Icon

kn_​automl_​h2o_​classification_​r

H2O.ai AutoML (wrapped with R) in KNIME for classification problems

H2O.ai AutoML (wrapped with R) in KNIME for classification problems - a powerful auto-machine-learning framework (https://hub.knime.com/mlauber71/spaces/Public/latest/automl/)
v 1.75

It features various models like Random Forest or XGBoost along with Deep Learning. It has warppers for R and Python but also could be used from KNIME. The results will be written to a folder and the models will be stored in MOJO format to be used in KNIME (as well as on a Big Data cluster via Sparkling Water). One major parameter to set is the running time the model has to test various models and do some hyper parameter optimization as well. The best model of each round is stored and some graphics are produced to see the results.

To run this workflow you have to install R and H2O.ai with several packages. Please refer to the green box on the right.

The results may be used also on Big Data clusters with the help of H2O.ai Sparkling Water (https://hub.knime.com/mlauber71/spaces/Public/latest/kn_example_h2o_sparkling_water)

# Run AutoML for 60 seconds or# 300 = 5 min, 600 = 10 min, 900 = 15 min, 1800 = 30 min, 3600 = 1 hour, # 7200 = 2 hours# 14400 = 4 hours# 16200 = 4.5 hours# 18000 = 5 Stunden# 21600 = 6 hours# 25200 = 7 hours# 28800 = 8 hours# 36000 = 10 hours H2O.ai AutoML (wrapped with R) in KNIME for classification problems - a powerful auto-machine-learning framework (https://hub.knime.com/mlauber71/spaces/Public/latest/automl/)v 1.75It features various models like Random Forest or XGBoost along with Deep Learning. It has warppers for R and Python but also could be used from KNIME. The results will be written to a folder and the models will be stored in MOJO format to be used in KNIME (as well as on a Big Data cluster via Sparkling Water). One major parameter to set is the running time the model has to test various models and do some hyper parameteroptimization as well. The best model of each round is stored and some graphics are produced to see the results.To run this workflow you have to install R and H2O.ai with several packages. Please refer to the green box on the right.The results may be used also on Big Data clusters with the help of H2O.ai Sparkling Water (https://hub.knime.com/mlauber71/spaces/Public/latest/kn_example_h2o_sparkling_water) which output is there to be interpretedmodel are stored in the folder /model/<full model name>/<model name>.zip-> as MOJO model format (certain model types cannot be stored and reused - so they are excluded as of now)/model/<full model name>/<model name>-> genuine H2O model stored in a folder (can be reused from H2O itself - also could store the Stacked andEnsemble models as well as XGBoost)/model/validate/h2o_list_of_models.csv -> list of all leading model from the runs with their RMSE (among other things) --- individual model results/model/validate/H2O_AutoML_Classification_yyyymmdd_hhmmh.txt-> capture of a print command describing the winning modelmodel_table_H2O_AutoML_Classification_yyyymmdd_hhmmh.table-> a KNIME table with a collection of parameters and information about the modelH2O_AutoML_Classification_yyyymmdd_hhmmh_-> CSVfiles containing important information among these: - _leaderboard = the list of all tested models in the run - _model_summary = the characteristic of the winning model (depth - _variable_importances = !!! check if the variable importance does make senseH2O_AutoML_Classification_yyyymmdd_hhmmh.xlsx-> an Excel file containing important information among these:- model_cutoff = check the best cutoff for your business case (depending on the number of different scores you willget the scores rounded to 0.1 or 0.10) (max_cohens_kappa = based on best Cohen's Cappa, max_f_measure = based on best F1 score)- model_cutoff_overview = compact overview of cut-off results---- 4 graphics for each model to have visual support when interpreting the results (needs R)model_graph_H2O_AutoML_Class_yyyymmdd_hhmmh_cutoff.png-> a graphic illustrating the consequences of two possible cut-offs (with statistics). Please note depending on yourbusiness needs you might choose completely different onesmodel_graph_H2O_AutoML_Class_yyyymmdd_hhmmh_roc.png-> a classic ROC (receiver operating characteristic) curve with statisticsmodel_graph_H2O_AutoML_Class_yyyymmdd_hhmmh_lift.png-> a classic lift curve with statistics. Illustrating how the TOP 10% of your score are doing compared to the restmodel_graph_H2O_AutoML_Class_yyyymmdd_hhmmh_ks.png-> two curves illustrating the Kolmogorov-Smirnov Goodness-of-Fit Test Subfolders to check/data/ contains the original data/model/contains the stored models in MOJO and H2O format/model/validate/contains the validations and graphics/script/'pure' R code (if you do not wish to use the KNIME wrapper) kn_automl_h2o_classification_r.Ra PDF with further informations about the methods usedH2O.ai AutoML in KNIME for classification problems.pdf input_table_2 <- knime.in# H2O.ai AutoML applied# library(arrow)library(h2o)# set the Java Home path to the Java version distributed with KNIMEvar_java_home = knime.flow.in[["java.home"]]Sys.setenv("JAVA_HOME" = c(var_java_home))# ---> EDIT:# workpath_r <- "/Users/m_lauber/Dropbox/knime-workspace/kaggle/census_income_bigdata_40/data/"# workpath_r <- "C:\\Users\\A9699459\\knime-workspace\\kaggle\\census_income_bigdata_40\\data"# workspace_name <- paste0(workpath_r, "workspace_410.RData")# setwd(workpath_r) # Set work directory# http://biostat.mc.vanderbilt.edu/wiki/pub/Main/ColeBeck/datestimes.pdfvar_timestamp_day <- format(as.POSIXct(Sys.Date(), format = "%Y-%m-%d"), format = "%Y%m%d")# print("var_timestamp_day: ", var_timestamp_day)var_timestamp_time <- format(as.POSIXct(Sys.time(), format = "%Y-%m-%d %H:%M"), format = "%H%M")# print("var_timestamp_time: ", var_timestamp_time)# _edit: if you want to have another model namevar_model_name <- "H2O_AutoML_Classification"var_model_name_full <- paste0(var_model_name , "_" , var_timestamp_day , "_" , var_timestamp_time)# print("var_model_name_full: ", var_model_name_full)# import the train and test file from local drive# v_data_path <- c("/Users/m_lauber/Dropbox/knime-workspace/hub/automl/kn_automl_h2o_classification_r/data/")v_data_path <- c(knime.flow.in[["var_path_data"]])# input_table_1 <- as.data.frame(read_parquet(paste0(v_data_path,"train.parquet")))# input_table_2 <- as.data.frame(read_parquet(paste0(v_data_path,"test.parquet")))# _edit:# list variable names to be removed / not being used in model# this is here to demonstrate how this would be donev_remove_variables <- c("Date1", "Location")# grab the columns from the 1st dataframe# x <- colnames(input_table_1)# name the target variablesy = 'Target'v_dropList <- c(v_remove_variables, y)# temporary file that only has the df_x_vars <- head(input_table_1[, !colnames(input_table_1) %in% v_dropList])x<- colnames(df_x_vars)# save and load the working environment# save.image(workspace_name)# load(workspace_name)# h2o.shutdown(prompt = FALSE)h2o.init()# https://forum.knime.com/t/python-script-and-h2o-data-frames-error-under-windows/21099/4?u=mlauber71h2o.no_progress()h2o.clusterStatus()# When launching nodes, we recommend allocating a total of four times the memory of your data.# h2o.init(ice_root=c(workpath_mac))# h2o.init(nthreads=20, max_mem_size="12G")train <- as.h2o(input_table_1)valid <- as.h2o(input_table_2)# For binary classification, response should be a factortrain[,y] <- as.factor(train[,y])valid[,y] <- as.factor(valid[,y])# https://www.h2o.ai/blog/h2o-release-3-24-yates/# http://docs.h2o.ai/h2o/latest-stable/h2o-docs/automl.html# https://www.rdocumentation.org/packages/h2o/versions/3.16.0.2/topics/h2o.automl# https://dzone.com/articles/a-trial-run-with-h2o-automl-automatic-machine-lear# "GLM", "DeepLearning", "DRF", "GBM", "StackedEnsemble", "XGBoost"# "GLM", "DeepLearning", "DRF", "GBM", "StackedEnsemble"# "DeepLearning", "StackedEnsemble", "XGBoost"# http://docs.h2o.ai/h2o/latest-stable/h2o-py/docs/_modules/h2o/automl/autoh2o.html# you could exclude algorithms as they might not be suitable # to be used in Big Data environments# exclude_algos =["GBM", "GLM", "DeepLearning", "DRF", "StackedEnsemble", "XGBoost"]# For binomial classification choose between "AUC", "logloss", "mean_per_class_error", "RMSE", "MSE". # For multinomial classification choose between "mean_per_class_error", "logloss", "RMSE", "MSE". # For regression choose between "deviance", "RMSE", "MSE", "MAE", "RMLSE".# Run AutoML for 60 seconds or# 300 = 5 min, 600 = 10 min, 900 = 15 min, 1800 = 30 min, 3600 = 1 hour, # 7200 = 2 hours# 14400 = 4 hours# 16200 = 4.5 hours# 18000 = 5 Stunden# 21600 = 6 hours# 25200 = 7 hours# 28800 = 8 hours# 36000 = 10 hours# http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/sort_metric.html# http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/exclude_algos.html# get the maximum runtime from the KNIME workflowmax_runtime_secs_opts <- knime.flow.in[["v_runtime_automl"]]aml <- h2o.automl(x = x, y = y, training_frame = train, validation_frame = valid, balance_classes=FALSE, max_runtime_secs=max_runtime_secs_opts, seed =1234, sort_metric ="AUC", stopping_metric ="AUC", stopping_tolerance =0.01, stopping_rounds =25, project_name =var_model_name_full , # exclude_algos =c("GBM", "GLM", "DeepLearning", "DRF", "StackedEnsemble") #, # exclude_algos =c("DRF", "GLM") exclude_algos =c("DeepLearning", "StackedEnsemble", "XGBoost"))# bring the leaderboard to a filelb <- aml@leaderboard# lb <- h2o.get_leaderboard(aml, extra_columns = "ALL")# print(lb)tb_leaderboard <- as.data.frame(lb)# var_selected_model <- "GBM_1_AutoML_20191214_123545"var_selected_model <- aml@leader@model_id# print("var_selected_model :", var_selected_model)# get the extracted modelextracted_model <- h2o.getModel(var_selected_model)# extract important tables from model to store latertb_variable_importances <- h2o.varimp(extracted_model)tb_model_summary <- as.data.frame(extracted_model@allparameters)# print(tb_variable_importances)# var_path_model <- c("/Users/m_lauber/Dropbox/knime-workspace/hub/automl/kn_automl_h2o_classification_r/model/")var_path_model <- c(knime.flow.in[["var_path_model"]])# var_path_validate <- c("/Users/m_lauber/Dropbox/knime-workspace/hub/automl/kn_automl_h2o_classification_r/model/validate/")var_path_validate <- c(knime.flow.in[["var_path_validate"]])v_csv_file_variable_importance <- paste0(var_path_validate, var_model_name_full, "_variable_importance.csv")write.table(tb_variable_importances, v_csv_file_variable_importance , sep="|")var_model_name_path <- paste0(var_path_model, var_model_name_full)# just save as 'normal' H2O Model - you can easily use later again with R# h2o.saveModel(aml@leader, path=c(var_loc_mojo_model), force=TRUE)h2o.saveModel(extracted_model, path=c(var_model_name_path), force=TRUE)# save the best model as MOJO model (seems to be a problem bringing it back from MOJO)# it seems you can save an ensemble but not bring it back, but anywayh2o.saveMojo(extracted_model, path = c(var_model_name_path), force=TRUE)var_model_file_name <- paste0(var_path_model, var_model_name_full, knime.flow.in[["file.separator"]] , var_selected_model)var_mojo_file_name <- paste0(var_path_model, var_model_name_full, knime.flow.in[["file.separator"]] , var_selected_model, ".zip")# saved_h2o_model<- h2o.loadModel(var_model_file_name)saved_mojo_model <- h2o.import_mojo(var_mojo_file_name)# https://www.kaggle.com/sramirez/deep-learning-in-r-with-h2otest_prediction <- as.data.frame(h2o.predict(saved_mojo_model, valid, type = 'raw'))output_table_2 <- cbind(input_table_2, test_prediction)v_csv_file_leaderboard <- paste0(var_path_validate, var_model_name_full, "_leaderboard.csv")write.table(tb_leaderboard, v_csv_file_leaderboard , sep="|")v_csv_file_model_summary <- paste0(var_path_validate, var_model_name_full, "_model_summary.csv")write.table(tb_model_summary, v_csv_file_model_summary , sep="|")var_txt_summary = paste0(var_path_validate, var_model_name_full, ".txt")sink(var_txt_summary)print(extracted_model)sink()# ------------------------------------------------------------------------# export the necessary variables as flow variables to KNIME# https://forum.knime.com/t/creating-flow-variables-in-r-scripting-nodes/5701/2?u=mlauber71# https://forum.knime.com/t/creating-flow-variables-with-r-snippet-works-but/9007?u=mlauber71knime.flow.out <- list(var_timestamp_day = var_timestamp_day, var_timestamp_time = var_timestamp_time, var_model_name = var_model_name, var_model_name_full = var_model_name_full, x = paste(c(x), collapse=', ' ), y = y, max_runtime_secs_opts = max_runtime_secs_opts, var_path_model = var_path_model, var_mojo_file_name = var_mojo_file_name, v_csv_file_variable_importance = v_csv_file_variable_importance, v_csv_file_leaderboard = v_csv_file_leaderboard, v_csv_file_model_summary = v_csv_file_model_summary, var_txt_summary = var_txt_summary)# shut down the H2O cluster in the end# h2o.shutdown(prompt = FALSE) # make sure you have R and the necessary R packages installed, also check aout the pdf in /script/conda install -c conda-forge r-h2o# Install R alongside KNIME on Windows and MacOS# https://forum.knime.com/t/install-r-alongside-knime-on-windows-and-macos/13287# R and Rtools# https://forum.knime.com/t/how-to-import-tables-from-docx-documents-via-r-snippet/19284/10# RServe 1.8.6 on MacOSX# https://forum.knime.com/t/installing-rserve-1-8-6-on-macos-10-15-catalina/20909/6?u=mlauber71library(h2o)# if you wish to use the 'pure' R code and import the data with parquetlibrary(arrow) additional R packages needed:ggplot2, lift, reshape2If you use the R wrapper you will need the h2o package and the arrow package if you plan on using the pure R script in the /script/subfolderhttp://docs.h2o.ai/h2o/latest-stable/h2o-docs/downloading.html Inspect the models so far and see to results. This will also give you a quick idea where you stand and what you would be able to achieve.Along with all parameters to load the respective model. Propagate R environmentfor KNIME on MacOS withMinicondaconfigure how to handle the environmentdefault = just check the namestrain.tableinput_table_2and mainR codewrapperinput_table_1scored outputedit: v_runtime_automlset the maximum runtime ofH2O.ai AutoML in SECONDSextract parametersfrom Rwhich have been usedto calculate the modelexclude pathsh2o_list_of_models.csvappend if CSV already exists to collect allmodel runsRead the MOJOmodelyou could check out this nodeScore the test tableyou might also use a third table to validatethat has not been used developing themodelvar_model_name_fullsolutionto string^(.*submission|solution).*$h2o_list_of_models.csvAUC DESCkeep best modelRead the MOJOmodelRead Variableimportancetest.tablePropagate R environmentfor KNIME on Windows withMinicondaconfigure how to handle the environmentdefault = just check the namesbinary classification modelswith Rset the parameters!java.homejava.homecreate initial Test andTraining dataCensus incomeclassification knime_r_environment Table Reader Add Table To R Table to R R to Table Integer Input collect meta data Merge Variables RowID Variable toTable Row Column Filter String to Path(Variable) CSV Writer Column Resorter H2O MOJO Reader Binary ClassificationInspector H2O MOJO Predictor(Classification) ROC Curve (local) ConstantValue Column Number To String Column Filter Column Rename CSV Reader String to Path(Variable) Sorter Row Filter Column Filter Table Rowto Variable H2O MOJO Reader CSV Reader Table Reader knime_r_environment_windows Model QualityClassification - Graphics Extract SystemProperties Table Columnto Variable Joiner Test Training # Run AutoML for 60 seconds or# 300 = 5 min, 600 = 10 min, 900 = 15 min, 1800 = 30 min, 3600 = 1 hour, # 7200 = 2 hours# 14400 = 4 hours# 16200 = 4.5 hours# 18000 = 5 Stunden# 21600 = 6 hours# 25200 = 7 hours# 28800 = 8 hours# 36000 = 10 hours H2O.ai AutoML (wrapped with R) in KNIME for classification problems - a powerful auto-machine-learning framework (https://hub.knime.com/mlauber71/spaces/Public/latest/automl/)v 1.75It features various models like Random Forest or XGBoost along with Deep Learning. It has warppers for R and Python but also could be used from KNIME. The results will be written to a folder and the models will be stored in MOJO format to be used in KNIME (as well as on a Big Data cluster via Sparkling Water). One major parameter to set is the running time the model has to test various models and do some hyper parameteroptimization as well. The best model of each round is stored and some graphics are produced to see the results.To run this workflow you have to install R and H2O.ai with several packages. Please refer to the green box on the right.The results may be used also on Big Data clusters with the help of H2O.ai Sparkling Water (https://hub.knime.com/mlauber71/spaces/Public/latest/kn_example_h2o_sparkling_water) which output is there to be interpretedmodel are stored in the folder /model/<full model name>/<model name>.zip-> as MOJO model format (certain model types cannot be stored and reused - so they are excluded as of now)/model/<full model name>/<model name>-> genuine H2O model stored in a folder (can be reused from H2O itself - also could store the Stacked andEnsemble models as well as XGBoost)/model/validate/h2o_list_of_models.csv -> list of all leading model from the runs with their RMSE (among other things) --- individual model results/model/validate/H2O_AutoML_Classification_yyyymmdd_hhmmh.txt-> capture of a print command describing the winning modelmodel_table_H2O_AutoML_Classification_yyyymmdd_hhmmh.table-> a KNIME table with a collection of parameters and information about the modelH2O_AutoML_Classification_yyyymmdd_hhmmh_-> CSVfiles containing important information among these: - _leaderboard = the list of all tested models in the run - _model_summary = the characteristic of the winning model (depth - _variable_importances = !!! check if the variable importance does make senseH2O_AutoML_Classification_yyyymmdd_hhmmh.xlsx-> an Excel file containing important information among these:- model_cutoff = check the best cutoff for your business case (depending on the number of different scores you willget the scores rounded to 0.1 or 0.10) (max_cohens_kappa = based on best Cohen's Cappa, max_f_measure = based on best F1 score)- model_cutoff_overview = compact overview of cut-off results---- 4 graphics for each model to have visual support when interpreting the results (needs R)model_graph_H2O_AutoML_Class_yyyymmdd_hhmmh_cutoff.png-> a graphic illustrating the consequences of two possible cut-offs (with statistics). Please note depending on yourbusiness needs you might choose completely different onesmodel_graph_H2O_AutoML_Class_yyyymmdd_hhmmh_roc.png-> a classic ROC (receiver operating characteristic) curve with statisticsmodel_graph_H2O_AutoML_Class_yyyymmdd_hhmmh_lift.png-> a classic lift curve with statistics. Illustrating how the TOP 10% of your score are doing compared to the restmodel_graph_H2O_AutoML_Class_yyyymmdd_hhmmh_ks.png-> two curves illustrating the Kolmogorov-Smirnov Goodness-of-Fit Test Subfolders to check/data/ contains the original data/model/contains the stored models in MOJO and H2O format/model/validate/contains the validations and graphics/script/'pure' R code (if you do not wish to use the KNIME wrapper)kn_automl_h2o_classification_r.Ra PDF with further informations about the methods usedH2O.ai AutoML in KNIME for classification problems.pdf input_table_2 <- knime.in# H2O.ai AutoML applied# library(arrow)library(h2o)# set the Java Home path to the Java version distributed with KNIMEvar_java_home = knime.flow.in[["java.home"]]Sys.setenv("JAVA_HOME" = c(var_java_home))# ---> EDIT:# workpath_r <- "/Users/m_lauber/Dropbox/knime-workspace/kaggle/census_income_bigdata_40/data/"# workpath_r <- "C:\\Users\\A9699459\\knime-workspace\\kaggle\\census_income_bigdata_40\\data"# workspace_name <- paste0(workpath_r, "workspace_410.RData")# setwd(workpath_r) # Set work directory# http://biostat.mc.vanderbilt.edu/wiki/pub/Main/ColeBeck/datestimes.pdfvar_timestamp_day <- format(as.POSIXct(Sys.Date(), format = "%Y-%m-%d"), format = "%Y%m%d")# print("var_timestamp_day: ", var_timestamp_day)var_timestamp_time <- format(as.POSIXct(Sys.time(), format = "%Y-%m-%d %H:%M"), format = "%H%M")# print("var_timestamp_time: ", var_timestamp_time)# _edit: if you want to have another model namevar_model_name <- "H2O_AutoML_Classification"var_model_name_full <- paste0(var_model_name , "_" , var_timestamp_day , "_" , var_timestamp_time)# print("var_model_name_full: ", var_model_name_full)# import the train and test file from local drive# v_data_path <- c("/Users/m_lauber/Dropbox/knime-workspace/hub/automl/kn_automl_h2o_classification_r/data/")v_data_path <- c(knime.flow.in[["var_path_data"]])# input_table_1 <- as.data.frame(read_parquet(paste0(v_data_path,"train.parquet")))# input_table_2 <- as.data.frame(read_parquet(paste0(v_data_path,"test.parquet")))# _edit:# list variable names to be removed / not being used in model# this is here to demonstrate how this would be donev_remove_variables <- c("Date1", "Location")# grab the columns from the 1st dataframe# x <- colnames(input_table_1)# name the target variablesy = 'Target'v_dropList <- c(v_remove_variables, y)# temporary file that only has the df_x_vars <- head(input_table_1[, !colnames(input_table_1) %in% v_dropList])x<- colnames(df_x_vars)# save and load the working environment# save.image(workspace_name)# load(workspace_name)# h2o.shutdown(prompt = FALSE)h2o.init()# https://forum.knime.com/t/python-script-and-h2o-data-frames-error-under-windows/21099/4?u=mlauber71h2o.no_progress()h2o.clusterStatus()# When launching nodes, we recommend allocating a total of four times the memory of your data.# h2o.init(ice_root=c(workpath_mac))# h2o.init(nthreads=20, max_mem_size="12G")train <- as.h2o(input_table_1)valid <- as.h2o(input_table_2)# For binary classification, response should be a factortrain[,y] <- as.factor(train[,y])valid[,y] <- as.factor(valid[,y])# https://www.h2o.ai/blog/h2o-release-3-24-yates/# http://docs.h2o.ai/h2o/latest-stable/h2o-docs/automl.html# https://www.rdocumentation.org/packages/h2o/versions/3.16.0.2/topics/h2o.automl# https://dzone.com/articles/a-trial-run-with-h2o-automl-automatic-machine-lear# "GLM", "DeepLearning", "DRF", "GBM", "StackedEnsemble", "XGBoost"# "GLM", "DeepLearning", "DRF", "GBM", "StackedEnsemble"# "DeepLearning", "StackedEnsemble", "XGBoost"# http://docs.h2o.ai/h2o/latest-stable/h2o-py/docs/_modules/h2o/automl/autoh2o.html# you could exclude algorithms as they might not be suitable # to be used in Big Data environments# exclude_algos =["GBM", "GLM", "DeepLearning", "DRF", "StackedEnsemble", "XGBoost"]# For binomial classification choose between "AUC", "logloss", "mean_per_class_error", "RMSE", "MSE". # For multinomial classification choose between "mean_per_class_error", "logloss", "RMSE", "MSE". # For regression choose between "deviance", "RMSE", "MSE", "MAE", "RMLSE".# Run AutoML for 60 seconds or# 300 = 5 min, 600 = 10 min, 900 = 15 min, 1800 = 30 min, 3600 = 1 hour, # 7200 = 2 hours# 14400 = 4 hours# 16200 = 4.5 hours# 18000 = 5 Stunden# 21600 = 6 hours# 25200 = 7 hours# 28800 = 8 hours# 36000 = 10 hours# http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/sort_metric.html# http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/exclude_algos.html# get the maximum runtime from the KNIME workflowmax_runtime_secs_opts <- knime.flow.in[["v_runtime_automl"]]aml <- h2o.automl(x = x, y = y, training_frame = train, validation_frame = valid, balance_classes=FALSE, max_runtime_secs=max_runtime_secs_opts, seed =1234, sort_metric ="AUC", stopping_metric ="AUC", stopping_tolerance =0.01, stopping_rounds =25, project_name =var_model_name_full , # exclude_algos =c("GBM", "GLM", "DeepLearning", "DRF", "StackedEnsemble") #, # exclude_algos =c("DRF", "GLM") exclude_algos =c("DeepLearning", "StackedEnsemble", "XGBoost"))# bring the leaderboard to a filelb <- aml@leaderboard# lb <- h2o.get_leaderboard(aml, extra_columns = "ALL")# print(lb)tb_leaderboard <- as.data.frame(lb)# var_selected_model <- "GBM_1_AutoML_20191214_123545"var_selected_model <- aml@leader@model_id# print("var_selected_model :", var_selected_model)# get the extracted modelextracted_model <- h2o.getModel(var_selected_model)# extract important tables from model to store latertb_variable_importances <- h2o.varimp(extracted_model)tb_model_summary <- as.data.frame(extracted_model@allparameters)# print(tb_variable_importances)# var_path_model <- c("/Users/m_lauber/Dropbox/knime-workspace/hub/automl/kn_automl_h2o_classification_r/model/")var_path_model <- c(knime.flow.in[["var_path_model"]])# var_path_validate <- c("/Users/m_lauber/Dropbox/knime-workspace/hub/automl/kn_automl_h2o_classification_r/model/validate/")var_path_validate <- c(knime.flow.in[["var_path_validate"]])v_csv_file_variable_importance <- paste0(var_path_validate, var_model_name_full, "_variable_importance.csv")write.table(tb_variable_importances, v_csv_file_variable_importance , sep="|")var_model_name_path <- paste0(var_path_model, var_model_name_full)# just save as 'normal' H2O Model - you can easily use later again with R# h2o.saveModel(aml@leader, path=c(var_loc_mojo_model), force=TRUE)h2o.saveModel(extracted_model, path=c(var_model_name_path), force=TRUE)# save the best model as MOJO model (seems to be a problem bringing it back from MOJO)# it seems you can save an ensemble but not bring it back, but anywayh2o.saveMojo(extracted_model, path = c(var_model_name_path), force=TRUE)var_model_file_name <- paste0(var_path_model, var_model_name_full, knime.flow.in[["file.separator"]] , var_selected_model)var_mojo_file_name <- paste0(var_path_model, var_model_name_full, knime.flow.in[["file.separator"]] , var_selected_model, ".zip")# saved_h2o_model<- h2o.loadModel(var_model_file_name)saved_mojo_model <- h2o.import_mojo(var_mojo_file_name)# https://www.kaggle.com/sramirez/deep-learning-in-r-with-h2otest_prediction <- as.data.frame(h2o.predict(saved_mojo_model, valid, type = 'raw'))output_table_2 <- cbind(input_table_2, test_prediction)v_csv_file_leaderboard <- paste0(var_path_validate, var_model_name_full, "_leaderboard.csv")write.table(tb_leaderboard, v_csv_file_leaderboard , sep="|")v_csv_file_model_summary <- paste0(var_path_validate, var_model_name_full, "_model_summary.csv")write.table(tb_model_summary, v_csv_file_model_summary , sep="|")var_txt_summary = paste0(var_path_validate, var_model_name_full, ".txt")sink(var_txt_summary)print(extracted_model)sink()# ------------------------------------------------------------------------# export the necessary variables as flow variables to KNIME# https://forum.knime.com/t/creating-flow-variables-in-r-scripting-nodes/5701/2?u=mlauber71# https://forum.knime.com/t/creating-flow-variables-with-r-snippet-works-but/9007?u=mlauber71knime.flow.out <- list(var_timestamp_day = var_timestamp_day, var_timestamp_time = var_timestamp_time, var_model_name = var_model_name, var_model_name_full = var_model_name_full, x = paste(c(x), collapse=', ' ), y = y, max_runtime_secs_opts = max_runtime_secs_opts, var_path_model = var_path_model, var_mojo_file_name = var_mojo_file_name, v_csv_file_variable_importance = v_csv_file_variable_importance, v_csv_file_leaderboard = v_csv_file_leaderboard, v_csv_file_model_summary = v_csv_file_model_summary, var_txt_summary = var_txt_summary)# shut down the H2O cluster in the end# h2o.shutdown(prompt = FALSE) # make sure you have R and the necessary R packages installed, also check aout the pdf in /script/conda install -c conda-forge r-h2o# Install R alongside KNIME on Windows and MacOS# https://forum.knime.com/t/install-r-alongside-knime-on-windows-and-macos/13287# R and Rtools# https://forum.knime.com/t/how-to-import-tables-from-docx-documents-via-r-snippet/19284/10# RServe 1.8.6 on MacOSX# https://forum.knime.com/t/installing-rserve-1-8-6-on-macos-10-15-catalina/20909/6?u=mlauber71library(h2o)# if you wish to use the 'pure' R code and import the data with parquetlibrary(arrow) additional R packages needed:ggplot2, lift, reshape2If you use the R wrapper you will need the h2o package and the arrow package if you plan on using the pure R script in the /script/subfolderhttp://docs.h2o.ai/h2o/latest-stable/h2o-docs/downloading.html Inspect the models so far and see to results. This will also give you a quick idea where you stand and what you would be able to achieve.Along with all parameters to load the respective model. Propagate R environmentfor KNIME on MacOS withMinicondaconfigure how to handle the environmentdefault = just check the namestrain.tableinput_table_2and mainR codewrapperinput_table_1scored outputedit: v_runtime_automlset the maximum runtime ofH2O.ai AutoML in SECONDSextract parametersfrom Rwhich have been usedto calculate the modelexclude pathsh2o_list_of_models.csvappend if CSV already exists to collect allmodel runsRead the MOJOmodelyou could check out this nodeScore the test tableyou might also use a third table to validatethat has not been used developing themodelvar_model_name_fullsolutionto string^(.*submission|solution).*$h2o_list_of_models.csvAUC DESCkeep best modelRead the MOJOmodelRead Variableimportancetest.tablePropagate R environmentfor KNIME on Windows withMinicondaconfigure how to handle the environmentdefault = just check the namesbinary classification modelswith Rset the parameters!java.homejava.homecreate initial Test andTraining dataCensus incomeclassificationknime_r_environment Table Reader Add Table To R Table to R R to Table Integer Input collect meta data Merge Variables RowID Variable toTable Row Column Filter String to Path(Variable) CSV Writer Column Resorter H2O MOJO Reader Binary ClassificationInspector H2O MOJO Predictor(Classification) ROC Curve (local) ConstantValue Column Number To String Column Filter Column Rename CSV Reader String to Path(Variable) Sorter Row Filter Column Filter Table Rowto Variable H2O MOJO Reader CSV Reader Table Reader knime_r_environment_windows Model QualityClassification - Graphics Extract SystemProperties Table Columnto Variable Joiner Test Training

Nodes

Extensions

Links

Download

To use this workflow in KNIME, download it from the below URL and open it in KNIME:

Download Workflow
Created by: m_lauber
Created at: 2020-01-23
On NodePit since: 2020-02-25
Last update: 2024-04-17
Created with KNIME version: v4.6.3
Tags: h2oautomlknimexgboostrrstudio

Deploy, schedule, execute, and monitor your KNIME workflows locally, in the cloud or on-premises – with our brand new NodePit Runner.

Try NodePit Runner!