I'm trying to predict whether or not an airline will add a route to their existing network by looking at their previous additions and training the model on what the previous year looked like. I've used xgboost before and it worked fine, but I removed a few cities and now xgboost is just predicting everything to be 50:50.
trainm <- sparse.model.matrix(add ~. -1, data = train)
train_label <- train[, "add"]
train_matrix <- xgb.DMatrix(data = (trainm), label = train_label)
testm <- sparse.model.matrix(add~. -1, data = test)
test_label <- test[, "add"]
test_matrix <- xgb.DMatrix(data = (testm), label = test_label)
nc <- length(unique(train_label))
xgb_params <- list("objective" = "binary:logistic",
"eval_metric" = "error",
"scale_pos_weight" = weight)
watchlist <- list(train = train_matrix, test = test_matrix)
bst_model <- xgb.train(params = xgb_params,
nthreads = 2,
data = train_matrix,
nrounds = 10,
watchlist = watchlist,
booster = 'gbtree'
)
outputs:
[1] train-error:0.972469 test-error:0.972580
[2] train-error:0.972469 test-error:0.972580
[3] train-error:0.972469 test-error:0.972580
[4] train-error:0.972469 test-error:0.972580
[5] train-error:0.972469 test-error:0.972580
[6] train-error:0.972469 test-error:0.972580
[7] train-error:0.972469 test-error:0.972580
[8] train-error:0.972469 test-error:0.972580
[9] train-error:0.972469 test-error:0.972580
[10] train-error:0.972469 test-error:0.972580
It is weighted because it is very imbalanced (~36 negative for every 1 positive) just don't know why it's suddenly not working.
Edit. It fixed itself and I have no idea why.
Edit2. It did it again and I have no idea why.
Edit3. I fixed it. It has to do with NA values in certain columns.
Related
I am seeing a peculiar behaviour from the R Seurat package, when trying to subset objects to specific sets of cells.
So, say that I generate three sets of random cell names from a Seurat object using sample
library(Seurat)
set.seed(12345)
ten_cells_id <- sample(Cells(pbmc_small), 10)
other_ten_ids <- sample(Cells(pbmc_small), 10)
and_other_ten <- sample(Cells(pbmc_small), 10)
I can now subset the object using [] and print the cell tags
Cells(pbmc_small[, ten_cells_id], pt.size=3)
Cells(pbmc_small[, other_ten_ids], pt.size=3)
Cells(pbmc_small[, and_other_ten], pt.size=3)
No surprises here; it yields three different things as expected.
> Cells(pbmc_small[, ten_cells_id], pt.size=3)
[1] "CATGAGACACGGGA" "CGTAGCCTGTATGC" "ACTCGCACGAAAGT" "CTAGGTGATGGTTG" "TTACGTACGTTCAG" "CATGGCCTGTGCAT"
[7] "ACAGGTACTGGTGT" "AATGTTGACAGTCA" "GATAGAGAAGGGTG" "CATTACACCAACTG"
> Cells(pbmc_small[, other_ten_ids], pt.size=3)
[1] "GGCATATGCTTATC" "ACAGGTACTGGTGT" "CATCAGGATGCACA" "ATGCCAGAACGACT" "GAGTTGTGGTAGCT" "GGCATATGGGGAGT"
[7] "AGAGATGATCTCGC" "GAACCTGATGAACC" "GATATAACACGCAT" "CATGAGACACGGGA"
> Cells(pbmc_small[, and_other_ten], pt.size=3)
[1] "GGGTAACTCTAGTG" "TTTAGCTGTACTCT" "TACATCACGCTAAC" "CTAAACCTGTGCAT" "ATACCACTCTAAGC" "CATGCGCTAGTCAC"
[7] "GATAGAGAAGGGTG" "ATTACCTGCCTTAT" "GCGCATCTTGCTCC" "ACAGGTACTGGTGT"
However, if I do
cells1 <- pbmc_small[, sample(Cells(pbmc_small), 10)]
cells2 <- pbmc_small[, sample(Cells(pbmc_small), 10)]
cells3 <- pbmc_small[, sample(Cells(pbmc_small), 10)]
Cells(cells1)
Cells(cells2)
Cells(cells3)
I get three times the same thing
> Cells(cells1)
[1] "GATAGAGATCACGA" "GGCATATGCTTATC" "ATGCCAGAACGACT" "AGATATACCCGTAA" "TACAATGATGCTAG" "CATGAGACACGGGA"
[7] "GCACTAGACCTTTA" "CGTAGCCTGTATGC" "TTACCATGAATCGC" "ATAAGTTGGTACGT"
> Cells(cells2)
[1] "GATAGAGATCACGA" "GGCATATGCTTATC" "ATGCCAGAACGACT" "AGATATACCCGTAA" "TACAATGATGCTAG" "CATGAGACACGGGA"
[7] "GCACTAGACCTTTA" "CGTAGCCTGTATGC" "TTACCATGAATCGC" "ATAAGTTGGTACGT"
> Cells(cells3)
[1] "GATAGAGATCACGA" "GGCATATGCTTATC" "ATGCCAGAACGACT" "AGATATACCCGTAA" "TACAATGATGCTAG" "CATGAGACACGGGA"
[7] "GCACTAGACCTTTA" "CGTAGCCTGTATGC" "TTACCATGAATCGC" "ATAAGTTGGTACGT"
The values are always the same, independently of the seed I use!
I guess that R is somehow resetting the seed each time. This is not an issue with [] as:
a <- 1:100
a[sample(1:100, 10)]
a[sample(1:100, 10)]
a[sample(1:100, 10)]
Returns three different values.
The only thing I can think of is that something strange is happening because Seurat overloads []. Any ideas?
It looks like this is because [.Seurat() calls subset.Seurat(), which in turn calls WhichCells(). WhichCells() has a seed argument, which defaults to 1. You can override this by setting it to NULL, and thankfully this will also filter through if you pass it to [ like so:
library(Seurat)
#> Attaching SeuratObject
#> Attaching sp
set.seed(12345)
cells1 <- pbmc_small[, sample(Cells(pbmc_small), 10), seed = NULL]
cells2 <- pbmc_small[, sample(Cells(pbmc_small), 10), seed = NULL]
cells3 <- pbmc_small[, sample(Cells(pbmc_small), 10), seed = NULL]
Cells(cells1)
#> [1] "GATAGAGATCACGA" "GGCATATGCTTATC" "ATGCCAGAACGACT" "AGATATACCCGTAA"
#> [5] "TACAATGATGCTAG" "CATGAGACACGGGA" "GCACTAGACCTTTA" "CGTAGCCTGTATGC"
#> [9] "TTACCATGAATCGC" "ATAAGTTGGTACGT"
Cells(cells2)
#> [1] "GTCATACTTCGCCT" "TGGTATCTAAACAG" "ATCATCTGACACCA" "GTTGACGATATCGG"
#> [5] "GACGCTCTCTCTCG" "AGATATACCCGTAA" "CTTCATGACCGAAT" "CTAACGGAACCGAT"
#> [9] "TACTCTGAATCGAC" "GCGTAAACACGGTT"
Cells(cells3)
#> [1] "GTCATACTTCGCCT" "GCTCCATGAGAAGT" "ACAGGTACTGGTGT" "TACATCACGCTAAC"
#> [5] "CCATCCGATTCGCC" "GACGCTCTCTCTCG" "CTTCATGACCGAAT" "GCGTAAACACGGTT"
#> [9] "CATTACACCAACTG" "CTTGATTGATCTTC"
Created on 2022-10-17 with reprex v2.0.2
In my opinion this is quite poorly documented, and the behaviour is confusing enough to possibly justify a new issue at the surat-object GitHub.
I'm having trouble with the trafo function for SMOTE {smotefamily}'s K parameter. In particular, when the number of nearest neighbours K is greater than or equal to the sample size, an error is returned (warning("k should be less than sample size!")) and the tuning process is terminated.
The user cannot control K to be smaller than the sample size during the internal resampling process. This would have to be controlled internally so that if, for instance, trafo_K = 2 ^ K >= sample_size for some value of K, then, say, trafo_K = sample_size - 1.
I was wondering if there's a solution to this or if one is already on its way?
library("mlr3") # mlr3 base package
library("mlr3misc") # contains some helper functions
library("mlr3pipelines") # create ML pipelines
library("mlr3tuning") # tuning ML algorithms
library("mlr3learners") # additional ML algorithms
library("mlr3viz") # autoplot for benchmarks
library("paradox") # hyperparameter space
library("OpenML") # to obtain data sets
library("smotefamily") # SMOTE algorithm for imbalance correction
# get list of curated binary classification data sets (see https://arxiv.org/abs/1708.03731v2)
ds = listOMLDataSets(
number.of.classes = 2,
number.of.features = c(1, 100),
number.of.instances = c(5000, 10000)
)
# select imbalanced data sets (without categorical features as SMOTE cannot handle them)
ds = subset(ds, minority.class.size / number.of.instances < 0.2 &
number.of.symbolic.features == 1)
ds
d = getOMLDataSet(980)
d
# make sure target is a factor and create mlr3 tasks
data = as.data.frame(d)
data[[d$target.features]] = as.factor(data[[d$target.features]])
task = TaskClassif$new(
id = d$desc$name, backend = data,
target = d$target.features)
task
# Code above copied from https://mlr3gallery.mlr-org.com/posts/2020-03-30-imbalanced-data/
class_counts <- table(task$truth())
majority_to_minority_ratio <- class_counts[class_counts == max(class_counts)] /
class_counts[class_counts == min(class_counts)]
# Pipe operator for SMOTE
po_smote <- po("smote", dup_size = round(majority_to_minority_ratio))
# Random Forest learner
rf <- lrn("classif.ranger", predict_type = "prob")
# Pipeline of Random Forest learner with SMOTE
graph <- po_smote %>>%
po('learner', rf, id = 'rf')
graph$plot()
# Graph learner
rf_smote <- GraphLearner$new(graph, predict_type = 'prob')
rf_smote$predict_type <- 'prob'
# Parameter set in data table format
ps_table <- as.data.table(rf_smote$param_set)
View(ps_table[, 1:4])
# Define parameter search space for the SMOTE parameters
param_set <- ps_table$id %>%
lapply(
function(x) {
if (grepl('smote.', x)) {
if (grepl('.dup_size', x)) {
ParamInt$new(x, lower = 1, upper = round(majority_to_minority_ratio))
} else if (grepl('.K', x)) {
ParamInt$new(x, lower = 1, upper = round(majority_to_minority_ratio))
}
}
}
)
param_set <- Filter(Negate(is.null), param_set)
param_set <- ParamSet$new(param_set)
# Apply transformation function on SMOTE's K (= The number of nearest neighbors used for sampling new values. See SMOTE().)
param_set$trafo <- function(x, param_set) {
index <- which(grepl('.K', names(x)))
if (sum(index) != 0){
x[[index]] <- round(3 ^ x[[index]]) # Intentionally define a trafo that won't work
}
x
}
# Define and instantiate resampling strategy to be applied within pipeline
cv <- rsmp("cv", folds = 2)
cv$instantiate(task)
# Set up tuning instance
instance <- TuningInstance$new(
task = task,
learner = rf_smote,
resampling = cv,
measures = msr("classif.bbrier"),
param_set,
terminator = term("evals", n_evals = 3),
store_models = TRUE)
tuner <- TunerRandomSearch$new()
# Tune pipe learner to find optimal SMOTE parameter values
tuner$optimize(instance)
And here's what happens
INFO [11:00:14.904] Benchmark with 2 resampling iterations
INFO [11:00:14.919] Applying learner 'smote.rf' on task 'optdigits' (iter 2/2)
Error in get.knnx(data, query, k, algorithm) : ANN: ERROR------->
In addition: Warning message:
In get.knnx(data, query, k, algorithm) : k should be less than sample size!
Session info
R version 3.6.2 (2019-12-12)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 16299)
Matrix products: default
locale:
[1] LC_COLLATE=English_United Kingdom.1252 LC_CTYPE=English_United Kingdom.1252
[3] LC_MONETARY=English_United Kingdom.1252 LC_NUMERIC=C
[5] LC_TIME=English_United Kingdom.1252
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] smotefamily_1.3.1 OpenML_1.10 mlr3viz_0.1.1.9002
[4] mlr3tuning_0.1.2-9000 mlr3pipelines_0.1.2.9000 mlr3misc_0.2.0
[7] mlr3learners_0.2.0 mlr3filters_0.2.0.9000 mlr3_0.2.0-9000
[10] paradox_0.2.0 yardstick_0.0.5 rsample_0.0.5
[13] recipes_0.1.9 parsnip_0.0.5 infer_0.5.1
[16] dials_0.0.4 scales_1.1.0 broom_0.5.4
[19] tidymodels_0.0.3 reshape2_1.4.3 janitor_1.2.1
[22] data.table_1.12.8 forcats_0.4.0 stringr_1.4.0
[25] dplyr_0.8.4 purrr_0.3.3 readr_1.3.1
[28] tidyr_1.0.2 tibble_3.0.1 ggplot2_3.3.0
[31] tidyverse_1.3.0
loaded via a namespace (and not attached):
[1] utf8_1.1.4 tidyselect_1.0.0 lme4_1.1-21
[4] htmlwidgets_1.5.1 grid_3.6.2 ranger_0.12.1
[7] pROC_1.16.1 munsell_0.5.0 codetools_0.2-16
[10] bbotk_0.1 DT_0.12 future_1.17.0
[13] miniUI_0.1.1.1 withr_2.2.0 colorspace_1.4-1
[16] knitr_1.28 uuid_0.1-4 rstudioapi_0.10
[19] stats4_3.6.2 bayesplot_1.7.1 listenv_0.8.0
[22] rstan_2.19.2 lgr_0.3.4 DiceDesign_1.8-1
[25] vctrs_0.2.4 generics_0.0.2 ipred_0.9-9
[28] xfun_0.12 R6_2.4.1 markdown_1.1
[31] mlr3measures_0.1.3-9000 rstanarm_2.19.2 lhs_1.0.1
[34] assertthat_0.2.1 promises_1.1.0 nnet_7.3-12
[37] gtable_0.3.0 globals_0.12.5 processx_3.4.1
[40] timeDate_3043.102 rlang_0.4.5 workflows_0.1.1
[43] BBmisc_1.11 splines_3.6.2 checkmate_2.0.0
[46] inline_0.3.15 yaml_2.2.1 modelr_0.1.5
[49] tidytext_0.2.2 threejs_0.3.3 crosstalk_1.0.0
[52] backports_1.1.6 httpuv_1.5.2 rsconnect_0.8.16
[55] tokenizers_0.2.1 tools_3.6.2 lava_1.6.6
[58] ellipsis_0.3.0 ggridges_0.5.2 Rcpp_1.0.4.6
[61] plyr_1.8.5 base64enc_0.1-3 visNetwork_2.0.9
[64] ps_1.3.0 prettyunits_1.1.1 rpart_4.1-15
[67] zoo_1.8-7 haven_2.2.0 fs_1.3.1
[70] furrr_0.1.0 magrittr_1.5 colourpicker_1.0
[73] reprex_0.3.0 GPfit_1.0-8 SnowballC_0.6.0
[76] packrat_0.5.0 matrixStats_0.55.0 tidyposterior_0.0.2
[79] hms_0.5.3 shinyjs_1.1 mime_0.8
[82] xtable_1.8-4 XML_3.99-0.3 tidypredict_0.4.3
[85] shinystan_2.5.0 readxl_1.3.1 gridExtra_2.3
[88] rstantools_2.0.0 compiler_3.6.2 crayon_1.3.4
[91] minqa_1.2.4 StanHeaders_2.21.0-1 htmltools_0.4.0
[94] later_1.0.0 lubridate_1.7.4 DBI_1.1.0
[97] dbplyr_1.4.2 MASS_7.3-51.4 boot_1.3-23
[100] Matrix_1.2-18 cli_2.0.1 parallel_3.6.2
[103] gower_0.2.1 igraph_1.2.4.2 pkgconfig_2.0.3
[106] xml2_1.2.2 foreach_1.4.7 dygraphs_1.1.1.6
[109] prodlim_2019.11.13 farff_1.1 rvest_0.3.5
[112] snakecase_0.11.0 janeaustenr_0.1.5 callr_3.4.1
[115] digest_0.6.25 cellranger_1.1.0 curl_4.3
[118] shiny_1.4.0 gtools_3.8.1 nloptr_1.2.1
[121] lifecycle_0.2.0 nlme_3.1-142 jsonlite_1.6.1
[124] fansi_0.4.1 pillar_1.4.3 lattice_0.20-38
[127] loo_2.2.0 fastmap_1.0.1 httr_1.4.1
[130] pkgbuild_1.0.6 survival_3.1-8 glue_1.4.0
[133] xts_0.12-0 FNN_1.1.3 shinythemes_1.1.2
[136] iterators_1.0.12 class_7.3-15 stringi_1.4.4
[139] memoise_1.1.0 future.apply_1.5.0
Many thanks.
I've found a workaround.
As pointed out earlier, the problem is that SMOTE {smotefamily}'s K cannot be greater than or equal to the sample size.
I dag into the process and disovered that SMOTE {smotefamily} uses knearest {smotefamily}, which uses knnx.index {FNN}, which in turn uses get.knn {FNN},
which is what returns the error warning("k should be less than sample size!") that terminates the tuning process in mlr3.
Now, within SMOTE {smotefamily}, the three arguments for knearest {smotefamily} are P_set, P_set and K. From an mlr3 resampling perspective,
data frame P_set is a subset of the cross-validation fold of the training data, filtered to only contain the records of the minority class. The 'sample size' that
the error is referring to is the number of rows of P_set.
Thus, it becomes more likely that K >= nrow(P_set) as K increases via a trafo such as some_integer ^ K (e.g. 2 ^ K).
We need to ensure that K will never be greater than or equal to P_set.
Here's my proposed solution:
Define a variable cv_folds before defining the CV resampling strategy with rsmp().
Define the CV resampling strategy where folds = cv_folds in rsmp(), before defining the trafo.
Instantiate the CV. Now, the dataset is split into training and test/valitation data in each fold.
Find the minimum sample size of the minority class among all training data folds and set that as the threshold for K:
smote_k_thresh <- 1:cv_folds %>%
lapply(
function(x) {
index <- cv$train_set(x)
aux <- as.data.frame(task$data())[index, task$target_names]
aux <- min(table(aux))
}
) %>%
bind_cols %>%
min %>%
unique
Now define the trafo as follows:
param_set$trafo <- function(x, param_set) {
index <- which(grepl('.K', names(x)))
if (sum(index) != 0){
aux <- round(2 ^ x[[index]])
if (aux < smote_k_thresh) {
x[[index]] <- aux
} else {
x[[index]] <- sample(smote_k_thresh - 1, 1)
}
}
x
}
In other words, when the trafoed K remains smaller than the sample size, keep it. Otherwise, set its value to be any number between 1 and smote_k_thresh - 1.
Implementation
Original code slightly modified to accommodate proposed tweaks:
library("mlr3learners") # additional ML algorithms
library("mlr3viz") # autoplot for benchmarks
library("paradox") # hyperparameter space
library("OpenML") # to obtain data sets
library("smotefamily") # SMOTE algorithm for imbalance correction
# get list of curated binary classification data sets (see https://arxiv.org/abs/1708.03731v2)
ds = listOMLDataSets(
number.of.classes = 2,
number.of.features = c(1, 100),
number.of.instances = c(5000, 10000)
)
# select imbalanced data sets (without categorical features as SMOTE cannot handle them)
ds = subset(ds, minority.class.size / number.of.instances < 0.2 &
number.of.symbolic.features == 1)
ds
d = getOMLDataSet(980)
d
# make sure target is a factor and create mlr3 tasks
data = as.data.frame(d)
data[[d$target.features]] = as.factor(data[[d$target.features]])
task = TaskClassif$new(
id = d$desc$name, backend = data,
target = d$target.features)
task
# Code above copied from https://mlr3gallery.mlr-org.com/posts/2020-03-30-imbalanced-data/
class_counts <- table(task$truth())
majority_to_minority_ratio <- class_counts[class_counts == max(class_counts)] /
class_counts[class_counts == min(class_counts)]
# Pipe operator for SMOTE
po_smote <- po("smote", dup_size = round(majority_to_minority_ratio))
# Define and instantiate resampling strategy to be applied within pipeline
# Do that BEFORE defining the trafo
cv_folds <- 2
cv <- rsmp("cv", folds = cv_folds)
cv$instantiate(task)
# Calculate max possible value for k-nearest neighbours
smote_k_thresh <- 1:cv_folds %>%
lapply(
function(x) {
index <- cv$train_set(x)
aux <- as.data.frame(task$data())[index, task$target_names]
aux <- min(table(aux))
}
) %>%
bind_cols %>%
min %>%
unique
# Random Forest learner
rf <- lrn("classif.ranger", predict_type = "prob")
# Pipeline of Random Forest learner with SMOTE
graph <- po_smote %>>%
po('learner', rf, id = 'rf')
graph$plot()
# Graph learner
rf_smote <- GraphLearner$new(graph, predict_type = 'prob')
rf_smote$predict_type <- 'prob'
# Parameter set in data table format
ps_table <- as.data.table(rf_smote$param_set)
View(ps_table[, 1:4])
# Define parameter search space for the SMOTE parameters
param_set <- ps_table$id %>%
lapply(
function(x) {
if (grepl('smote.', x)) {
if (grepl('.dup_size', x)) {
ParamInt$new(x, lower = 1, upper = round(majority_to_minority_ratio))
} else if (grepl('.K', x)) {
ParamInt$new(x, lower = 1, upper = round(majority_to_minority_ratio))
}
}
}
)
param_set <- Filter(Negate(is.null), param_set)
param_set <- ParamSet$new(param_set)
# Apply transformation function on SMOTE's K while ensuring it never equals or exceeds the sample size
param_set$trafo <- function(x, param_set) {
index <- which(grepl('.K', names(x)))
if (sum(index) != 0){
aux <- round(5 ^ x[[index]]) # Try a large value here for the sake of the example
if (aux < smote_k_thresh) {
x[[index]] <- aux
} else {
x[[index]] <- sample(smote_k_thresh - 1, 1)
}
}
x
}
# Set up tuning instance
instance <- TuningInstance$new(
task = task,
learner = rf_smote,
resampling = cv,
measures = msr("classif.bbrier"),
param_set,
terminator = term("evals", n_evals = 10),
store_models = TRUE)
tuner <- TunerRandomSearch$new()
# Tune pipe learner to find optimal SMOTE parameter values
tuner$optimize(instance)
# Here are the original K values
instance$archive$data
# And here are their transformations
instance$archive$data$opt_x
Hi I am programming here in R, and I want to use the xgboost function for predicting a dummy variable.
That's the code:
library(xgboost)
library(Matrix)
mydata<-read.csv(file.choose(),header = TRUE,sep=",")
names(mydata)
[1] "Factor_Check" "Cor_Check" "Cor_Check4"
[4] "Cor_Check2" "n_tokens_title" "n_tokens_content"
[7] "n_unique_tokens" "n_non_stop_words" "n_non_stop_unique_tokens"
[10] "num_hrefs" "num_self_hrefs" "num_imgs"
[13] "num_videos" "average_token_length" "num_keywords"
[16] "data_channel_is_lifestyle" "data_channel_is_entertainment" "data_channel_is_bus"
[19] "data_channel_is_socmed" "data_channel_is_tech" "data_channel_is_world"
[22] "kw_min_min" "kw_max_min" "kw_avg_min"
[25] "kw_min_max" "kw_max_max" "kw_avg_max"
[28] "kw_min_avg" "kw_max_avg" "kw_avg_avg"
[31] "self_reference_min_shares" "self_reference_max_shares" "self_reference_avg_sharess"
[34] "weekday_is_monday" "weekday_is_tuesday" "weekday_is_wednesday"
[37] "weekday_is_thursday" "weekday_is_friday" "weekday_is_saturday"
[40] "weekday_is_sunday" "is_weekend" "LDA_00"
[43] "LDA_01" "LDA_02" "LDA_03"
[46] "LDA_04" "global_subjectivity" "global_sentiment_polarity"
[49] "global_rate_positive_words" "global_rate_negative_words" "rate_positive_words"
[52] "rate_negative_words" "avg_positive_polarity" "min_positive_polarity"
[55] "max_positive_polarity" "avg_negative_polarity" "min_negative_polarity"
[58] "max_negative_polarity" "title_subjectivity" "title_sentiment_polarity"
[61] "abs_title_subjectivity" "abs_title_sentiment_polarity" "TargetVarCont"
[64] "TargetVar1" "TargetVar2"
Factor Check is Factor the rest are numeric
output.var <- "TargetVar2"
vars.to.exclude <- c("Factor_Check","Cor_Check","Cor_Check4","Cor_Check2","TargetVar1", "TargetVarCont")
Building the model based on 80% of the data
train<-mydata[(1:round(nrow(mydata)*(0.8))),]
train<-train[,!(names(train) %in% vars.to.exclude)]
Train<- Matrix::sparse.model.matrix(~.-1 , data=train)
xgb <- xgboost(data = Train[,!(names(Train) %in% output.var)], label = Train[,output.var],max.depth = 2, eta = 1, nthread = 2, nround = 2, objective = "binary:logistic")
Train
Error: shinyjs: could not find the Shiny session object. This usually
happens when a shinyjs function is called from a context that wasn't
set up by a Shiny session.
Does anyone know why I am getting this error?
I'm trying to understand the intuition about what is going on in the xgb.dump of a binary classification with an interaction depth of 1. Specifically how the same split is used twiced in a row (f38 < 2.5) (code lines 2 and 6)
The resulting output looks like this:
xgb.dump(model_2,with.stats=T)
[1] "booster[0]"
[2] "0:[f38<2.5] yes=1,no=2,missing=1,gain=173.793,cover=6317"
[3] "1:leaf=-0.0366182,cover=3279.75"
[4] "2:leaf=-0.0466305,cover=3037.25"
[5] "booster[1]"
[6] "0:[f38<2.5] yes=1,no=2,missing=1,gain=163.887,cover=6314.25"
[7] "1:leaf=-0.035532,cover=3278.65"
[8] "2:leaf=-0.0452568,cover=3035.6"
Is the difference between the first use of f38 and the second use of f38 simply the residual fitting going on? At first it seemed weird to me, and trying to understand exactly what's going on here!
Thanks!
Is the difference between the first use of f38 and the second use of f38 simply the residual fitting going on?
most likely yes - its updating the gradient after the first round and finding the same feature with split point in your example
Here's a reproducible example.
Note how I lower the learning rate in the second example and its finds the same feature, same split point again for all three rounds. In the first example it uses different features in all 3 rounds.
require(xgboost)
data(agaricus.train, package='xgboost')
train <- agaricus.train
dtrain <- xgb.DMatrix(data = train$data, label=train$label)
#high learning rate, finds different first split feature (f55,f28,f66) in each tree
bst <- xgboost(data = train$data, label = train$label, max_depth = 2, eta = 1, nrounds = 3,nthread = 2, objective = "binary:logistic")
xgb.dump(model = bst)
# [1] "booster[0]" "0:[f28<-9.53674e-07] yes=1,no=2,missing=1"
# [3] "1:[f55<-9.53674e-07] yes=3,no=4,missing=3" "3:leaf=1.71218"
# [5] "4:leaf=-1.70044" "2:[f108<-9.53674e-07] yes=5,no=6,missing=5"
# [7] "5:leaf=-1.94071" "6:leaf=1.85965"
# [9] "booster[1]" "0:[f59<-9.53674e-07] yes=1,no=2,missing=1"
# [11] "1:[f28<-9.53674e-07] yes=3,no=4,missing=3" "3:leaf=0.784718"
# [13] "4:leaf=-0.96853" "2:leaf=-6.23624"
# [15] "booster[2]" "0:[f101<-9.53674e-07] yes=1,no=2,missing=1"
# [17] "1:[f66<-9.53674e-07] yes=3,no=4,missing=3" "3:leaf=0.658725"
# [19] "4:leaf=5.77229" "2:[f110<-9.53674e-07] yes=5,no=6,missing=5"
# [21] "5:leaf=-0.791407" "6:leaf=-9.42142"
## changed eta to lower learning rate, finds same feature(f55) in first split of each tree
bst2 <- xgboost(data = train$data, label = train$label, max_depth = 2, eta = .01, nrounds = 3,nthread = 2, objective = "binary:logistic")
xgb.dump(model = bst2)
# [1] "booster[0]" "0:[f28<-9.53674e-07] yes=1,no=2,missing=1"
# [3] "1:[f55<-9.53674e-07] yes=3,no=4,missing=3" "3:leaf=0.0171218"
# [5] "4:leaf=-0.0170044" "2:[f108<-9.53674e-07] yes=5,no=6,missing=5"
# [7] "5:leaf=-0.0194071" "6:leaf=0.0185965"
# [9] "booster[1]" "0:[f28<-9.53674e-07] yes=1,no=2,missing=1"
# [11] "1:[f55<-9.53674e-07] yes=3,no=4,missing=3" "3:leaf=0.016952"
# [13] "4:leaf=-0.0168371" "2:[f108<-9.53674e-07] yes=5,no=6,missing=5"
# [15] "5:leaf=-0.0192151" "6:leaf=0.0184251"
# [17] "booster[2]" "0:[f28<-9.53674e-07] yes=1,no=2,missing=1"
# [19] "1:[f55<-9.53674e-07] yes=3,no=4,missing=3" "3:leaf=0.0167863"
# [21] "4:leaf=-0.0166737" "2:[f108<-9.53674e-07] yes=5,no=6,missing=5"
# [23] "5:leaf=-0.0190286" "6:leaf=0.0182581"
Wondering how to generate time series and assign dates at the same time. I am trying this
series = as.ts(arima.sim(model = list(ar = c(0.12, -.36)),
n = 1990 - 1875, sd = sqrt(4)),
start = 1875, deltat = 1)
But this does not return a ts object that counts the years from 1875. By my reckoning this should work. Any advice appreciated.
You are correct. I re-typed vs cut/paste your code and it's both the start parameter and the use of as.ts vs just ts that's the issue:
asim <- arima.sim(model = list(ar = c(0.12, -.36)),
n = 1990 - 1875, sd = sqrt(4))
series <- ts(asim, start = c(1875, 1), deltat = 1)
print(series)
Time Series:
Start = 1875
End = 1989
Frequency = 1
[1] -1.22873543 -2.87876290 -3.00367322 -0.93120214 1.76854684 0.93874091 -2.32494289
[8] 1.14892019 1.87773156 1.48735536 -0.84149973 -3.69650397 1.20710878 2.14151424
[15] -2.58376182 -2.97501726 2.77019523 4.50829433 0.35603642 -1.95517140 -1.12792253
[22] 1.64063413 2.25654663 -0.51293345 1.07829896 -1.77134896 2.38908172 4.29362478
[29] -1.55577635 1.17953083 3.39823289 1.11846543 -0.92758706 -1.24158935 -2.39831233
[36] 4.24302415 2.93797283 -0.75916084 -0.66967525 2.85022663 -0.18190842 -5.39057660
[43] 0.08454559 2.01667062 -3.17054706 -3.77788365 0.19987174 2.87106608 -0.33844973
[50] 1.20917997 -1.00509230 3.23130604 5.80269444 3.33781468 2.67050526 1.85130774
[57] -0.46065144 -2.79539368 0.29784271 -4.51945793 0.61091013 2.56372897 -4.66101520
[64] 2.43024521 0.04428268 -1.19454953 -3.10583191 4.55208114 6.00037902 -3.32996632
[71] 2.22167610 1.07499343 1.89873604 2.04067084 -3.43648828 -0.53093294 0.66225057
[78] -2.30214366 0.78945348 0.35241170 -0.68250626 1.39801271 -1.01914282 -0.33615058
[85] 0.92311887 1.66289752 -0.83158693 -0.74454853 6.53884660 1.53567335 -2.16745416
[92] -0.01540633 -1.25032821 -0.02958796 3.18116493 -2.07512219 -1.40620668 -0.78869155
[99] 2.30251140 -2.23997817 0.34824690 4.81898402 -0.38751197 -5.74540148 -0.37754295
[106] 2.59869857 -1.90175430 0.37994317 -1.27326292 -3.96302760 -2.01928982 2.57643462
[113] 2.62600151 -4.20987173 0.46388883
Since asim is already a ts class object, as.ts is pulling the tsp attribute from it vs creating it from the input parameters. Using ts creates a new tsp attribute.