Forecasting_Model_based_methods.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "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": "3.7.4"
    },
    "colab": {
      "name": "Forecasting_Model based methods.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "sxXLV1QPv_3s"
      },
      "source": [
        "import numpy as np\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GB2YSGsgv_3s"
      },
      "source": [
        "Walmart = pd.read_csv(\"footfalls.csv\")\n",
        "Walmart.Footfalls.plot()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "shifEIr3v_3s"
      },
      "source": [
        "Walmart.head()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RJlEJSYPv_3s"
      },
      "source": [
        "Walmart[\"Date\"] = pd.to_datetime(Walmart.Month,format=\"%b-%y\")\n",
        "#look for c standard format codes\n",
        "\n",
        "# Extracting Day, weekday name, month name, year from the Date column using \n",
        "# Date functions from pandas \n",
        "\n",
        "Walmart[\"month\"] = Walmart.Date.dt.strftime(\"%b\") # month extraction\n",
        "Walmart[\"year\"] = Walmart.Date.dt.strftime(\"%Y\") # year extraction\n",
        "\n",
        "#Walmart[\"Day\"] = Walmart.Date.dt.strftime(\"%d\") # Day extraction\n",
        "#Walmart[\"wkday\"] = Walmart.Date.dt.strftime(\"%A\") # weekday extraction"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Mv-5gU3Sv_3t"
      },
      "source": [
        "plt.figure(figsize=(12,8))\n",
        "heatmap_y_month = pd.pivot_table(data=Walmart,values=\"Footfalls\",index=\"year\",columns=\"month\",aggfunc=\"mean\",fill_value=0)\n",
        "sns.heatmap(heatmap_y_month,annot=True,fmt=\"g\") #fmt is format of the grid values\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YtQlhEN8v_3t"
      },
      "source": [
        "# Boxplot for ever\n",
        "plt.figure(figsize=(8,6))\n",
        "plt.subplot(211)\n",
        "sns.boxplot(x=\"month\",y=\"Footfalls\",data=Walmart)\n",
        "plt.subplot(212)\n",
        "sns.boxplot(x=\"year\",y=\"Footfalls\",data=Walmart)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5V-97vg9v_3t"
      },
      "source": [
        "#month =['Jan','Feb','Mar','Apr','May','Jun','Jul','Aug','Sep','Oct','Nov','Dec'] \n",
        "#import numpy as np\n",
        "#p = Walmart[\"Month\"][0]\n",
        "#p[0:3]\n",
        "#Walmart['months']= 0\n",
        "\n",
        "#for i in range(159):\n",
        "#    p = Walmart[\"Month\"][i]\n",
        "#    Walmart['months'][i]= p[0:3]\n",
        "    \n",
        "#month_dummies = pd.DataFrame(pd.get_dummies(Walmart['months']))\n",
        "#Walmart1 = pd.concat([footfalls,month_dummies],axis = 1)\n",
        "\n",
        "#Walmart1[\"t\"] = np.arange(1,160)\n",
        "\n",
        "#Walmart1[\"t_squared\"] = Walmart1[\"t\"]*Walmart1[\"t\"]\n",
        "#Walmart1.columns\n",
        "#Walmart1[\"log_footfalls\"] = np.log(Walmart1[\"Footfalls \"])\n",
        "#Walmart1.rename(columns={\"Footfalls \": 'Footfalls'}, inplace=True)\n",
        "#Walmart1.Footfalls.plot()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "gisksCHUv_3t"
      },
      "source": [
        "plt.figure(figsize=(12,3))\n",
        "sns.lineplot(x=\"year\",y=\"Footfalls\",data=Walmart)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "t7AtBl_Sv_3t"
      },
      "source": [
        "# Splitting data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KuEW71Spv_3t"
      },
      "source": [
        "Train = Walmart.head(147)\n",
        "Test = Walmart.tail(12)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "o5pwEbMLv_3t"
      },
      "source": [
        "#Linear Model\n",
        "import statsmodels.formula.api as smf \n",
        "\n",
        "linear_model = smf.ols('Footfalls~t',data=Train).fit()\n",
        "pred_linear =  pd.Series(linear_model.predict(pd.DataFrame(Test['t'])))\n",
        "rmse_linear = np.sqrt(np.mean((np.array(Test['Footfalls'])-np.array(pred_linear))**2))\n",
        "rmse_linear"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Kqv2o-nEv_3t"
      },
      "source": [
        "#Exponential\n",
        "\n",
        "Exp = smf.ols('log_footfalls~t',data=Train).fit()\n",
        "pred_Exp = pd.Series(Exp.predict(pd.DataFrame(Test['t'])))\n",
        "rmse_Exp = np.sqrt(np.mean((np.array(Test['Footfalls'])-np.array(np.exp(pred_Exp)))**2))\n",
        "rmse_Exp\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nIOnHc8Jv_3t"
      },
      "source": [
        "#Quadratic \n",
        "\n",
        "Quad = smf.ols('Footfalls~t+t_square',data=Train).fit()\n",
        "pred_Quad = pd.Series(Quad.predict(Test[[\"t\",\"t_square\"]]))\n",
        "rmse_Quad = np.sqrt(np.mean((np.array(Test['Footfalls'])-np.array(pred_Quad))**2))\n",
        "rmse_Quad"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "I8SOV_a9v_3t"
      },
      "source": [
        "#Additive seasonality \n",
        "\n",
        "add_sea = smf.ols('Footfalls~Jan+Feb+Mar+Apr+May+Jun+Jul+Aug+Sep+Oct+Nov',data=Train).fit()\n",
        "pred_add_sea = pd.Series(add_sea.predict(Test[['Jan','Feb','Mar','Apr','May','Jun','Jul','Aug','Sep','Oct','Nov']]))\n",
        "rmse_add_sea = np.sqrt(np.mean((np.array(Test['Footfalls'])-np.array(pred_add_sea))**2))\n",
        "rmse_add_sea"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "l83nPnNUv_3t"
      },
      "source": [
        "#Additive Seasonality Quadratic \n",
        "\n",
        "add_sea_Quad = smf.ols('Footfalls~t+t_square+Jan+Feb+Mar+Apr+May+Jun+Jul+Aug+Sep+Oct+Nov',data=Train).fit()\n",
        "pred_add_sea_quad = pd.Series(add_sea_Quad.predict(Test[['Jan','Feb','Mar','Apr','May','Jun','Jul','Aug','Sep','Oct','Nov','t','t_square']]))\n",
        "rmse_add_sea_quad = np.sqrt(np.mean((np.array(Test['Footfalls'])-np.array(pred_add_sea_quad))**2))\n",
        "rmse_add_sea_quad"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "POeCiSPVv_3t"
      },
      "source": [
        "##Multiplicative Seasonality\n",
        "\n",
        "Mul_sea = smf.ols('log_footfalls~Jan+Feb+Mar+Apr+May+Jun+Jul+Aug+Sep+Oct+Nov',data = Train).fit()\n",
        "pred_Mult_sea = pd.Series(Mul_sea.predict(Test))\n",
        "rmse_Mult_sea = np.sqrt(np.mean((np.array(Test['Footfalls'])-np.array(np.exp(pred_Mult_sea)))**2))\n",
        "rmse_Mult_sea"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FFgAJMTQv_3t"
      },
      "source": [
        "#Multiplicative Additive Seasonality \n",
        "\n",
        "Mul_Add_sea = smf.ols('log_footfalls~t+Jan+Feb+Mar+Apr+May+Jun+Jul+Aug+Sep+Oct+Nov',data = Train).fit()\n",
        "pred_Mult_add_sea = pd.Series(Mul_Add_sea.predict(Test))\n",
        "rmse_Mult_add_sea = np.sqrt(np.mean((np.array(Test['Footfalls'])-np.array(np.exp(pred_Mult_add_sea)))**2))\n",
        "rmse_Mult_add_sea "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1XShRA2-v_3t"
      },
      "source": [
        "#Compare the results \n",
        "\n",
        "data = {\"MODEL\":pd.Series([\"rmse_linear\",\"rmse_Exp\",\"rmse_Quad\",\"rmse_add_sea\",\"rmse_add_sea_quad\",\"rmse_Mult_sea\",\"rmse_Mult_add_sea\"]),\"RMSE_Values\":pd.Series([rmse_linear,rmse_Exp,rmse_Quad,rmse_add_sea,rmse_add_sea_quad,rmse_Mult_sea,rmse_Mult_add_sea])}\n",
        "table_rmse=pd.DataFrame(data)\n",
        "table_rmse.sort_values(['RMSE_Values'])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oZbRNcv_v_3t"
      },
      "source": [
        "#### Predict for new time period"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sFtvLjSMv_3t"
      },
      "source": [
        "predict_data = pd.read_csv(\"Predict_new.csv\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "o9Cu5tjZv_3t"
      },
      "source": [
        "predict_data"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eY504_-xv_3t"
      },
      "source": [
        "#Build the model on entire data set\n",
        "model_full = smf.ols('Footfalls~t+t_square+Jan+Feb+Mar+Apr+May+Jun+Jul+Aug+Sep+Oct+Nov',data=Walmart).fit()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yMVSLkK3v_3t"
      },
      "source": [
        "pred_new  = pd.Series(model_full.predict(predict_data))\n",
        "pred_new"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1D1R_RIvv_3t"
      },
      "source": [
        "predict_data[\"forecasted_Footfalls\"] = pd.Series(pred_new)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Fml94P-Rv_3t"
      },
      "source": [
        "predict_data"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ciZMtQz_v_3t"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}