en anuncios vs Ventas del producto');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>TV</th>\n",
       "      <th>Radio</th>\n",
       "      <th>Newspaper</th>\n",
       "      <th>Sales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>TV</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.054809</td>\n",
       "      <td>0.056648</td>\n",
       "      <td>0.782224</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Radio</th>\n",
       "      <td>0.054809</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.354104</td>\n",
       "      <td>0.576223</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Newspaper</th>\n",
       "      <td>0.056648</td>\n",
       "      <td>0.354104</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.228299</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Sales</th>\n",
       "      <td>0.782224</td>\n",
       "      <td>0.576223</td>\n",
       "      <td>0.228299</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                 TV     Radio  Newspaper     Sales\n",
       "TV         1.000000  0.054809   0.056648  0.782224\n",
       "Radio      0.054809  1.000000   0.354104  0.576223\n",
       "Newspaper  0.056648  0.354104   1.000000  0.228299\n",
       "Sales      0.782224  0.576223   0.228299  1.000000"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "## Pandas tiene un forma sencilla de realizar. volveremos a cargar el dataset\n",
    "\n",
    "df = pd.read_csv('/Users/fsanmartin/python-ml-course-master/datasets/ads/Advertising.csv')\n",
    "df.corr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x1e26b1f4fd0>"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Y una forma de representar esta matriz de correlación es:\n",
    "\n",
    "plt.matshow(df.corr()) # El color verde indica mayor correlación"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}