library(h2o)
h2o.init()
Connection successful!
R is connected to the H2O cluster:
H2O cluster uptime: 1 days 15 minutes
H2O cluster version: 3.14.0.3
H2O cluster version age: 12 days
H2O cluster name: H2O_started_from_R_r631758_lcl606
H2O cluster total nodes: 1
H2O cluster total memory: 3.16 GB
H2O cluster total cores: 8
H2O cluster allowed cores: 8
H2O cluster healthy: TRUE
H2O Connection ip: localhost
H2O Connection port: 54321
H2O Connection proxy: NA
H2O Internal Security: FALSE
H2O API Extensions: Algos, AutoML, Core V3, Core V4
R Version: R version 3.4.2 (2017-09-28)
h2o.removeAll()
[1] 0
load sample dataset
titanic<-h2o.importFile(path="http://s3.amazonaws.com/h2o-public-test-data/smalldata/gbm_test/titanic.csv")
|
| | 0%
|
|=====================================================================================| 100%
dim(titanic)
[1] 1309 14
head(titanic)
tail(titanic)
summary(titanic, exact_quantiles=T)
pclass survived name sex age sibsp
Min. :1.000 Min. :0.000 male :843 Min. : 0.1667 Min. :0.0000
1st Qu.:2.000 1st Qu.:0.000 female:466 1st Qu.:21.0000 1st Qu.:0.0000
Median :3.000 Median :0.000 Median :28.0000 Median :0.0000
Mean :2.295 Mean :0.382 Mean :29.8811 Mean :0.4989
3rd Qu.:3.000 3rd Qu.:1.000 3rd Qu.:39.0000 3rd Qu.:1.0000
Max. :3.000 Max. :1.000 Max. :80.0000 Max. :8.0000
NA's :263
parch ticket fare cabin embarked
Min. :0.000 Min. : 680 Min. : 0.000 C23 C25 C27 : 6 S :914
1st Qu.:0.000 1st Qu.: 19950 1st Qu.: 7.896 B57 B59 B63 B66: 5 C :270
Median :0.000 Median : 234604 Median : 14.454 G6 : 5 Q :123
Mean :0.385 Mean : 249039 Mean : 33.295 B96 B98 : 4 NA: 2
3rd Qu.:0.000 3rd Qu.: 347468 3rd Qu.: 31.275 C22 C26 : 4
Max. :9.000 Max. :3101298 Max. :512.329 C78 : 4
NA's :352 NA's :1 NA :1014
boat body home.dest
Min. : 1.000 Min. : 1.0 New York NY : 64
1st Qu.: 5.000 1st Qu.: 72.0 London : 14
Median :10.000 Median :155.0 Montreal PQ : 10
Mean : 9.405 Mean :160.8 Cornwall / Akron OH: 9
3rd Qu.:13.000 3rd Qu.:256.0 Paris France : 9
Max. :16.000 Max. :328.0 Philadelphia PA : 8
NA's :911 NA's :1188 NA :564
str(titanic)
Class 'H2OFrame' <environment: 0x0000000020321850>
- attr(*, "op")= chr "Parse"
- attr(*, "id")= chr "Key_Frame__http___s3_amazonaws_com_h2o_public_test_data_smalldata_gbm_test_titanic.hex_sid_888a_41"
- attr(*, "eval")= logi FALSE
- attr(*, "nrow")= int 1309
- attr(*, "ncol")= int 14
- attr(*, "types")=List of 14
..$ : chr "int"
..$ : chr "int"
..$ : chr "string"
..$ : chr "enum"
..$ : chr "real"
..$ : chr "int"
..$ : chr "int"
..$ : chr "int"
..$ : chr "real"
..$ : chr "enum"
..$ : chr "enum"
..$ : chr "int"
..$ : chr "int"
..$ : chr "enum"
- attr(*, "data")='data.frame': 10 obs. of 14 variables:
..$ pclass : num 1 1 1 1 1 1 1 1 1 1
..$ survived : num 1 1 0 0 0 1 1 0 1 0
..$ name : chr "Allen Miss. Elisabeth Walton" "Allison Master. Hudson Trevor" "Allison Miss. Helen Loraine" "Allison Mr. Hudson Joshua Creighton" ...
..$ sex : Factor w/ 2 levels "female","male": 1 2 1 2 1 2 1 2 1 2
..$ age : num 29 0.917 2 30 25 ...
..$ sibsp : num 0 1 1 1 1 0 1 0 2 0
..$ parch : num 0 2 2 2 2 0 0 0 0 0
..$ ticket : num 24160 113781 113781 113781 113781 ...
..$ fare : num 211 152 152 152 152 ...
..$ cabin : Factor w/ 186 levels "A10","A11","A14",..: 44 80 80 80 80 150 146 16 62 NA
..$ embarked : Factor w/ 3 levels "C","Q","S": 3 3 3 3 3 3 3 3 3 1
..$ boat : num 2 11 NaN NaN NaN 3 10 NaN NaN NaN
..$ body : num NaN NaN NaN 135 NaN NaN NaN NaN NaN 22
..$ home.dest: Factor w/ 369 levels "?Havana Cuba",..: 309 231 231 231 231 237 162 24 22 229
set response and predictors
titanic$survived<-as.factor(titanic$survived)
response<-"survived"
predictors<-paste( colnames(titanic[,-c(2,3)]), sep="")
predictors
[1] "pclass" "sex" "age" "sibsp" "parch" "ticket" "fare"
[8] "cabin" "embarked" "boat" "body" "home.dest"
split data
splits<-h2o.splitFrame(data=titanic, ratios=c(0.6,0.2), destination_frames = c("train.hex", "valid.hex", "test.hex"), seed=1234)
train<-splits[[1]]
valid<-splits[[2]]
test<-splits[[3]]
start with very basic model
model.gbm1<-h2o.gbm(x=predictors, y=response, training_frame = train)
|
| | 0%
|
|================= | 20%
|
|=====================================================================================| 100%
model.gbm1
Model Details:
==============
H2OBinomialModel: gbm
Model ID: GBM_model_R_1507141024239_18396
Model Summary:
number_of_trees number_of_internal_trees model_size_in_bytes min_depth max_depth mean_depth
1 50 50 22845 2 5 4.94000
min_leaves max_leaves mean_leaves
1 3 21 13.02000
H2OBinomialMetrics: gbm
** Reported on training data. **
MSE: 0.02096719
RMSE: 0.1448005
LogLoss: 0.08788473
Mean Per-Class Error: 0.02596078
AUC: 0.9960535
Gini: 0.992107
Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:
0 1 Error Rate
0 478 1 0.002088 =1/479
1 15 286 0.049834 =15/301
Totals 493 287 0.020513 =16/780
Maximum Metrics: Maximum metrics at their respective thresholds
metric threshold value idx
1 max f1 0.499288 0.972789 165
2 max f2 0.140574 0.970684 191
3 max f0point5 0.499288 0.986888 165
4 max accuracy 0.499288 0.979487 165
5 max precision 0.996316 1.000000 0
6 max recall 0.056272 1.000000 235
7 max specificity 0.996316 1.000000 0
8 max absolute_mcc 0.499288 0.957042 165
9 max min_per_class_accuracy 0.275850 0.966777 174
10 max mean_per_class_accuracy 0.499288 0.974039 165
Gains/Lift Table: Extract with `h2o.gainsLift(<model>, <data>)` or `h2o.gainsLift(<model>, valid=<T/F>, xval=<T/F>)`
get AUC
h2o.auc(h2o.performance(model.gbm1, newdata=valid))
[1] 0.9500141
trained with 80% of the data
model.gbm2<-h2o.gbm(x=predictors, y=response, training_frame = h2o.rbind(train, valid), nfolds=4, seed=100000)
|
| | 0%
|
|===================== | 25%
|
|=================================================================================== | 98%
|
|=====================================================================================| 100%
Show a detailed summary of the cross validation metrics
This gives you an idea of the variance between the folds
model.gbm2@model$cross_validation_metrics_summary
Cross-Validation Metrics Summary:
mean sd cv_1_valid cv_2_valid cv_3_valid
accuracy 0.93055546 0.008629293 0.9166667 0.92134833 0.94716984
auc 0.9485867 0.0090652285 0.9281929 0.9501641 0.95242697
err 0.069444545 0.008629293 0.083333336 0.07865169 0.05283019
err_count 18.25 2.0841665 21.0 21.0 14.0
f0point5 0.93016684 0.008899874 0.925 0.9259259 0.95137423
f1 0.9050068 0.014668015 0.87573963 0.8955224 0.92783505
f2 0.8818632 0.02512356 0.83146065 0.867052 0.9054326
lift_top_group 2.6012814 0.063610904 2.7391305 2.518868 2.6237624
logloss 0.23374325 0.026640536 0.28481033 0.25281268 0.20978114
max_per_class_error 0.13256821 0.032050658 0.19565217 0.1509434 0.10891089
mcc 0.8534678 0.018669134 0.82108814 0.83597547 0.88815725
mean_per_class_accuracy 0.91830176 0.012959264 0.8927989 0.90900034 0.9363982
mean_per_class_error 0.08169827 0.012959264 0.107201084 0.09099965 0.063601784
mse 0.0636791 0.008084995 0.07762853 0.072123334 0.054258037
precision 0.948204 0.013882402 0.96103895 0.94736844 0.9677419
r2 0.7304616 0.035632342 0.6651004 0.6987226 0.7699668
recall 0.8674318 0.032050658 0.8043478 0.8490566 0.8910891
rmse 0.25132284 0.016061164 0.27861896 0.26855788 0.23293355
specificity 0.9691717 0.010593492 0.98125 0.9689441 0.98170733
cv_4_valid
accuracy 0.93703705
auc 0.96356285
err 0.062962964
err_count 17.0
f0point5 0.9183673
f1 0.9209302
f2 0.92350745
lift_top_group 2.5233645
logloss 0.18756889
max_per_class_error 0.07476635
mcc 0.8686504
mean_per_class_accuracy 0.9350095
mean_per_class_error 0.06499054
mse 0.05070648
precision 0.9166667
r2 0.78805673
recall 0.92523366
rmse 0.22518098
specificity 0.9447853
get the cross-validated AUC by scoring the combined holdout predictions
h2o.auc(h2o.performance(model.gbm2))
[1] 0.9953856
h2o.auc(h2o.performance(model.gbm2,xval=TRUE))
[1] 0.9474834
fine tuning parameters
starttime<-Sys.time()
model.gbm3<-h2o.gbm(x=predictors, y=response, training_frame = train, validation_frame = valid, ntrees=10000, learn_rate=0.01, stopping_rounds=5, stopping_tolerance = 1e-4, stopping_metric="AUC",sample_rate=0.8, col_sample_rate = 0.8, seed=1234, score_tree_interval = 10)
|
| | 0%
|
|=====================================================================================| 100%
gbm3_time<-Sys.time()-starttime
print(paste("Took", round(gbm3_time, digits=2), units(gbm3_time), "to build GBM3 model."))
[1] "Took 1.54 secs to build GBM3 model."
get the AUC on the validation set
result does not get better
h2o.auc(h2o.performance(model.gbm3, valid=TRUE))
[1] 0.9424908
exploring more fine tuning parameters in depth
depth 10 is usually plenty of depth for most datasets, but you never know
hyper.params=list(max_depth=seq(1,29,2))
grid<-h2o.grid(hyper_params=hyper.params,
search_criteria =list(strategy="Cartesian"),
algorithm="gbm",
grid_id="depth_grid",
x=predictors,
y=response,
training_frame=train,
validation_frame=valid,
ntrees=10000,
learn_rate=0.05,
learn_rate_annealing=0.99,
sample_rate=0.8,
col_sample_rate=0.8,
seed=1234,
stopping_rounds=5,
stopping_tolerance=1e-4,
stopping_metric="AUC",
score_tree_interval=10
)
|
| | 0%
|
|====== | 7%
|
|=========== | 13%
|
|============ | 14%
|
|================= | 20%
|
|======================= | 27%
|
|============================ | 33%
|
|============================= | 34%
|
|================================== | 40%
|
|======================================== | 47%
|
|============================================= | 53%
|
|============================================= | 54%
|
|============================================== | 54%
|
|=================================================== | 60%
|
|========================================================= | 67%
|
|============================================================== | 73%
|
|============================================================== | 74%
|
|=============================================================== | 74%
|
|==================================================================== | 80%
|
|========================================================================== | 87%
|
|=============================================================================== | 93%
|
|=============================================================================== | 94%
|
|================================================================================ | 94%
|
|=====================================================================================| 100%
grid
H2O Grid Details
================
Grid ID: depth_grid
Used hyper parameters:
- max_depth
Number of models: 15
Number of failed models: 0
Hyper-Parameter Search Summary: ordered by increasing logloss
max_depth model_ids logloss
1 13 depth_grid_model_6 0.20109637892392768
2 9 depth_grid_model_4 0.20160720998146248
3 7 depth_grid_model_3 0.2024624226746261
4 5 depth_grid_model_2 0.2029008034398234
5 11 depth_grid_model_5 0.20343494648988525
6 29 depth_grid_model_14 0.20446595941168913
7 19 depth_grid_model_9 0.20446595941168913
8 21 depth_grid_model_10 0.20446595941168913
9 25 depth_grid_model_12 0.20446595941168913
10 27 depth_grid_model_13 0.20446595941168913
11 23 depth_grid_model_11 0.20446595941168913
12 17 depth_grid_model_8 0.2044659596864782
13 15 depth_grid_model_7 0.20463752833415869
14 3 depth_grid_model_1 0.20971798928576324
15 1 depth_grid_model_0 0.23401163708609685
sort the grid models by decreasing AUC
sortedGrid<-h2o.getGrid("depth_grid", sort_by="AUC", decreasing=TRUE)
sortedGrid
H2O Grid Details
================
Grid ID: depth_grid
Used hyper parameters:
- max_depth
Number of models: 15
Number of failed models: 0
Hyper-Parameter Search Summary: ordered by decreasing AUC
max_depth model_ids auc
1 13 depth_grid_model_6 0.9525218371372218
2 9 depth_grid_model_4 0.9519019442096365
3 11 depth_grid_model_5 0.9512820512820513
4 7 depth_grid_model_3 0.9512256973795435
5 5 depth_grid_model_2 0.9511411665257818
6 29 depth_grid_model_14 0.9505494505494505
7 17 depth_grid_model_8 0.9505494505494505
8 19 depth_grid_model_9 0.9505494505494505
9 21 depth_grid_model_10 0.9505494505494505
10 25 depth_grid_model_12 0.9505494505494505
11 27 depth_grid_model_13 0.9505494505494505
12 23 depth_grid_model_11 0.9505494505494505
13 15 depth_grid_model_7 0.9503240349394196
14 1 depth_grid_model_0 0.9462383770076077
15 3 depth_grid_model_1 0.9458157227387998
find the range of max_depth for the top 5 models
topDepths=sortedGrid@summary_table$max_depth[1:5]
minDepth=min(as.numeric(topDepths))
maxDepth=max(as.numeric(topDepths))
minDepth
[1] 5
maxDepth
[1] 13
select sequencially
hyper.params = list(
## restrict the search to the range of max_depth established above
max_depth = seq(minDepth,maxDepth,1),
## search a large space of row sampling rates per tree
sample_rate = seq(0.2,1,0.01),
## search a large space of column sampling rates per split
col_sample_rate = seq(0.2,1,0.01),
## search a large space of column sampling rates per tree
col_sample_rate_per_tree = seq(0.2,1,0.01),
## search a large space of how column sampling per split should change as a function of the depth of the split
col_sample_rate_change_per_level = seq(0.9,1.1,0.01),
## search a large space of the number of min rows in a terminal node
min_rows = 2^seq(0,log2(nrow(train))-1,1),
## search a large space of the number of bins for split-finding for continuous and integer columns
nbins = 2^seq(4,10,1),
## search a large space of the number of bins for split-finding for categorical columns
nbins_cats = 2^seq(4,12,1),
## search a few minimum required relative error improvement thresholds for a split to happen
min_split_improvement = c(0,1e-8,1e-6,1e-4),
## try all histogram types (QuantilesGlobal and RoundRobin are good for numeric columns with outliers)
histogram_type = c("UniformAdaptive","QuantilesGlobal","RoundRobin")
)
search.criteria=list(
strategy = "RandomDiscrete",
## limit the runtime to 60 minutes
max_runtime_secs = 3600,
## build no more than 100 models
max_models = 100,
## random number generator seed to make sampling of parameter combinations reproducible
seed = 1234,
## early stopping once the leaderboard of the top 5 models is converged to 0.1% relative difference
stopping_rounds = 5,
stopping_metric = "AUC",
stopping_tolerance = 1e-3
)
grid<-h2o.grid(
## hyper parameters
hyper_params = hyper.params,
## hyper-parameter search configuration (see above)
search_criteria = search.criteria,
## which algorithm to run
algorithm = "gbm",
## identifier for the grid, to later retrieve it
grid_id = "final_grid",
## standard model parameters
x = predictors,
y = response,
training_frame = train,
validation_frame = valid,
## more trees is better if the learning rate is small enough
## use "more than enough" trees - we have early stopping
ntrees = 10000,
## smaller learning rate is better
## since we have learning_rate_annealing, we can afford to start with a bigger learning rate
learn_rate = 0.05,
## learning rate annealing: learning_rate shrinks by 1% after every tree
## (use 1.00 to disable, but then lower the learning_rate)
learn_rate_annealing = 0.99,
## early stopping based on timeout (no model should take more than 1 hour - modify as needed)
max_runtime_secs = 600,
## early stopping once the validation AUC doesn't improve by at least 0.01% for 5 consecutive scoring events
stopping_rounds = 5, stopping_tolerance = 1e-4, stopping_metric = "AUC",
## score every 10 trees to make early stopping reproducible (it depends on the scoring interval)
score_tree_interval = 10,
## base random number generator seed for each model (automatically gets incremented internally for each model)
seed = 1234
)
|
| | 0%
|
|= | 1%
|
|=== | 4%
|
|==== | 5%
|
|===== | 6%
|
|======= | 8%
|
|======== | 9%
|
|========= | 10%
|
|========= | 11%
|
|========== | 12%
|
|=========== | 13%
|
|============ | 14%
|
|============= | 15%
|
|============== | 16%
|
|=============== | 18%
|
|================ | 19%
|
|================= | 20%
|
|================== | 21%
|
|=================== | 22%
|
|==================== | 23%
|
|====================== | 26%
|
|======================== | 28%
|
|========================= | 29%
|
|========================== | 30%
|
|=========================== | 32%
|
|============================ | 33%
|
|============================== | 35%
|
|=============================== | 36%
|
|=============================== | 37%
|
|================================ | 38%
|
|================================= | 39%
|
|================================== | 40%
|
|=================================== | 41%
|
|==================================== | 42%
|
|===================================== | 44%
|
|====================================== | 45%
|
|======================================= | 46%
|
|======================================== | 47%
|
|========================================== | 49%
|
|=========================================== | 50%
|
|============================================ | 52%
|
|============================================= | 53%
|
|============================================== | 54%
|
|=============================================== | 55%
|
|================================================ | 56%
|
|================================================ | 57%
|
|================================================= | 58%
|
|================================================== | 59%
|
|=================================================== | 60%
|
|==================================================== | 61%
|
|===================================================== | 62%
|
|====================================================== | 63%
|
|====================================================== | 64%
|
|======================================================= | 65%
|
|======================================================== | 66%
|
|========================================================= | 67%
|
|========================================================== | 68%
|
|=========================================================== | 69%
|
|============================================================= | 72%
|
|============================================================== | 73%
|
|=============================================================== | 74%
|
|================================================================ | 75%
|
|================================================================= | 76%
|
|================================================================= | 77%
|
|================================================================== | 78%
|
|=================================================================== | 79%
|
|==================================================================== | 80%
|
|===================================================================== | 81%
|
|====================================================================== | 82%
|
|======================================================================= | 83%
|
|======================================================================= | 84%
|
|========================================================================= | 86%
|
|========================================================================== | 87%
|
|=========================================================================== | 88%
|
|============================================================================= | 90%
|
|============================================================================= | 91%
|
|============================================================================== | 92%
|
|================================================================================ | 94%
|
|================================================================================== | 96%
|
|================================================================================== | 97%
|
|==================================================================================== | 99%
|
|=====================================================================================| 100%
Sort the grid models by AUC
sortedGrid <- h2o.getGrid("final_grid", sort_by = "auc", decreasing = TRUE)
sortedGrid
H2O Grid Details
================
Grid ID: final_grid
Used hyper parameters:
- col_sample_rate
- col_sample_rate_change_per_level
- col_sample_rate_per_tree
- histogram_type
- max_depth
- min_rows
- min_split_improvement
- nbins
- nbins_cats
- sample_rate
Number of models: 100
Number of failed models: 0
Hyper-Parameter Search Summary: ordered by decreasing auc
col_sample_rate col_sample_rate_change_per_level col_sample_rate_per_tree histogram_type
1 0.92 0.93 0.56 QuantilesGlobal
2 0.49 1.04 0.94 QuantilesGlobal
3 0.35 1.09 0.83 QuantilesGlobal
4 0.61 1.04 0.61 UniformAdaptive
5 0.81 0.94 0.89 QuantilesGlobal
max_depth min_rows min_split_improvement nbins nbins_cats sample_rate model_ids
1 6 4.0 0.0 128 128 0.93 final_grid_model_96
2 9 2.0 0.0 32 256 0.86 final_grid_model_68
3 5 4.0 1.0E-8 64 128 0.69 final_grid_model_38
4 11 1.0 1.0E-4 64 16 0.69 final_grid_model_81
5 8 16.0 1.0E-8 1024 32 0.71 final_grid_model_69
auc
1 0.974218089602705
2 0.9738799661876585
3 0.9698224852071006
4 0.9691462383770075
5 0.9684699915469147
---
col_sample_rate col_sample_rate_change_per_level col_sample_rate_per_tree histogram_type
95 0.7 1.08 0.99 QuantilesGlobal
96 0.5 1.03 0.45 RoundRobin
97 0.87 1.0 0.2 RoundRobin
98 0.24 1.08 0.3 UniformAdaptive
99 0.57 1.1 0.68 RoundRobin
100 0.96 0.94 0.62 QuantilesGlobal
max_depth min_rows min_split_improvement nbins nbins_cats sample_rate model_ids
95 7 256.0 1.0E-4 32 16 0.49 final_grid_model_86
96 13 256.0 1.0E-8 512 16 0.28 final_grid_model_58
97 12 256.0 1.0E-6 512 1024 0.97 final_grid_model_51
98 5 256.0 1.0E-4 32 64 0.97 final_grid_model_44
99 12 256.0 0.0 16 4096 0.58 final_grid_model_8
100 8 256.0 1.0E-6 64 4096 0.57 final_grid_model_95
auc
95 0.8014370245139476
96 0.7997464074387151
97 0.7965624119470274
98 0.7854888701042547
99 0.7836573682727528
100 0.7608058608058608
goe best 5 model AUC
for (i in 1:5) {
gbm<-h2o.getModel(sortedGrid@model_ids[[i]])
print(h2o.auc(h2o.performance(gbm,valid=TRUE)))
}
[1] 0.9742181
[1] 0.97388
[1] 0.9698225
[1] 0.9691462
[1] 0.96847
apply the best model to test data
gbm<-h2o.getModel(sortedGrid@model_ids[[1]])
print(h2o.auc(h2o.performance(gbm,newdata=test)))
[1] 0.9824898
gbm@parameters
$model_id
[1] "final_grid_model_96"
$training_frame
[1] "train.hex"
$validation_frame
[1] "valid.hex"
$score_tree_interval
[1] 10
$ntrees
[1] 10000
$max_depth
[1] 6
$min_rows
[1] 4
$nbins
[1] 128
$nbins_cats
[1] 128
$stopping_rounds
[1] 5
$stopping_metric
[1] "AUC"
$stopping_tolerance
[1] 1e-04
$max_runtime_secs
[1] 600
$seed
[1] 1234
$learn_rate
[1] 0.05
$learn_rate_annealing
[1] 0.99
$distribution
[1] "bernoulli"
$sample_rate
[1] 0.93
$col_sample_rate
[1] 0.92
$col_sample_rate_change_per_level
[1] 0.93
$col_sample_rate_per_tree
[1] 0.56
$min_split_improvement
[1] 0
$histogram_type
[1] "QuantilesGlobal"
$x
[1] "pclass" "sex" "age" "sibsp" "parch" "ticket" "fare"
[8] "cabin" "embarked" "boat" "body" "home.dest"
$y
[1] "survived"
Now we can confirm that these parameters are generally sound, by building a GBM model on the whole dataset (instead of the 60%) and using internal 5-fold cross-validation (re-using all other parameters including the seed):
model<-do.call(h2o.gbm,
{
p<-gbm@parameters
p$model_id=NULL ## do not overwrite the original grid model
p$training_frame=titanic ## use the full dataset
p$validation_frame=NULL ## no validation frame
p$nfolds=5 ## cross-validation
p
})
|
| | 0%
|
|= | 1%
|
|== | 2%
|
|============== | 17%
|
|============================ | 34%
|
|=========================================== | 50%
|
|=========================================== | 51%
|
|========================================================= | 67%
|
|======================================================================= | 84%
|
|=====================================================================================| 100%
model@model$cross_validation_metrics_summary
Cross-Validation Metrics Summary:
mean sd cv_1_valid cv_2_valid cv_3_valid
accuracy 0.94809973 0.003993344 0.9400749 0.94833946 0.9457364
auc 0.9743477 0.006040138 0.9674539 0.9610417 0.9794005
err 0.051900264 0.003993344 0.059925094 0.051660515 0.054263566
err_count 13.6 1.1489125 16.0 14.0 14.0
f0point5 0.95091534 0.011208471 0.9623016 0.95454544 0.944206
f1 0.9295287 0.0049488554 0.9238095 0.9230769 0.9263158
f2 0.9096094 0.013264459 0.88827837 0.89361703 0.90909094
lift_top_group 2.6258688 0.099894695 2.3839285 2.8229167 2.632653
logloss 0.19542515 0.015181999 0.20480314 0.23214972 0.19031271
max_per_class_error 0.102922216 0.019955961 0.13392857 0.125 0.10204082
mcc 0.89094704 0.0075886473 0.8800855 0.8873967 0.8845431
mean_per_class_accuracy 0.9385944 0.0053582946 0.9298099 0.9317857 0.93647957
mean_per_class_error 0.061405573 0.0053582946 0.070190094 0.06821428 0.06352041
mse 0.051655047 0.0043810816 0.05615356 0.061237488 0.049908444
precision 0.9660636 0.018879278 0.9897959 0.9767442 0.95652175
r2 0.7805782 0.019705383 0.7694049 0.73230106 0.788131
recall 0.8970778 0.019955961 0.8660714 0.875 0.8979592
rmse 0.22687589 0.009549358 0.23696741 0.2474621 0.22340198
specificity 0.98011106 0.011789808 0.9935484 0.9885714 0.975
cv_4_valid cv_5_valid
accuracy 0.9488189 0.95752895
auc 0.9819927 0.98184973
err 0.051181104 0.042471044
err_count 13.0 11.0
f0point5 0.92402464 0.96949893
f1 0.93264246 0.9417989
f2 0.9414226 0.91563785
lift_top_group 2.6736841 2.6161616
logloss 0.17594479 0.17391542
max_per_class_error 0.05263158 0.1010101
mcc 0.89166886 0.911041
mean_per_class_accuracy 0.948527 0.94636995
mean_per_class_error 0.05147302 0.05363005
mse 0.04610445 0.0448713
precision 0.9183673 0.98888886
r2 0.80308014 0.809974
recall 0.94736844 0.8989899
rmse 0.21471947 0.21182847
specificity 0.9496855 0.99375
to save time, let’s just scan through the top 5 models and cross-validate their parameters with nfolds=5 on the entire dataset:
for ( i in 1:5){
gbm<-h2o.getModel(sortedGrid@model_ids[[i]])
cvgbm<-do.call(h2o.gbm,
{
p <- gbm@parameters
p$model_id = NULL ## do not overwrite the original grid model
p$training_frame = titanic ## use the full dataset
p$validation_frame = NULL ## no validation frame
p$nfolds = 5 ## cross-validation
p
})
print(gbm@model_id)
print(cvgbm@model$cross_validation_metrics_summary[2,]) ## Pick out the "AUC" row
}
|
| | 0%
|
|= | 1%
|
|== | 2%
|
|============== | 17%
|
|============================= | 34%
|
|=========================================== | 50%
|
|=========================================== | 51%
|
|========================================================= | 67%
|
|======================================================================= | 83%
|
|======================================================================= | 84%
|
|=====================================================================================| 100%
[1] "final_grid_model_96"
Cross-Validation Metrics Summary:
mean sd cv_1_valid cv_2_valid cv_3_valid cv_4_valid cv_5_valid
auc 0.9743477 0.006040138 0.9674539 0.9610417 0.9794005 0.9819927 0.98184973
|
| | 0%
|
|= | 1%
|
|== | 2%
|
|============== | 17%
|
|=============== | 17%
|
|============================ | 33%
|
|============================= | 34%
|
|=========================================== | 50%
|
|========================================================= | 67%
|
|======================================================================= | 83%
|
|======================================================================= | 84%
|
|=====================================================================================| 100%
[1] "final_grid_model_68"
Cross-Validation Metrics Summary:
mean sd cv_1_valid cv_2_valid cv_3_valid cv_4_valid cv_5_valid
auc 0.9741264 0.0058573517 0.96854836 0.9610417 0.97665817 0.9807349 0.983649
|
| | 0%
|
|= | 1%
|
|== | 2%
|
|=== | 3%
|
|============== | 17%
|
|=============== | 17%
|
|============================= | 34%
|
|=========================================== | 50%
|
|=========================================== | 51%
|
|========================================================= | 67%
|
|======================================================================= | 84%
|
|=====================================================================================| 100%
[1] "final_grid_model_38"
Cross-Validation Metrics Summary:
mean sd cv_1_valid cv_2_valid cv_3_valid cv_4_valid cv_5_valid
auc 0.9724971 0.0057914597 0.9625576 0.9624107 0.97927296 0.97835153 0.9798927
|
| | 0%
|
|= | 1%
|
|== | 2%
|
|============== | 17%
|
|============================ | 33%
|
|=========================================== | 50%
|
|=========================================== | 51%
|
|========================================================= | 67%
|
|======================================================================= | 83%
|
|======================================================================= | 84%
|
|=====================================================================================| 100%
[1] "final_grid_model_81"
Cross-Validation Metrics Summary:
mean sd cv_1_valid cv_2_valid cv_3_valid cv_4_valid cv_5_valid
auc 0.9689918 0.006921219 0.96209675 0.9530357 0.97786987 0.9755048 0.976452
|
| | 0%
|
|= | 1%
|
|== | 2%
|
|============== | 17%
|
|============================ | 33%
|
|============================ | 34%
|
|============================= | 34%
|
|=========================================== | 50%
|
|=========================================== | 51%
|
|========================================================= | 67%
|
|======================================================================= | 83%
|
|======================================================================= | 84%
|
|=====================================================================================| 100%
[1] "final_grid_model_69"
Cross-Validation Metrics Summary:
mean sd cv_1_valid cv_2_valid cv_3_valid cv_4_valid cv_5_valid
auc 0.97103506 0.0053187287 0.96313363 0.96068454 0.97589284 0.9776233 0.9778409
apply the best model to test data
gbm<-h2o.getModel(sortedGrid@model_ids[[1]])
preds<-h2o.predict(gbm,test)
|
| | 0%
|
|=====================================================================================| 100%
head(preds)
gbm@model$validation_metrics@metrics$max_criteria_and_metric_scores
Maximum Metrics: Maximum metrics at their respective thresholds
metric threshold value idx
1 max f1 0.415247 0.929293 92
2 max f2 0.207864 0.924528 109
3 max f0point5 0.523349 0.970149 90
4 max accuracy 0.523349 0.948905 90
5 max precision 0.990276 1.000000 0
6 max recall 0.057998 1.000000 205
7 max specificity 0.990276 1.000000 0
8 max absolute_mcc 0.523349 0.894631 90
9 max min_per_class_accuracy 0.207864 0.928994 109
10 max mean_per_class_accuracy 0.415247 0.935137 92
While this is running, we can actually look at the model.
To do this we simply need a new connection to H2O.
This R console will run the model, so we need either another R console
or the web browser (or python, etc.).
In the demo, we will use Flow in our web browser
And the focus will be to look at model performance, since we are using R to
control H2O. So we can simply type in:
getModel “final_grid_model_96”
---
title: "Fine tuning GBM"
output: html_notebook
---


```{r}
library(h2o)
h2o.init()
h2o.removeAll()
```

#load sample dataset
```{r}
titanic<-h2o.importFile(path="http://s3.amazonaws.com/h2o-public-test-data/smalldata/gbm_test/titanic.csv")
dim(titanic)
head(titanic)
tail(titanic)
summary(titanic, exact_quantiles=T)
str(titanic)
```


#set response and predictors
```{r}
titanic$survived<-as.factor(titanic$survived)
response<-"survived"
predictors<-paste( colnames(titanic[,-c(2,3)]), sep="")
predictors
```

#split data
```{r}
splits<-h2o.splitFrame(data=titanic, ratios=c(0.6,0.2), destination_frames = c("train.hex", "valid.hex", "test.hex"), seed=1234)
train<-splits[[1]]
valid<-splits[[2]]
test<-splits[[3]]
```

#start with very basic model
```{r}
model.gbm1<-h2o.gbm(x=predictors, y=response, training_frame = train)
 model.gbm1
```
#get AUC
```{r}
h2o.auc(h2o.performance(model.gbm1, newdata=valid))
```
#trained with 80% of the data
```{r}
model.gbm2<-h2o.gbm(x=predictors, y=response, training_frame = h2o.rbind(train, valid), nfolds=4, seed=100000)
```

## Show a detailed summary of the cross validation metrics
## This gives you an idea of the variance between the folds
```{r}
model.gbm2@model$cross_validation_metrics_summary
```

#get the cross-validated AUC by scoring the combined holdout predictions
```{r}
h2o.auc(h2o.performance(model.gbm2))
h2o.auc(h2o.performance(model.gbm2,xval=TRUE))
```

#fine tuning parameters
```{r}
starttime<-Sys.time()
model.gbm3<-h2o.gbm(x=predictors, y=response, training_frame = train, validation_frame = valid, ntrees=10000, learn_rate=0.01, stopping_rounds=5, stopping_tolerance = 1e-4, stopping_metric="AUC",sample_rate=0.8, col_sample_rate = 0.8, seed=1234, score_tree_interval = 10)
gbm3_time<-Sys.time()-starttime
print(paste("Took", round(gbm3_time, digits=2), units(gbm3_time), "to build GBM3 model."))
```
#get the AUC on the validation set
#result does not get better
```{r}
h2o.auc(h2o.performance(model.gbm3, valid=TRUE))
```

#exploring more fine tuning parameters in depth
#depth 10 is usually plenty of depth for most datasets, but you never know
```{r}
hyper.params=list(max_depth=seq(1,29,2))
grid<-h2o.grid(hyper_params=hyper.params,
               search_criteria =list(strategy="Cartesian"),
               algorithm="gbm",
               grid_id="depth_grid",
               x=predictors,
               y=response,
               training_frame=train,
               validation_frame=valid,
               ntrees=10000,
               learn_rate=0.05,
               learn_rate_annealing=0.99, 
               sample_rate=0.8,
               col_sample_rate=0.8,
               seed=1234,
               stopping_rounds=5,
               stopping_tolerance=1e-4,
               stopping_metric="AUC",
               score_tree_interval=10
               )
grid
```

#sort the grid models by decreasing AUC
```{r}
sortedGrid<-h2o.getGrid("depth_grid", sort_by="AUC", decreasing=TRUE)
sortedGrid
```
#find the range of max_depth for the top 5 models
```{r}
topDepths=sortedGrid@summary_table$max_depth[1:5]
minDepth=min(as.numeric(topDepths))
maxDepth=max(as.numeric(topDepths))
minDepth
maxDepth
```

#select sequencially
```{r}
hyper.params = list( 
  ## restrict the search to the range of max_depth established above
  max_depth = seq(minDepth,maxDepth,1),                                      
  
  ## search a large space of row sampling rates per tree
  sample_rate = seq(0.2,1,0.01),                                             
  
  ## search a large space of column sampling rates per split
  col_sample_rate = seq(0.2,1,0.01),                                         
  
  ## search a large space of column sampling rates per tree
  col_sample_rate_per_tree = seq(0.2,1,0.01),                                
  
  ## search a large space of how column sampling per split should change as a function of the depth of the split
  col_sample_rate_change_per_level = seq(0.9,1.1,0.01),                      
  
  ## search a large space of the number of min rows in a terminal node
  min_rows = 2^seq(0,log2(nrow(train))-1,1),                                 
  
  ## search a large space of the number of bins for split-finding for continuous and integer columns
  nbins = 2^seq(4,10,1),                                                     
  
  ## search a large space of the number of bins for split-finding for categorical columns
  nbins_cats = 2^seq(4,12,1),                                                
  
  ## search a few minimum required relative error improvement thresholds for a split to happen
  min_split_improvement = c(0,1e-8,1e-6,1e-4),                               
  
  ## try all histogram types (QuantilesGlobal and RoundRobin are good for numeric columns with outliers)
  histogram_type = c("UniformAdaptive","QuantilesGlobal","RoundRobin")       
)
search.criteria=list(
  strategy = "RandomDiscrete",      
  
  ## limit the runtime to 60 minutes
  max_runtime_secs = 3600,         
  
  ## build no more than 100 models
  max_models = 100,                  
  
  ## random number generator seed to make sampling of parameter combinations reproducible
  seed = 1234,                        
  
  ## early stopping once the leaderboard of the top 5 models is converged to 0.1% relative difference
  stopping_rounds = 5,                
  stopping_metric = "AUC",
  stopping_tolerance = 1e-3
)
grid<-h2o.grid(
  ## hyper parameters
  hyper_params = hyper.params,
  
  ## hyper-parameter search configuration (see above)
  search_criteria = search.criteria,
  
  ## which algorithm to run
  algorithm = "gbm",
  
  ## identifier for the grid, to later retrieve it
  grid_id = "final_grid", 
  
  ## standard model parameters
  x = predictors, 
  y = response, 
  training_frame = train, 
  validation_frame = valid,
  
  ## more trees is better if the learning rate is small enough
  ## use "more than enough" trees - we have early stopping
  ntrees = 10000,                                                            
  
  ## smaller learning rate is better
  ## since we have learning_rate_annealing, we can afford to start with a bigger learning rate
  learn_rate = 0.05,                                                         
  
  ## learning rate annealing: learning_rate shrinks by 1% after every tree 
  ## (use 1.00 to disable, but then lower the learning_rate)
  learn_rate_annealing = 0.99,                                               
  
  ## early stopping based on timeout (no model should take more than 1 hour - modify as needed)
  max_runtime_secs = 600,                                                 
  
  ## early stopping once the validation AUC doesn't improve by at least 0.01% for 5 consecutive scoring events
  stopping_rounds = 5, stopping_tolerance = 1e-4, stopping_metric = "AUC", 
  
  ## score every 10 trees to make early stopping reproducible (it depends on the scoring interval)
  score_tree_interval = 10,                                                
  
  ## base random number generator seed for each model (automatically gets incremented internally for each model)
  seed = 1234                
)

```

## Sort the grid models by AUC
```{r}
sortedGrid <- h2o.getGrid("final_grid", sort_by = "auc", decreasing = TRUE)    
sortedGrid
```

#goe best 5 model AUC
```{r}
for (i in 1:5) {
  gbm<-h2o.getModel(sortedGrid@model_ids[[i]])
  print(h2o.auc(h2o.performance(gbm,valid=TRUE)))
}
```

#apply the best model to test data
```{r}
gbm<-h2o.getModel(sortedGrid@model_ids[[1]])
print(h2o.auc(h2o.performance(gbm,newdata=test)))
gbm@parameters
```

#Now we can confirm that these parameters are generally sound, by building a GBM model on the whole dataset (instead of the 60%) and using internal 5-fold cross-validation (re-using all other parameters including the seed):

```{r}
model<-do.call(h2o.gbm, 
               {
                 p<-gbm@parameters 
                 p$model_id=NULL          ## do not overwrite the original grid model
                 p$training_frame=titanic ## use the full dataset
                 p$validation_frame=NULL  ## no validation frame
                 p$nfolds=5               ## cross-validation
                 p
               })
model@model$cross_validation_metrics_summary
```

#to save time, let's just scan through the top 5 models and cross-validate their parameters with nfolds=5 on the entire dataset:
```{r}
for ( i in 1:5){
  gbm<-h2o.getModel(sortedGrid@model_ids[[i]])
          cvgbm<-do.call(h2o.gbm,
                                   {
          p <- gbm@parameters
          p$model_id = NULL          ## do not overwrite the original grid model
          p$training_frame = titanic     ## use the full dataset
          p$validation_frame = NULL  ## no validation frame
          p$nfolds = 5               ## cross-validation
          p  
                                     
                                   })
                    
    print(gbm@model_id)
  print(cvgbm@model$cross_validation_metrics_summary[2,]) ## Pick out the "AUC" row
                    
}
```

#apply the best model to test data
```{r}
gbm<-h2o.getModel(sortedGrid@model_ids[[1]])
preds<-h2o.predict(gbm,test)
head(preds)
gbm@model$validation_metrics@metrics$max_criteria_and_metric_scores
```

#### While this is running, we can actually look at the model.
#### To do this we simply need a new connection to H2O.
#### This R console will run the model, so we need either another R console
####   or the web browser (or python, etc.).
#### In the demo, we will use Flow in our web browser
####  http://localhost:54321
#### And the focus will be to look at model performance, since we are using R to 
####  control H2O. So we can simply type in:
####  getModel "final_grid_model_96"


#https://github.com/h2oai/h2o-3/blob/master/h2o-docs/src/product/tutorials/gbm/gbmTuning.Rmd
