model_defaults.ipynb

{
 "cells": [
  {
   "metadata": {
    "trusted": true,
    "collapsed": false
   },
   "cell_type": "code",
   "source": "import math\n\ndef clean_mtry(x,mid,nfeat):\n    ret = []\n    for i in x:\n        if i < mid:\n            ret.append(max(2,math.floor(i)))\n        else:\n            ret.append(min(nfeat,math.ceil(i)))\n    return sorted(list(set(ret)))\n\n\ndef get_defaults(n_features, task_type=\"classification\",\n                 model_type=\"random_forest\",\n                 opt_criteria=\"fast\"):\n\n    \"\"\"\n    n_features       number of features\n    task_type        classification, regression\n    model_type       random_forest\n                  SVM [not implemented yet]\n\n    opt_criteria     quick = do only a few model searches\n                  comprehensive = do fine grained searches\n\n    \"\"\"\n          \n    # TODO...add more defaults for different model types\n    if model_type != \"random_forest\":\n        raise\n\n    bootstrap = [True]\n    oob_score = [True]\n    n_jobs = -1\n    max_depth = [None]\n\n    if opt_criteria == \"fast\":\n        n_estimators = [50]\n        min_samples_split = [2]\n        min_samples_leaf  = [1]\n\n        if task_type == \"classification\":\n            criterion = [\"gini\",\"entropy\"]\n            mid = math.sqrt(n_features)\n        elif task_type == \"regression\":\n            criterion = [\"mse\"]\n            mid = max(1, math.ceil(n_features / 3.0))\n        else:\n             raise\n        \n        initial = [0.85*mid,mid,1.15*mid]\n        max_features = clean_mtry(initial,mid,n_features)\n        \n    elif opt_criteria == \"comprehensive\":\n        n_estimators = [100,250]\n        min_samples_split = [2]\n        min_samples_leaf  = [1,2,5]\n\n        if task_type == \"classification\":\n            criterion = [\"gini\",\"entropy\"]\n            mid = math.sqrt(n_features)\n\n        elif task_type == \"regression\":\n            criterion = [\"mse\",\"mae\"]\n            mid = max(1, math.ceil(n_features / 3.0))\n        else:\n             raise\n\n        initial = [0.5*mid,0.85*mid,mid,1.15*mid,1.3*mid,1.5*mid,2*mid]\n        max_features = clean_mtry(initial,mid,n_features)\n                 \n    ret = {\"bootstrap\": bootstrap,\n           \"oob_score\": oob_score,\n           \"n_jobs\": -1,\n           \"max_depth\": max_depth,\n           \"max_features\": max_features,\n           \"criterion\": criterion,\n           \"min_samples_leaf\": min_samples_leaf,\n           \"min_samples_split\": min_samples_split,\n           \"n_estimators\": n_estimators}\n\n    return ret\n",
   "execution_count": 45,
   "outputs": []
  },
  {
   "metadata": {
    "trusted": true,
    "collapsed": false
   },
   "cell_type": "code",
   "source": "get_defaults(100, task_type=\"classification\",\n                 model_type=\"random_forest\",\n                 opt_criteria=\"fast\")",
   "execution_count": 46,
   "outputs": [
    {
     "execution_count": 46,
     "output_type": "execute_result",
     "data": {
      "text/plain": "{'bootstrap': [True],\n 'criterion': ['gini', 'entropy'],\n 'max_depth': [None],\n 'max_features': [8, 10, 12],\n 'min_samples_leaf': [1],\n 'min_samples_split': [2],\n 'n_estimators': [50],\n 'n_jobs': -1,\n 'oob_score': [True]}"
     },
     "metadata": {}
    }
   ]
  },
  {
   "metadata": {
    "trusted": true,
    "collapsed": false
   },
   "cell_type": "code",
   "source": "get_defaults(100, task_type=\"classification\",\n                 model_type=\"random_forest\",\n                 opt_criteria=\"comprehensive\")",
   "execution_count": 47,
   "outputs": [
    {
     "execution_count": 47,
     "output_type": "execute_result",
     "data": {
      "text/plain": "{'bootstrap': [True],\n 'criterion': ['gini', 'entropy'],\n 'max_depth': [None],\n 'max_features': [5, 8, 10, 12, 13, 15, 20],\n 'min_samples_leaf': [1, 2, 5],\n 'min_samples_split': [2],\n 'n_estimators': [100, 250],\n 'n_jobs': -1,\n 'oob_score': [True]}"
     },
     "metadata": {}
    }
   ]
  },
  {
   "metadata": {
    "trusted": true,
    "collapsed": false
   },
   "cell_type": "code",
   "source": "get_defaults(2000, task_type=\"classification\",\n                 model_type=\"random_forest\",\n                 opt_criteria=\"comprehensive\")",
   "execution_count": 48,
   "outputs": [
    {
     "execution_count": 48,
     "output_type": "execute_result",
     "data": {
      "text/plain": "{'bootstrap': [True],\n 'criterion': ['gini', 'entropy'],\n 'max_depth': [None],\n 'max_features': [22, 38, 45, 52, 59, 68, 90],\n 'min_samples_leaf': [1, 2, 5],\n 'min_samples_split': [2],\n 'n_estimators': [100, 250],\n 'n_jobs': -1,\n 'oob_score': [True]}"
     },
     "metadata": {}
    }
   ]
  },
  {
   "metadata": {
    "trusted": true,
    "collapsed": false
   },
   "cell_type": "code",
   "source": "get_defaults(8, task_type=\"classification\",\n                 model_type=\"random_forest\",\n                 opt_criteria=\"comprehensive\")",
   "execution_count": 49,
   "outputs": [
    {
     "execution_count": 49,
     "output_type": "execute_result",
     "data": {
      "text/plain": "{'bootstrap': [True],\n 'criterion': ['gini', 'entropy'],\n 'max_depth': [None],\n 'max_features': [2, 3, 4, 5, 6],\n 'min_samples_leaf': [1, 2, 5],\n 'min_samples_split': [2],\n 'n_estimators': [100, 250],\n 'n_jobs': -1,\n 'oob_score': [True]}"
     },
     "metadata": {}
    }
   ]
  },
  {
   "metadata": {
    "trusted": true,
    "collapsed": false
   },
   "cell_type": "code",
   "source": "get_defaults(100, task_type=\"regression\",\n                 model_type=\"random_forest\",\n                 opt_criteria=\"comprehensive\")",
   "execution_count": 51,
   "outputs": [
    {
     "execution_count": 51,
     "output_type": "execute_result",
     "data": {
      "text/plain": "{'bootstrap': [True],\n 'criterion': ['mse', 'mae'],\n 'max_depth': [None],\n 'max_features': [17, 28, 34, 40, 45, 51, 68],\n 'min_samples_leaf': [1, 2, 5],\n 'min_samples_split': [2],\n 'n_estimators': [100, 250],\n 'n_jobs': -1,\n 'oob_score': [True]}"
     },
     "metadata": {}
    }
   ]
  },
  {
   "metadata": {
    "trusted": true,
    "collapsed": true
   },
   "cell_type": "code",
   "source": "",
   "execution_count": null,
   "outputs": []
  }
 ],
 "metadata": {
  "language_info": {
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "mimetype": "text/x-python",
   "file_extension": ".py",
   "name": "python",
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "version": "3.5.2"
  },
  "kernelspec": {
   "name": "py3k",
   "display_name": "Python 3",
   "language": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}