trylib_examples.ipynb

{
 "cells": [
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "# Trylib example"
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "## RF"
  },
  {
   "metadata": {
    "collapsed": true,
    "trusted": false
   },
   "cell_type": "code",
   "source": "param = {'n_estimators': 100,\n             'criterion' : 'gini',\n             'max_depth':None,\n             'max_features' : 'auto',\n             'min_samples_leaf' : 1,\n         \n             'n_jobs':-1,\n}\n\nres = trylib(X,Y,'rf',param,one = False,skf_seed = 12, skf = 10, X_test= X_test).res",
   "execution_count": null,
   "outputs": []
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "# ET"
  },
  {
   "metadata": {
    "collapsed": true,
    "trusted": false
   },
   "cell_type": "code",
   "source": "param = {'n_estimators': 100,\n         'criterion' : 'gini',\n         'max_depth':None,\n         'max_features' : 'auto',\n         'min_samples_leaf' : 1,\n         \n         'n_jobs':-1,\n}\n\nres = trylib(X,Y,'et',param,one = False,skf_seed = 12, skf = 10, X_test= X_test).res",
   "execution_count": null,
   "outputs": []
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "## XGBoost"
  },
  {
   "metadata": {
    "collapsed": true,
    "trusted": false
   },
   "cell_type": "code",
   "source": "%%capture --no-stdout --no-display\nparam = {'num_round': 200,\n         \n             'seed' : 2441,\n             'max_depth':4,\n             'gamma': 0,\n             'eta':0.02,\n             'min_child_weight':1,\n             'silent':1, \n\n             'objective':'binary:logistic',\n             'scale_pos_weight':3,\n\n             'subsample' : 0.95,\n             'colsample_bytree' : 0.2,\n             #'base_score':Y.mean()\n           }\n\n# Objectives:\n\n#  -- 'objective':'binary:logistic',\n#  -- 'objective':'multi:softprob',\n#  -- 'objective':'rank:pairwise',\n#  -- 'objective':'reg:linear',\n\n# eval_metrics:\n\n#  -- 'eval_metric':'mlogloss',\n#  -- 'eval_metric':'auc',\n\n# num_class:\n\n#  -- 'num_class':9\n\nres = trylib(X,Y,'xg',param,one = False,skf_seed = 12, skf = 10, X_test= X_test).res",
   "execution_count": null,
   "outputs": []
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "## Linear SVC"
  },
  {
   "metadata": {
    "collapsed": true,
    "trusted": false
   },
   "cell_type": "code",
   "source": "param = {'C':1.0,\n         'penalty':'l2',\n         'loss':'squared_hinge',\n         'dual':True,\n         'multi_class':'ovr',\n         'class_weight': None,\n         'random_state':329,\n        }\n\nres = trylib(X,Y,'lsvc',param,one = False,skf_seed = 12, skf = 10, X_test= X_test).res",
   "execution_count": null,
   "outputs": []
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "## KNN"
  },
  {
   "metadata": {
    "collapsed": true,
    "trusted": false
   },
   "cell_type": "code",
   "source": "param = {'n_neighbors':5,\n         'weights':'uniform',\n         'metric':'minkowski',\n        }\n\n# weights:\n\n# -- 'weights':'uniform'\n# -- 'weights':'distance'\n# [callable]\n\n# metrics:\n\n# -- 'metric':'euclidean'\n# -- 'metric':'cityblock'\n# -- 'metric':'canberra'\n# -- 'metric':'braycurtis'\n# cosine ? \n\nres = trylib(X,Y,'knn',param,one = False,skf_seed = 12, skf = 10, X_test= X_test).res",
   "execution_count": null,
   "outputs": []
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "# Holdout validation"
  },
  {
   "metadata": {
    "collapsed": true,
    "trusted": false
   },
   "cell_type": "code",
   "source": "%%capture --no-stdout --no-display\nfrom sklearn.cross_validation import train_test_split\n\nfor i in range(4):\n    x_train, x_ho, y_train, y_ho = train_test_split(X, Y, \n                                                test_size=0.25, \n                                                random_state=i)\n#     x_train  = x_train.values\n#     x_hohout = x_hohout.values\n#     y_train  = y_train.values\n#     y_ho     = y_ho.values\n    \n    param = { }\n    \n    \n    res = trylib_r(x_train,y_train,'rf',param,one = True,skf = 5,X_test=x_ho,skf_seed = 12).res\n\n    print 'cv ',res['loss']\n    print 'ho ',RMSE(y_ho,res['pr_test'])    \n    \n    #print 'cv classes ', res['accuracy']\n    #print 'ho classes ', accuracy_score(y_ho,np.argmax(res['pr_test'],axis=1)) ",
   "execution_count": null,
   "outputs": []
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "# 2 level"
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "### 1"
  },
  {
   "metadata": {
    "collapsed": true,
    "trusted": false
   },
   "cell_type": "code",
   "source": "%%capture --no-stdout --no-display\n\n# X = \n# Y = \n# X_test =\n\n# ============ \ny_metas = []\ny_tests = []\n\nxg_param = {'num_round': 50,\n             'seed' : 21,\n             'max_depth':3,\n             'gamma': 0.01,\n             'eta':0.3,\n             'min_child_weight':1,\n             'silent':1, \n             'objective':'multi:softprob',\n             'num_round':100,\n             'num_class' : #9,\n             'subsample' : 0.8,\n             'colsample_bytree' : 0.2}\n\nclfs = [\n             ('xg', xg_param),\n             ('lsvc', {'C' : 100}),\n             ('knn', {'n_neighbors' : 10 , 'metric': 'canberra'}),\n             ('knn', {'n_neighbors' : 1 , 'metric': 'canberra'}),\n             ('nusvc',{'nu' : 0.007}),\n             ('rf',{'max_features' : 0.01,'n_estimators' : 4000}),\n             ('rf',{'max_features' : 0.3,'n_estimators' : 4000}),\n             ('et',{'max_features' : 0.1,'n_estimators' : 4000}),\n    ]\n\nfor c in clfs:\n    print ' ==== ', c[0], ' ====='\n        \n    res = trylib(X,Y,c[0],c[1],skf =10, X_test = X_test).res\n    print \" - %s: log_loss: %f, acc: %f\" %(c[0], res['loss'],res['accuracy'])\n    y_metas.append(res['pr'])\n    y_tests.append(res['pr_test'])\n\ny_meta = res['y']",
   "execution_count": null,
   "outputs": []
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "### 2"
  },
  {
   "metadata": {
    "collapsed": true,
    "trusted": false
   },
   "cell_type": "code",
   "source": "%%capture --no-stdout --no-display\n\nxg_param = {\n         'seed' : 21,\n         'max_depth':3,\n         'gamma': 0.01,\n         'eta':0.1,\n         'silent':1, \n         'objective':'multi:softprob',\n         'num_round':200,\n         'num_class' : 9,\n         'subsample' : 0.8,\n         'colsample_bytree' : 0.1}\nr = None\n\nmetas = np.hstack(y_metas)\ntest_metas = np.hstack(y_tests)\n\ny_out = []\nfor i in range(10):\n    xg_param['seed'] = (i+34)*15\n    \n    res = trylib(metas,y_meta,'xg',xg_param,skf = 10,one = False, X_test= test_metas).res\n    y_out.append(res['pr_test'])\n    \n    if r is None:\n        r = res['pr']\n    else:\n        r += res['pr']\n    print \"%d %f\" % (i, log_loss(y_val,r/(i+1)))\n\npr_test_final = np.mean(y_out,axis = 0)\npr_test_final.shape\n    \n#res2 = trylib(np.concatenate([d.values[res['val_index'],:],np.hstack(prs)],axis = 1),y_val,'et',{'max_features' : 0.1,'n_estimators' : 20000, 'compute_importances': True},skf = 10,one = False).res\n\n#prs.append(res['pr_val'])\n#y_val = res['y_val']",
   "execution_count": null,
   "outputs": []
  },
  {
   "metadata": {},
   "cell_type": "markdown",
   "source": "# CV template"
  },
  {
   "metadata": {
    "collapsed": true,
    "trusted": false
   },
   "cell_type": "code",
   "source": "skf = KFold(Y.shape[0], n_folds=10, shuffle=True, random_state=11)\n\nfor it, (train_index, test_index) in enumerate(skf):\n    X_train, X_val = X[train_index], X[test_index]\n    y_train, y_val = Y[train_index], Y[test_index]",
   "execution_count": null,
   "outputs": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "name": "python2",
   "display_name": "Python 2",
   "language": "python"
  },
  "language_info": {
   "mimetype": "text/x-python",
   "nbconvert_exporter": "python",
   "name": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.10",
   "file_extension": ".py",
   "codemirror_mode": {
    "version": 2,
    "name": "ipython"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}