apply_model/fast_light_gbm_applier.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Save LightGBM model in CatBoost format to use fast CatBoost appliers\n",
    "\n",
    "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/catboost/tutorials/blob/master/apply_model/fast_light_gbm_applier.ipynb)\n",
    "\n",
    "To save LightGBM in CatBoost format you need to convert LightGBM model to ONNX, and then to convert the model from ONNX to CatBoost.\n",
    "+ Save LightGBM model in the ONNX format\n",
    "+ Load the ONNX model into CatBoost using the load_model() method\n",
    "+ Apply your model in CatBoost using the predict() method or save it as file and use with other appliers\n",
    "\n",
    "Note, that this tutorial will only work for Python 3.* since onnxmltools \n",
    "\n",
    "Let us follow this scenario step-by-step for a LightGBM model."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "Download the MSRank dataset and import the necessary packages:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from catboost import datasets, CatBoostRegressor\n",
    "\n",
    "from lightgbm import LGBMRegressor\n",
    "\n",
    "import onnxmltools\n",
    "from onnxconverter_common import *\n",
    "\n",
    "\n",
    "train_df, _ = datasets.msrank_10k()\n",
    "X, Y = train_df[train_df.columns[1:]], train_df[train_df.columns[0]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "Build a model:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1.30604501 1.60390655 0.35207384 ... 1.18672199 0.55631924 0.54655847]\n"
     ]
    }
   ],
   "source": [
    "model = LGBMRegressor()\n",
    "model.fit(X, Y)\n",
    "predict = model.predict(X)\n",
    "print(predict)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "Save the model in the ONNX format:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "The maximum opset needed by this model is only 1.\n",
      "The maximum opset needed by this model is only 1.\n"
     ]
    }
   ],
   "source": [
    "features_count = len(X.columns)\n",
    "onnx_model = onnxmltools.convert_lightgbm(model, name='LightGBM', initial_types=[['input', FloatTensorType([0, features_count])]])\n",
    "onnxmltools.utils.save_model(onnx_model, 'model.onnx')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "Load the ONNX model into CatBoost and print predictions to make sure they are correct."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1.30604502 1.60390654 0.35207381 ... 1.18672202 0.55631925 0.54655849]\n"
     ]
    }
   ],
   "source": [
    "catboost_model = CatBoostRegressor()\n",
    "catboost_model.load_model('model.onnx', format='onnx')\n",
    "catboost_predict = catboost_model.predict(X)\n",
    "print(catboost_predict)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Save the model in CatBoost format for later use in C++, C#, Java or Python code with fast CatBoost applier."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "catboost_model.save_model('model.bin')"
   ]
  }
 ],
 "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"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}