{"cells":[
 {
  "cell_type": "code",
  "execution_count": null,
  "metadata": {
   "collapsed": false
  },
  "outputs": [],
  "source": "library(readr)\n\n# The competition datafiles are in the directory ../input\n# Read competition data files:\ntrain <- read_csv(\"../input/train.csv\")\ntest <- read_csv(\"../input/test.csv\")\n\n\nlibrary(h2o)\n\nlocalH2O = h2o.init(nthreads = -1) # use all CPUs (8 on my personal computer :3)\n\n\n#Now, I just convert the `train` and `test` sets into the h2o format and set up the model. \n## MNIST data as H2O\ntrain[,1] = as.factor(train[,1]) # convert digit labels to factor for classification\ntrain_h2o = as.h2o(train)\ntest_h2o = as.h2o(test)\n\ns <- proc.time()\n\nmodel =\n  h2o.randomForest(x = 2:785,  # column numbers for predictors\n                   y = 1,   # column number for label\n                   training_frame = train_h2o, # data in H2O format\n                   sample_rate = 0.8,\n                   mtries = -1, \n                   build_tree_one_node = FALSE, \n                   ntrees = 100,\n                   max_depth = 100, \n                   min_rows = 1, \n                   nbins = 30, \n                   nbins_cats = 1024, \n                   binomial_double_trees = TRUE,\n                   max_after_balance_size = 5, \n                   nfolds = 0,\n                   stopping_tolerance = 0.001)\n\nh2o.confusionMatrix(model)\n\n## print time elapsed\ns - proc.time()\n\n## classify test set\nh2o_y_test <- h2o.predict(model, test_h2o)\n\n## convert H2O format into data frame and  save as csv\ndf_y_test = as.data.frame(h2o_y_test)\ndf_y_test = data.frame(ImageId = seq(1,length(df_y_test$predict)), Label = df_y_test$predict)\nwrite.csv(df_y_test, file = \"submission-r-h2o-v2.csv\", row.names=F)"
 }
],"metadata":{"kernelspec":{"display_name":"R","language":"R","name":"ir"}}, "nbformat": 4, "nbformat_minor": 0}