tests/test_fakedata_01.r

In bert9bert/ParallelForest: Random Forest Classification with Parallel Computing

#------------------------------------------------------------------------------
#   Test Fortran wrappers.
#   Copyright (C) 2014
#------------------------------------------------------------------------------

library(ParallelForest)


PERFORM_TREE_TESTS = FALSE
PERFORM_FOREST_TESTS = TRUE


### SETUP ###

### PREPARE FAKE DATA DATASET ####
# load training data
data(easy_2var_data)
df = easy_2var_data

# create testing data
xnew = matrix(c(
        0.06,  0.03,
        0.05,  0.02,
        0.05,  0.05,
        0.01,  0.03,
        0.09,  0.04,
        0.05, -1000,
        0.05,  1000,
       -1000,  0.04,
        1000,  0.03
        ), 
    nrow=9, ncol=2, byrow=TRUE
    )
xnew = as.data.frame(xnew)
colnames(xnew) = c("X1","X2")

ynew = c(1, 0, 1, 0, 1, 0, 1, 1, 1)



### FITTING PARAMETERS ###
min_node_obs = 1
max_depth = 10



##### TREES #####
if(PERFORM_TREE_TESTS){
    ftree = grow.tree(Y~X1+X2, data=df, min_node_obs=min_node_obs, max_depth=max_depth)
    ftree_samepred = predict(ftree, df)
    ftree_ynewhat = predict(ftree, xnew)

    if(!all(df$Y==ftree_samepred)) {
       stop("Tree prediction on training data is different than training data.")
    }

    if(!all(ynew==ftree_ynewhat)) {
       stop("Tree prediction on new testing data is different than expected.")
    }
}

##### TESTS ON GROWING AND PREDICTING FORESTS #####
if(PERFORM_FOREST_TESTS){
    numsamps=150
    numvars=1
    numboots=20

    ### TEST 01 (ON EASY TO FIT DATA) ###
    fforest = grow.forest(Y~X1+X2, data=df, min_node_obs=min_node_obs, max_depth=max_depth,
        numsamps=numsamps, numvars=numvars, numboots=numboots,
        model=TRUE, x=TRUE, y=TRUE)
    fforest_samepred = predict(fforest, df)

    # test failure conditions
    if(sum(df$Y==fforest_samepred)/nrow(df) < 0.65) {
       stop("Forest prediction on training data performs worse than threshold.")
    }

    ### TEST 02 (NEW DATA) ###

    # same as above, except now with new data
    fforest_ynewhat = predict(fforest, xnew)

    # test failure conditions
    if(sum(ynew==fforest_ynewhat)/length(ynew) < 0.65) {
       stop("Forest prediction on testing data performs worse than threshold.")
    }

}