2018-06-30 R机器学习中的应用

From shirinsplayground,非常好的机器学习的文章,保存下来,慢慢学习。

https://shirinsplayground.netlify.com/2018/06/intro_to_ml_workshop_heidelberg/

Code for Workshop: Introduction to Machine Learning with R

June 29, 2018 in Rmachine learning

These are the slides from my workshop: Introduction to Machine Learning with R which I gave at the University of Heidelberg, Germany on June 28th 2018. The entire code accompanying the workshop can be found below the video.

The workshop covered the basics of machine learning. With an example dataset I went through a standard machine learning workflow in R with the packages caret and h2o:

reading in data

exploratory data analysis

missingness

feature engineering

training and test split

model training with Random Forests, Gradient Boosting, Neural Nets, etc.

hyperparameter tuning

Workshop - Introduction to Machine Learning with R from Shirin Glander

Setup

All analyses are done in R using RStudio. For detailed session information including R version, operating system and package versions, see the sessionInfo() output at the end of this document.

All figures are produced with ggplot2.

libraries

library(tidyverse)# for tidy data analysislibrary(readr)# for fast reading of input fileslibrary(mice)# mice package for Multivariate Imputation by Chained Equations (MICE)

Data preparation

The dataset

The dataset I am using in these example analyses, is the Breast Cancer Wisconsin (Diagnostic) Dataset. The data was downloaded from the UC Irvine Machine Learning Repository.

The first dataset looks at the predictor classes:

malignant or

benign breast mass.

The features characterise cell nucleus properties and were generated from image analysis of fine needle aspirates (FNA) of breast masses:

Sample ID (code number)

Clump thickness

Uniformity of cell size

Uniformity of cell shape

Marginal adhesion

Single epithelial cell size

Number of bare nuclei

Bland chromatin

Number of normal nuclei

Mitosis

Classes, i.e. diagnosis

bc_data <- read_delim("/Users/shiringlander/Documents/Github/intro_to_ml_workshop/intro_to_ml_uni_heidelberg/datasets/breast-cancer-wisconsin.data.txt",                      delim =",",                      col_names = c("sample_code_number","clump_thickness","uniformity_of_cell_size","uniformity_of_cell_shape","marginal_adhesion","single_epithelial_cell_size","bare_nuclei","bland_chromatin","normal_nucleoli","mitosis","classes")) %>%  mutate(bare_nuclei =as.numeric(bare_nuclei),        classes = ifelse(classes =="2","benign",                          ifelse(classes =="4","malignant", NA)))

summary(bc_data)

##  sample_code_number clump_thickness  uniformity_of_cell_size##  Min.  :  61634  Min.  : 1.000  Min.  : 1.000        ##  1st Qu.:  870688  1st Qu.: 2.000  1st Qu.: 1.000        ##  Median : 1171710  Median : 4.000  Median : 1.000        ##  Mean  : 1071704  Mean  : 4.418  Mean  : 3.134        ##  3rd Qu.: 1238298  3rd Qu.: 6.000  3rd Qu.: 5.000        ##  Max.  :13454352  Max.  :10.000  Max.  :10.000        ##                                                            ##  uniformity_of_cell_shape marginal_adhesion single_epithelial_cell_size##  Min.  : 1.000          Min.  : 1.000    Min.  : 1.000            ##  1st Qu.: 1.000          1st Qu.: 1.000    1st Qu.: 2.000            ##  Median : 1.000          Median : 1.000    Median : 2.000            ##  Mean  : 3.207          Mean  : 2.807    Mean  : 3.216            ##  3rd Qu.: 5.000          3rd Qu.: 4.000    3rd Qu.: 4.000            ##  Max.  :10.000          Max.  :10.000    Max.  :10.000            ##                                                                        ##  bare_nuclei    bland_chromatin  normal_nucleoli    mitosis      ##  Min.  : 1.000  Min.  : 1.000  Min.  : 1.000  Min.  : 1.000  ##  1st Qu.: 1.000  1st Qu.: 2.000  1st Qu.: 1.000  1st Qu.: 1.000  ##  Median : 1.000  Median : 3.000  Median : 1.000  Median : 1.000  ##  Mean  : 3.545  Mean  : 3.438  Mean  : 2.867  Mean  : 1.589  ##  3rd Qu.: 6.000  3rd Qu.: 5.000  3rd Qu.: 4.000  3rd Qu.: 1.000  ##  Max.  :10.000  Max.  :10.000  Max.  :10.000  Max.  :10.000  ##  NA's  :16                                                        ##    classes        ##  Length:699        ##  Class :character  ##  Mode  :character  ##                    ##                    ##                    ##

Missing data

# how many NAs are in the datamd.pattern(bc_data, plot =FALSE)

##    sample_code_number clump_thickness uniformity_of_cell_size## 683                  1              1                      1## 16                  1              1                      1##                      0              0                      0##    uniformity_of_cell_shape marginal_adhesion single_epithelial_cell_size## 683                        1                1                          1## 16                        1                1                          1##                            0                0                          0##    bland_chromatin normal_nucleoli mitosis classes bare_nuclei  ## 683              1              1      1      1          1  0## 16                1              1      1      1          0  1##                  0              0      0      0          16 16

bc_data<- bc_data %>%  drop_na() %>%select(classes, everything(), -sample_code_number)head(bc_data)

## # A tibble: 6 x 10##  classes  clump_thickness uniformity_of_cell_si… uniformity_of_cell_sha…#### 1 benign                  5                      1                      1## 2 benign                  5                      4                      4## 3 benign                  3                      1                      1## 4 benign                  6                      8                      8## 5 benign                  4                      1                      1## 6 malignant              8                    10                      10## # ... with 6 more variables: marginal_adhesion,## #  single_epithelial_cell_size, bare_nuclei,## #  bland_chromatin, normal_nucleoli, mitosis

Missing values can be imputed with the mice package.

More info and tutorial with code: https://shirinsplayground.netlify.com/2018/04/flu_prediction/

Data exploration

Response variable for classification

ggplot(bc_data, aes(x = classes, fill = classes)) +

  geom_bar()

More info on dealing with unbalanced classes: https://shiring.github.io/machine_learning/2017/04/02/unbalanced

Response variable for regression

ggplot(bc_data, aes(x = clump_thickness)) +

  geom_histogram(bins = 10)

Features

gather(bc_data, x, y, clump_thickness:mitosis) %>%ggplot(aes(x = y, color = classes, fill = classes)) +geom_density(alpha = 0.3) +facet_wrap( ~ x, scales ="free", ncol = 3)

Correlation graphs

co_mat_benign <- filter(bc_data, classes =="benign") %>%  select(-1) %>%  cor()co_mat_malignant <- filter(bc_data, classes =="malignant") %>%  select(-1) %>%  cor()library(igraph)g_benign <- graph.adjacency(co_mat_benign,                        weighted =TRUE,                        diag =FALSE,                        mode ="upper")g_malignant <- graph.adjacency(co_mat_malignant,                        weighted =TRUE,                        diag =FALSE,                        mode ="upper")# http://kateto.net/networks-r-igraphcut.off_b <- mean(E(g_benign)$weight)cut.off_m <- mean(E(g_malignant)$weight)g_benign_2 <- delete_edges(g_benign, E(g_benign)[weight < cut.off_b])g_malignant_2 <- delete_edges(g_malignant, E(g_malignant)[weight < cut.off_m])c_g_benign_2 <- cluster_fast_greedy(g_benign_2) c_g_malignant_2 <- cluster_fast_greedy(g_malignant_2)

par(mfrow = c(1,2))plot(c_g_benign_2, g_benign_2,    vertex.size = colSums(co_mat_benign) * 10,    vertex.frame.color = NA,      vertex.label.color ="black",      vertex.label.cex = 0.8,    edge.width = E(g_benign_2)$weight* 15,    layout = layout_with_fr(g_benign_2),    main ="Benign tumors")plot(c_g_malignant_2, g_malignant_2,    vertex.size = colSums(co_mat_malignant) * 10,    vertex.frame.color = NA,      vertex.label.color ="black",      vertex.label.cex = 0.8,    edge.width = E(g_malignant_2)$weight* 15,    layout = layout_with_fr(g_malignant_2),    main ="Malignant tumors")

Principal Component Analysis

library(ellipse)# perform pca and extract scorespcaOutput <- prcomp(as.matrix(bc_data[,-1]), scale =TRUE, center =TRUE)pcaOutput2 <-as.data.frame(pcaOutput$x)# define groups for plottingpcaOutput2$groups <- bc_data$classes  centroids <- aggregate(cbind(PC1, PC2) ~ groups, pcaOutput2, mean)conf.rgn  <-do.call(rbind, lapply(unique(pcaOutput2$groups),function(t)data.frame(groups = as.character(t),            ellipse(cov(pcaOutput2[pcaOutput2$groups == t,1:2]),                  centre = as.matrix(centroids[centroids$groups == t,2:3]),                  level =0.95),            stringsAsFactors = FALSE)))ggplot(data = pcaOutput2, aes(x = PC1, y = PC2, group = groups, color = groups))+geom_polygon(data = conf.rgn, aes(fill = groups), alpha =0.2)+geom_point(size =2, alpha =0.6)+labs(color ="",        fill ="")

Multidimensional Scaling

select(bc_data,-1) %>%  dist() %>%  cmdscale %>%as.data.frame() %>%  mutate(group= bc_data$classes) %>%  ggplot(aes(x = V1, y = V2, color =group)) +    geom_point()

t-SNE dimensionality reduction

library(tsne)select(bc_data,-1) %>%  dist() %>%  tsne() %>%as.data.frame() %>%  mutate(group= bc_data$classes) %>%  ggplot(aes(x = V1, y = V2, color =group)) +    geom_point()

Machine Learning packages for R

caret

# configure multicorelibrary(doParallel)cl <- makeCluster(detectCores())registerDoParallel(cl)library(caret)

Training, validation and test data

set.seed(42)index<- createDataPartition(bc_data$classes, p =0.7,list=FALSE)train_data <- bc_data[index, ]test_data  <- bc_data[-index, ]

bind_rows(data.frame(group="train", train_data),      data.frame(group="test", test_data)) %>%  gather(x, y, clump_thickness:mitosis) %>%  ggplot(aes(x = y, color =group, fill =group)) +    geom_density(alpha =0.3) +    facet_wrap( ~ x, scales ="free", ncol =3)

Regression

set.seed(42)model_glm<-caret::train(clump_thickness~.,data=train_data,method="glm",preProcess=c("scale", "center"),trControl=trainControl(method="repeatedcv",number=10,repeats=10,savePredictions=TRUE,verboseIter=FALSE))

model_glm

## Generalized Linear Model ## ## 479 samples##  9 predictor## ## Pre-processing: scaled (9), centered (9) ## Resampling: Cross-Validated (10 fold, repeated 10 times) ## Summary of sample sizes: 432, 431, 432, 431, 431, 431, ... ## Resampling results:## ##  RMSE      Rsquared  MAE    ##  1.972314  0.5254215  1.648832

predictions<- predict(model_glm, test_data)

# model_glm$finalModel$linear.predictors == model_glm$finalModel$fitted.valuesdata.frame(residuals = resid(model_glm),          predictors = model_glm$finalModel$linear.predictors) %>%  ggplot(aes(x = predictors, y = residuals)) +    geom_jitter() +    geom_smooth(method ="lm")

# y == train_data$clump_thicknessdata.frame(residuals = resid(model_glm),          y = model_glm$finalModel$y) %>%  ggplot(aes(x = y, y = residuals)) +    geom_jitter() +    geom_smooth(method ="lm")

data.frame(actual = test_data$clump_thickness,          predicted = predictions) %>%  ggplot(aes(x = actual, y = predicted)) +    geom_jitter() +    geom_smooth(method ="lm")

Classification

Decision trees

rpart

library(rpart)library(rpart.plot)set.seed(42)fit<-rpart(classes~.,data=train_data,method="class",control=rpart.control(xval=10,minbucket=2,cp=0),parms=list(split="information"))rpart.plot(fit,extra=100)

Random Forests

Random Forests predictions are based on the generation of multiple classification trees. They can be used for both, classification and regression tasks. Here, I show a classification task.

set.seed(42)model_rf<-caret::train(classes~.,data=train_data,method="rf",preProcess=c("scale", "center"),trControl=trainControl(method="repeatedcv",number=5,repeats=3,savePredictions=TRUE,verboseIter=FALSE))

When you specify savePredictions = TRUE, you can access the cross-validation resuls with model_rf$pred.

model_rf

## Random Forest ## ## 479 samples##  9 predictor##  2 classes: 'benign', 'malignant' ## ## Pre-processing: scaled (9), centered (9) ## Resampling: Cross-Validated (10 fold, repeated 10 times) ## Summary of sample sizes: 432, 431, 431, 431, 431, 431, ... ## Resampling results across tuning parameters:## ##  mtry  Accuracy  Kappa    ##  2    0.9776753  0.9513499##  5    0.9757957  0.9469999##  9    0.9714200  0.9370285## ## Accuracy was used to select the optimal model using the largest value.## The final value used for the model was mtry = 2.

model_rf$finalModel$confusion

##          benign malignant class.error## benign      304        7  0.02250804## malignant      5      163  0.02976190

Dealing with unbalanced data

Luckily, caret makes it very easy to incorporate over- and under-sampling techniques with cross-validation resampling. We can simply add the sampling option to our trainControl and choose down for under- (also called down-) sampling. The rest stays the same as with our original model.

set.seed(42)model_rf_down<-caret::train(classes~.,data=train_data,method="rf",preProcess=c("scale", "center"),trControl=trainControl(method="repeatedcv",number=10,repeats=10,savePredictions=TRUE,verboseIter=FALSE,sampling="down"))

model_rf_down

## Random Forest ## ## 479 samples##  9 predictor##  2 classes: 'benign', 'malignant' ## ## Pre-processing: scaled (9), centered (9) ## Resampling: Cross-Validated (10 fold, repeated 10 times) ## Summary of sample sizes: 432, 431, 431, 431, 431, 431, ... ## Addtional sampling using down-sampling prior to pre-processing## ## Resampling results across tuning parameters:## ##  mtry  Accuracy  Kappa    ##  2    0.9797503  0.9563138##  5    0.9741198  0.9438326##  9    0.9699578  0.9346310## ## Accuracy was used to select the optimal model using the largest value.## The final value used for the model was mtry = 2.

Feature Importance

imp<- model_rf$finalModel$importanceimp[order(imp, decreasing = TRUE), ]

##    uniformity_of_cell_size    uniformity_of_cell_shape ##                  43.936945                  39.840595 ##                bare_nuclei            bland_chromatin ##                  33.820345                  31.984813 ##            normal_nucleoli single_epithelial_cell_size ##                  21.686039                  17.761202 ##            clump_thickness          marginal_adhesion ##                  16.318817                    9.518437 ##                    mitosis ##                    2.220633

# estimate variable importanceimportance<- varImp(model_rf, scale = TRUE)plot(importance)

predicting test data

confusionMatrix(predict(model_rf, test_data),as.factor(test_data$classes))

## Confusion Matrix and Statistics## ##            Reference## Prediction  benign malignant##  benign      128        4##  malignant      5        67##                                          ##                Accuracy : 0.9559          ##                  95% CI : (0.9179, 0.9796)##    No Information Rate : 0.652          ##    P-Value [Acc > NIR] : <2e-16          ##                                          ##                  Kappa : 0.9031          ##  Mcnemar's Test P-Value : 1              ##                                          ##            Sensitivity : 0.9624          ##            Specificity : 0.9437          ##          Pos Pred Value : 0.9697          ##          Neg Pred Value : 0.9306          ##              Prevalence : 0.6520          ##          Detection Rate : 0.6275          ##    Detection Prevalence : 0.6471          ##      Balanced Accuracy : 0.9530          ##                                          ##        'Positive' Class : benign          ##

results <- data.frame(actual = test_data$classes,                      predict(model_rf, test_data, type ="prob"))results$prediction <- ifelse(results$benign >0.5,"benign",                            ifelse(results$malignant >0.5,"malignant", NA))results$correct <- ifelse(results$actual == results$prediction,TRUE,FALSE)ggplot(results, aes(x = prediction, fill = correct)) +  geom_bar(position ="dodge")

ggplot(results, aes(x = prediction, y = benign, color = correct, shape = correct)) +

  geom_jitter(size = 3, alpha = 0.6)

Extreme gradient boosting trees

Extreme gradient boosting (XGBoost) is a faster and improved implementation of gradient boosting for supervised learning.

“XGBoost uses a more regularized model formalization to control over-fitting, which gives it better performance.” Tianqi Chen, developer of xgboost

XGBoost is a tree ensemble model, which means the sum of predictions from a set of classification and regression trees (CART). In that, XGBoost is similar to Random Forests but it uses a different approach to model training. Can be used for classification and regression tasks. Here, I show a classification task.

set.seed(42)model_xgb<-caret::train(classes~.,data=train_data,method="xgbTree",preProcess=c("scale", "center"),trControl=trainControl(method="repeatedcv",number=5,repeats=3,savePredictions=TRUE,verboseIter=FALSE))

model_xgb

## eXtreme Gradient Boosting ## ## 479 samples##  9 predictor##  2 classes: 'benign', 'malignant' ## ## Pre-processing: scaled (9), centered (9) ## Resampling: Cross-Validated (10 fold, repeated 10 times) ## Summary of sample sizes: 432, 431, 431, 431, 431, 431, ... ## Resampling results across tuning parameters:## ##  eta  max_depth  colsample_bytree  subsample  nrounds  Accuracy ##  0.3  1          0.6              0.50        50      0.9567788##  0.3  1          0.6              0.50      100      0.9544912##  0.3  1          0.6              0.50      150      0.9513572##  0.3  1          0.6              0.75        50      0.9576164##  0.3  1          0.6              0.75      100      0.9536448##  0.3  1          0.6              0.75      150      0.9525987##  0.3  1          0.6              1.00        50      0.9559409##  0.3  1          0.6              1.00      100      0.9555242##  0.3  1          0.6              1.00      150      0.9551031##  0.3  1          0.8              0.50        50      0.9718588##  0.3  1          0.8              0.50      100      0.9720583##  0.3  1          0.8              0.50      150      0.9699879##  0.3  1          0.8              0.75        50      0.9726964##  0.3  1          0.8              0.75      100      0.9724664##  0.3  1          0.8              0.75      150      0.9705868##  0.3  1          0.8              1.00        50      0.9714202##  0.3  1          0.8              1.00      100      0.9710035##  0.3  1          0.8              1.00      150      0.9705866##  0.3  2          0.6              0.50        50      0.9559448##  0.3  2          0.6              0.50      100      0.9565397##  0.3  2          0.6              0.50      150      0.9555063##  0.3  2          0.6              0.75        50      0.9530150##  0.3  2          0.6              0.75      100      0.9550985##  0.3  2          0.6              0.75      150      0.9551070##  0.3  2          0.6              1.00        50      0.9532320##  0.3  2          0.6              1.00      100      0.9551072##  0.3  2          0.6              1.00      150      0.9557237##  0.3  2          0.8              0.50        50      0.9720583##  0.3  2          0.8              0.50      100      0.9735166##  0.3  2          0.8              0.50      150      0.9720540##  0.3  2          0.8              0.75        50      0.9722494##  0.3  2          0.8              0.75      100      0.9726703##  0.3  2          0.8              0.75      150      0.9716374##  0.3  2          0.8              1.00        50      0.9716327##  0.3  2          0.8              1.00      100      0.9724622##  0.3  2          0.8              1.00      150      0.9718416##  0.3  3          0.6              0.50        50      0.9548905##  0.3  3          0.6              0.50      100      0.9557237##  0.3  3          0.6              0.50      150      0.9555198##  0.3  3          0.6              0.75        50      0.9561404##  0.3  3          0.6              0.75      100      0.9546820##  0.3  3          0.6              0.75      150      0.9552982##  0.3  3          0.6              1.00        50      0.9577983##  0.3  3          0.6              1.00      100      0.9573819##  0.3  3          0.6              1.00      150      0.9567655##  0.3  3          0.8              0.50        50      0.9733131##  0.3  3          0.8              0.50      100      0.9728829##  0.3  3          0.8              0.50      150      0.9718499##  0.3  3          0.8              0.75        50      0.9751879##  0.3  3          0.8              0.75      100      0.9743546##  0.3  3          0.8              0.75      150      0.9735212##  0.3  3          0.8              1.00        50      0.9743372##  0.3  3          0.8              1.00      100      0.9737122##  0.3  3          0.8              1.00      150      0.9743461##  0.4  1          0.6              0.50        50      0.9548861##  0.4  1          0.6              0.50      100      0.9528290##  0.4  1          0.6              0.50      150      0.9498772##  0.4  1          0.6              0.75        50      0.9557239##  0.4  1          0.6              0.75      100      0.9513529##  0.4  1          0.6              0.75      150      0.9492779##  0.4  1          0.6              1.00        50      0.9559365##  0.4  1          0.6              1.00      100      0.9551031##  0.4  1          0.6              1.00      150      0.9536361##  0.4  1          0.8              0.50        50      0.9710164##  0.4  1          0.8              0.50      100      0.9697577##  0.4  1          0.8              0.50      150      0.9687074##  0.4  1          0.8              0.75        50      0.9710122##  0.4  1          0.8              0.75      100      0.9707996##  0.4  1          0.8              0.75      150      0.9691455##  0.4  1          0.8              1.00        50      0.9705911##  0.4  1          0.8              1.00      100      0.9697446##  0.4  1          0.8              1.00      150      0.9697576##  0.4  2          0.6              0.50        50      0.9544866##  0.4  2          0.6              0.50      100      0.9542694##  0.4  2          0.6              0.50      150      0.9536357##  0.4  2          0.6              0.75        50      0.9540611##  0.4  2          0.6              0.75      100      0.9542694##  0.4  2          0.6              0.75      150      0.9549033##  0.4  2          0.6              1.00        50      0.9540653##  0.4  2          0.6              1.00      100      0.9555239##  0.4  2          0.6              1.00      150      0.9546818##  0.4  2          0.8              0.50        50      0.9720670##  0.4  2          0.8              0.50      100      0.9695629##  0.4  2          0.8              0.50      150      0.9702006##  0.4  2          0.8              0.75        50      0.9722627##  0.4  2          0.8              0.75      100      0.9720500##  0.4  2          0.8              0.75      150      0.9716289##  0.4  2          0.8              1.00        50      0.9726705##  0.4  2          0.8              1.00      100      0.9708042##  0.4  2          0.8              1.00      150      0.9708129##  0.4  3          0.6              0.50        50      0.9555150##  0.4  3          0.6              0.50      100      0.9553021##  0.4  3          0.6              0.50      150      0.9548943##  0.4  3          0.6              0.75        50      0.9555281##  0.4  3          0.6              0.75      100      0.9563662##  0.4  3          0.6              0.75      150      0.9555324##  0.4  3          0.6              1.00        50      0.9575900##  0.4  3          0.6              1.00      100      0.9571735##  0.4  3          0.6              1.00      150      0.9559104##  0.4  3          0.8              0.50        50      0.9737255##  0.4  3          0.8              0.50      100      0.9745501##  0.4  3          0.8              0.50      150      0.9730874##  0.4  3          0.8              0.75        50      0.9747539##  0.4  3          0.8              0.75      100      0.9724664##  0.4  3          0.8              0.75      150      0.9720498##  0.4  3          0.8              1.00        50      0.9747539##  0.4  3          0.8              1.00      100      0.9749624##  0.4  3          0.8              1.00      150      0.9734996##  Kappa    ##  0.9050828##  0.8999999##  0.8930637##  0.9067208##  0.8982284##  0.8959903##  0.9028825##  0.9022543##  0.9014018##  0.9382467##  0.9386326##  0.9340573##  0.9400323##  0.9395968##  0.9353783##  0.9372262##  0.9362148##  0.9353247##  0.9032270##  0.9047203##  0.9024465##  0.8968511##  0.9015282##  0.9016169##  0.8971329##  0.9015111##  0.9028614##  0.9387022##  0.9419143##  0.9387792##  0.9391933##  0.9401872##  0.9379714##  0.9377309##  0.9397601##  0.9384827##  0.9008861##  0.9029797##  0.9024531##  0.9037859##  0.9004226##  0.9019909##  0.9074584##  0.9064701##  0.9051441##  0.9414031##  0.9405025##  0.9380734##  0.9456856##  0.9438986##  0.9419994##  0.9438642##  0.9426000##  0.9439780##  0.9007223##  0.8964381##  0.8897615##  0.9027951##  0.8931520##  0.8886910##  0.9030461##  0.9014362##  0.8982364##  0.9363059##  0.9334254##  0.9311383##  0.9361883##  0.9357131##  0.9320657##  0.9353688##  0.9333607##  0.9334467##  0.8999756##  0.8997888##  0.8983861##  0.8991356##  0.8998960##  0.9013529##  0.8990428##  0.9023340##  0.9004889##  0.9387165##  0.9332663##  0.9345567##  0.9393855##  0.9389455##  0.9380863##  0.9401366##  0.9361847##  0.9361724##  0.9021263##  0.9017938##  0.9010613##  0.9025263##  0.9043436##  0.9024744##  0.9069828##  0.9059579##  0.9031829##  0.9424523##  0.9442537##  0.9410193##  0.9447486##  0.9397683##  0.9388701##  0.9449064##  0.9454375##  0.9422358## ## Tuning parameter 'gamma' was held constant at a value of 0## ## Tuning parameter 'min_child_weight' was held constant at a value of 1## Accuracy was used to select the optimal model using the largest value.## The final values used for the model were nrounds = 50, max_depth = 3,##  eta = 0.3, gamma = 0, colsample_bytree = 0.8, min_child_weight = 1##  and subsample = 0.75.

Feature Importance

importance<-varImp(model_xgb,scale=TRUE)plot(importance)

predicting test data

confusionMatrix(predict(model_xgb, test_data),as.factor(test_data$classes))

## Confusion Matrix and Statistics## ##            Reference## Prediction  benign malignant##  benign      128        3##  malignant      5        68##                                          ##                Accuracy : 0.9608          ##                  95% CI : (0.9242, 0.9829)##    No Information Rate : 0.652          ##    P-Value [Acc > NIR] : <2e-16          ##                                          ##                  Kappa : 0.9142          ##  Mcnemar's Test P-Value : 0.7237          ##                                          ##            Sensitivity : 0.9624          ##            Specificity : 0.9577          ##          Pos Pred Value : 0.9771          ##          Neg Pred Value : 0.9315          ##              Prevalence : 0.6520          ##          Detection Rate : 0.6275          ##    Detection Prevalence : 0.6422          ##      Balanced Accuracy : 0.9601          ##                                          ##        'Positive' Class : benign          ##

results <- data.frame(actual = test_data$classes,                      predict(model_xgb, test_data, type ="prob"))results$prediction <- ifelse(results$benign >0.5,"benign",                            ifelse(results$malignant >0.5,"malignant", NA))results$correct <- ifelse(results$actual == results$prediction,TRUE,FALSE)ggplot(results, aes(x = prediction, fill = correct)) +  geom_bar(position ="dodge")

ggplot(results, aes(x = prediction, y = benign, color = correct, shape = correct)) +

  geom_jitter(size = 3, alpha = 0.6)

Available models in caret

https://topepo.github.io/caret/available-models.html

Feature Selection

Performing feature selection on the whole dataset would lead to prediction bias, we therefore need to run the whole modeling process on the training data alone!

Correlation

Correlations between all features are calculated and visualised with the corrplot package. I am then removing all features with a correlation higher than 0.7, keeping the feature with the lower mean.

library(corrplot)# calculate correlation matrixcorMatMy<- cor(train_data[, -1])corrplot(corMatMy, order ="hclust")

#Applycorrelationfilterat0.70,highlyCor<-colnames(train_data[, -1])[findCorrelation(corMatMy, cutoff = 0.7, verbose = TRUE)]

## Compare row 2  and column  3 with corr  0.908 ##  Means:  0.709 vs 0.594 so flagging column 2 ## Compare row 3  and column  7 with corr  0.749 ##  Means:  0.67 vs 0.569 so flagging column 3 ## All correlations <= 0.7

# which variables are flagged for removal?highlyCor

## [1]"uniformity_of_cell_size""uniformity_of_cell_shape"

#then we remove these variablestrain_data_cor <- train_data[,which(!colnames(train_data) %in% highlyCor)]

Recursive Feature Elimination (RFE)

Another way to choose features is with Recursive Feature Elimination. RFE uses a Random Forest algorithm to test combinations of features and rate each with an accuracy score. The combination with the highest score is usually preferential.

set.seed(7)results_rfe<-rfe(x=train_data[,-1],y=as.factor(train_data$classes),sizes=c(1:9),rfeControl=rfeControl(functions=rfFuncs,method="cv",number=10))

# chosen featurespredictors(results_rfe)

## [1]"bare_nuclei""clump_thickness"## [3]"uniformity_of_cell_size""uniformity_of_cell_shape"## [5]"bland_chromatin""normal_nucleoli"## [7]"marginal_adhesion""single_epithelial_cell_size"

train_data_rfe <- train_data[, c(1,which(colnames(train_data) %in% predictors(results_rfe)))]

Genetic Algorithm (GA)

The Genetic Algorithm (GA) has been developed based on evolutionary principles of natural selection: It aims to optimize a population of individuals with a given set of genotypes by modeling selection over time. In each generation (i.e. iteration), each individual’s fitness is calculated based on their genotypes. Then, the fittest individuals are chosen to produce the next generation. This subsequent generation of individuals will have genotypes resulting from (re-) combinations of the parental alleles. These new genotypes will again determine each individual’s fitness. This selection process is iterated for a specified number of generations and (ideally) leads to fixation of the fittest alleles in the gene pool.

This concept of optimization can be applied to non-evolutionary models as well, like feature selection processes in machine learning.

set.seed(27)model_ga <- gafs(x = train_data[,-1],                  y = as.factor(train_data$classes),                iters =10, # generationsofalgorithm                popSize =10, # populationsizeforeachgenerationlevels= c("malignant","benign"),                gafsControl = gafsControl(functions = rfGA, # Assess fitnesswithRF                                          method ="cv",    #10foldcrossvalidationgenParallel =TRUE, #Useparallelprogramming                                          allowParallel =TRUE))

plot(model_ga)# Plot mean fitness (AUC) by generation

train_data_ga <- train_data[, c(1,which(colnames(train_data) %in% model_ga$ga$final))]

Hyperparameter tuning with caret

Cartesian Grid

mtry: Number of variables randomly sampled as candidates at each split.

set.seed(42)grid <- expand.grid(mtry = c(1:10))model_rf_tune_man <- caret::train(classes ~ .,data= train_data,                        method ="rf",                        preProcess = c("scale","center"),                        trControl = trainControl(method ="repeatedcv",number=10,                                                  repeats =10,                                                  savePredictions =TRUE,                                                  verboseIter =FALSE),                        tuneGrid = grid)

model_rf_tune_man

## Random Forest ## ## 479 samples##  9 predictor##  2 classes: 'benign', 'malignant' ## ## Pre-processing: scaled (9), centered (9) ## Resampling: Cross-Validated (10 fold, repeated 10 times) ## Summary of sample sizes: 432, 431, 431, 431, 431, 431, ... ## Resampling results across tuning parameters:## ##  mtry  Accuracy  Kappa    ##    1    0.9785044  0.9532161##    2    0.9772586  0.9504377##    3    0.9774625  0.9508246##    4    0.9766333  0.9488778##    5    0.9753789  0.9460274##    6    0.9737078  0.9422613##    7    0.9730957  0.9408547##    8    0.9714155  0.9371611##    9    0.9718280  0.9380578##  10    0.9718280  0.9380135## ## Accuracy was used to select the optimal model using the largest value.## The final value used for the model was mtry = 1.

plot(model_rf_tune_man)

Random Search

set.seed(42)model_rf_tune_auto<-caret::train(classes~.,data=train_data,method="rf",preProcess=c("scale", "center"),trControl=trainControl(method="repeatedcv",number=10,repeats=10,savePredictions=TRUE,verboseIter=FALSE,search="random"),tuneGrid=grid,tuneLength=15)

model_rf_tune_auto

## Random Forest ## ## 479 samples##  9 predictor##  2 classes: 'benign', 'malignant' ## ## Pre-processing: scaled (9), centered (9) ## Resampling: Cross-Validated (10 fold, repeated 10 times) ## Summary of sample sizes: 432, 431, 431, 431, 431, 431, ... ## Resampling results across tuning parameters:## ##  mtry  Accuracy  Kappa    ##    1    0.9785044  0.9532161##    2    0.9772586  0.9504377##    3    0.9774625  0.9508246##    4    0.9766333  0.9488778##    5    0.9753789  0.9460274##    6    0.9737078  0.9422613##    7    0.9730957  0.9408547##    8    0.9714155  0.9371611##    9    0.9718280  0.9380578##  10    0.9718280  0.9380135## ## Accuracy was used to select the optimal model using the largest value.## The final value used for the model was mtry = 1.

plot(model_rf_tune_auto)

Grid search with h2o

The R package h2o provides a convenient interface to H2O, which is an open-source machine learning and deep learning platform. H2O distributes a wide range of common machine learning algorithms for classification, regression and deep learning.

library(h2o)

h2o.init(nthreads = -1)

##  Connection successful!## ## R is connected to the H2O cluster: ##    H2O cluster uptime:        26 minutes 45 seconds ##    H2O cluster timezone:      Europe/Berlin ##    H2O data parsing timezone:  UTC ##    H2O cluster version:        3.20.0.2 ##    H2O cluster version age:    13 days  ##    H2O cluster name:          H2O_started_from_R_shiringlander_jrj894 ##    H2O cluster total nodes:    1 ##    H2O cluster total memory:  3.24 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:        XGBoost, Algos, AutoML, Core V3, Core V4 ##    R Version:                  R version 3.5.0 (2018-04-23)

h2o.no_progress()bc_data_hf<-as.h2o(bc_data)

h2o.describe(bc_data_hf) %>%  gather(x, y, Zeros:Sigma) %>%  mutate(group= ifelse(x %in% c("Min","Max","Mean"),"min, mean, max",                        ifelse(x %in% c("NegInf","PosInf"),"Inf","sigma, zeros"))) %>%  ggplot(aes(x = Label, y =as.numeric(y), color = x)) +    geom_point(size =4, alpha =0.6) +    scale_color_brewer(palette ="Set1") +    theme(axis.text.x = element_text(angle =45, vjust =1, hjust =1)) +    facet_grid(group~ ., scales ="free") +    labs(x ="Feature",        y ="Value",        color ="")

library(reshape2)# for meltingbc_data_hf[,1] <- h2o.asfactor(bc_data_hf[,1])cor <- h2o.cor(bc_data_hf)rownames(cor) <- colnames(cor)melt(cor) %>%  mutate(Var2 = rep(rownames(cor), nrow(cor))) %>%  mutate(Var2 = factor(Var2, levels = colnames(cor))) %>%  mutate(variable = factor(variable, levels = colnames(cor))) %>%  ggplot(aes(x = variable, y = Var2, fill = value)) +    geom_tile(width =0.9, height =0.9) +    scale_fill_gradient2(low ="white", high ="red", name ="Cor.") +    theme(axis.text.x = element_text(angle =90, vjust =0.5, hjust =1)) +labs(x ="",          y ="")

Training, validation and test data

splits <- h2o.splitFrame(bc_data_hf,                          ratios = c(0.7, 0.15),                          seed = 1)train <- splits[[1]]valid <- splits[[2]]test<- splits[[3]]response <-"classes"features <- setdiff(colnames(train), response)

summary(as.factor(train$classes), exact_quantiles =TRUE)

##  classes      ##  benign  :313 ##  malignant:167

summary(as.factor(valid$classes), exact_quantiles =TRUE)

##  classes      ##  benign  :64 ##  malignant:38

summary(as.factor(test$classes), exact_quantiles =TRUE)

##  classes      ##  benign  :67 ##  malignant:34

pca<- h2o.prcomp(training_frame = train,          x = features,          validation_frame = valid,          transform ="NORMALIZE",          impute_missing = TRUE,          k =3,          seed =42)eigenvec <- as.data.frame(pca@model$eigenvectors)eigenvec$label<- featureslibrary(ggrepel)ggplot(eigenvec, aes(x = pc1, y = pc2, label = label)) +  geom_point(color ="navy", alpha =0.7) +  geom_text_repel()

Classification

Random Forest

hyper_params <-list(                    ntrees = c(25,50,75,100),                    max_depth = c(10,20,30),                    min_rows = c(1,3,5)                    )search_criteria <-list(                        strategy ="RandomDiscrete",                        max_models =50,                        max_runtime_secs =360,                        stopping_rounds =5,                                  stopping_metric ="AUC",                              stopping_tolerance =0.0005,                        seed =42)

rf_grid<-h2o.grid(algorithm="randomForest", #h2o.randomForest,                                                #alternativelyh2o.gbm#forGradientboostingtreesx=features,y=response,grid_id="rf_grid",training_frame=train,validation_frame=valid,nfolds=25,fold_assignment="Stratified",hyper_params=hyper_params,search_criteria=search_criteria,seed=42)

# performance metrics where smaller is better -> order with decreasing = FALSEsort_options_1 <- c("mean_per_class_error","mse","err","logloss")for(sort_by_1insort_options_1) {    grid <- h2o.getGrid("rf_grid", sort_by = sort_by_1, decreasing =FALSE)    model_ids <- grid@model_ids  best_model <- h2o.getModel(model_ids[[1]])    h2o.saveModel(best_model, path="models", force =TRUE)  }# performance metrics where bigger is better -> order with decreasing = TRUEsort_options_2 <- c("auc","precision","accuracy","recall","specificity")for(sort_by_2insort_options_2) {    grid <- h2o.getGrid("rf_grid", sort_by = sort_by_2, decreasing =TRUE)    model_ids <- grid@model_ids  best_model <- h2o.getModel(model_ids[[1]])    h2o.saveModel(best_model, path ="models", force =TRUE)  }

files <- list.files(path ="/Users/shiringlander/Documents/Github/intro_to_ml_workshop/intro_to_ml_uni_heidelberg/models")rf_models <- files[grep("rf_grid_model", files)]for(model_idinrf_models) {    path <- paste0("/Users/shiringlander/Documents/Github/intro_to_ml_workshop/intro_to_ml_uni_heidelberg","/models/", model_id)  best_model <- h2o.loadModel(path)  mse_auc_test <- data.frame(model_id = model_id,                              mse = h2o.mse(h2o.performance(best_model,test)),                            auc = h2o.auc(h2o.performance(best_model,test)))if(model_id == rf_models[[1]]) {        mse_auc_test_comb <- mse_auc_test      }else{        mse_auc_test_comb <- rbind(mse_auc_test_comb, mse_auc_test)      }}mse_auc_test_comb %>%  gather(x, y, mse:auc) %>%  ggplot(aes(x = model_id, y = y, fill = model_id)) +    facet_grid(x ~ ., scales ="free") +    geom_bar(stat="identity", alpha = 0.8, position ="dodge") +    scale_fill_brewer(palette ="Set1") +    theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust = 1),          plot.margin = unit(c(0.5, 0, 0, 1.5),"cm")) +    labs(x ="", y ="value", fill ="")

for(model_idinrf_models) {    best_model <- h2o.getModel(model_id)    finalRf_predictions <- data.frame(model_id = rep(best_model@model_id,                                                    nrow(test)),                                    actual = as.vector(test$classes),                                    as.data.frame(h2o.predict(object = best_model,                                                              newdata =test)))    finalRf_predictions$accurate<- ifelse(finalRf_predictions$actual==                                            finalRf_predictions$predict,"yes","no")    finalRf_predictions$predict_stringent<- ifelse(finalRf_predictions$benign> 0.8,"benign",                                                  ifelse(finalRf_predictions$malignant> 0.8,"malignant","uncertain"))    finalRf_predictions$accurate_stringent<- ifelse(finalRf_predictions$actual==                                                      finalRf_predictions$predict_stringent,"yes",                                          ifelse(finalRf_predictions$predict_stringent=="uncertain","na","no"))if(model_id == rf_models[[1]]) {        finalRf_predictions_comb <- finalRf_predictions      }else{        finalRf_predictions_comb <- rbind(finalRf_predictions_comb, finalRf_predictions)      }}

finalRf_predictions_comb%>%  ggplot(aes(x = actual, fill = accurate)) +    geom_bar(position ="dodge") +    scale_fill_brewer(palette ="Set1") +    facet_wrap(~ model_id,ncol =2) +    labs(fill ="Were\npredictions\naccurate?",        title ="Default predictions")

finalRf_predictions_comb%>%  subset(accurate_stringent !="na") %>%  ggplot(aes(x = actual, fill = accurate_stringent)) +    geom_bar(position ="dodge") +    scale_fill_brewer(palette ="Set1") +    facet_wrap(~ model_id,ncol =2) +    labs(fill ="Were\npredictions\naccurate?",        title ="Stringent predictions")

rf_model<- h2o.loadModel("/Users/shiringlander/Documents/Github/intro_to_ml_workshop/intro_to_ml_uni_heidelberg/models/rf_grid_model_0")

h2o.varimp_plot(rf_model)

#h2o.varimp(rf_model)

h2o.mean_per_class_error(rf_model, train =TRUE, valid =TRUE, xval =TRUE)

##trainvalidxval## 0.021962460.023437500.02515735

h2o.confusionMatrix(rf_model, valid =TRUE)

## Confusion Matrix (vertical: actual; across: predicted)  for max f1 @ threshold = 0.533333333333333:##          benign malignant    Error    Rate## benign        61        3 0.046875  =3/64## malignant      0        38 0.000000  =0/38## Totals        61        41 0.029412  =3/102

plot(rf_model,timestep="number_of_trees",metric="classification_error")

plot(rf_model,timestep="number_of_trees",metric="logloss")

plot(rf_model,timestep="number_of_trees",metric="AUC")

plot(rf_model,timestep="number_of_trees",metric="rmse")

h2o.auc(rf_model, train =TRUE)

##[1]0.9907214

h2o.auc(rf_model, valid =TRUE)

##[1]0.9829359

h2o.auc(rf_model, xval =TRUE)

##[1]0.9903005

perf <- h2o.performance(rf_model,test)perf

## H2OBinomialMetrics: drf## ## MSE:  0.03258482## RMSE:  0.1805127## LogLoss:  0.1072519## Mean Per-Class Error:  0.02985075## AUC:  0.9916594## Gini:  0.9833187## ## Confusion Matrix (vertical: actual; across: predicted) for F1-optimal threshold:##          benign malignant    Error    Rate## benign        63        4 0.059701  =4/67## malignant      0        34 0.000000  =0/34## Totals        63        38 0.039604  =4/101## ## Maximum Metrics: Maximum metrics at their respective thresholds##                        metric threshold    value idx## 1                      max f1  0.306667 0.944444  18## 2                      max f2  0.306667 0.977011  18## 3                max f0point5  0.720000 0.933735  13## 4                max accuracy  0.533333 0.960396  16## 5                max precision  1.000000 1.000000  0## 6                  max recall  0.306667 1.000000  18## 7              max specificity  1.000000 1.000000  0## 8            max absolute_mcc  0.306667 0.917235  18## 9  max min_per_class_accuracy  0.533333 0.955224  16## 10 max mean_per_class_accuracy  0.306667 0.970149  18## ## Gains/Lift Table: Extract with `h2o.gainsLift(,)` or `h2o.gainsLift(, valid=, xval=)`

plot(perf)

perf@metrics$thresholds_and_metric_scores%>%  ggplot(aes(x = fpr, y = tpr)) +    geom_point() +    geom_line() +    geom_abline(slope =1, intercept =0) +    labs(x ="False Positive Rate",        y ="True Positive Rate")

h2o.logloss(perf)

##[1]0.1072519

h2o.mse(perf)

##[1]0.03258482

h2o.auc(perf)

##[1]0.9916594

head(h2o.metric(perf))

##MetricsforThresholds:Binomialmetricsasafunctionofclassificationthresholds##thresholdf1f2f0point5accuracyprecisionrecall## 1  1.0000000.5833330.4666670.7777780.8019801.0000000.411765## 2  0.9866670.6666670.5555560.8333330.8316831.0000000.500000## 3  0.9733330.7169810.6129030.8636360.8514851.0000000.558824## 4  0.9600000.7407410.6410260.8771930.8613861.0000000.588235## 5  0.9466670.7636360.6687900.8898310.8712871.0000000.617647## 6  0.9200000.8070180.7232700.9126980.8910891.0000000.676471##specificityabsolute_mccmin_per_class_accuracymean_per_class_accuracy## 1    1.0000000.5631220.4117650.705882## 2    1.0000000.6315140.5000000.750000## 3    1.0000000.6757220.5588240.779412## 4    1.0000000.6975420.5882350.794118## 5    1.0000000.7192210.6176470.808824## 6    1.0000000.7622800.6764710.838235##tnsfnsfpstpstnrfnrfprtpridx## 1  67  20  0  14 1.0000000.5882350.0000000.4117650## 2  67  17  0  17 1.0000000.5000000.0000000.5000001## 3  67  15  0  19 1.0000000.4411760.0000000.5588242## 4  67  14  0  20 1.0000000.4117650.0000000.5882353## 5  67  13  0  21 1.0000000.3823530.0000000.6176474## 6  67  11  0  23 1.0000000.3235290.0000000.6764715

finalRf_predictions <- data.frame(actual =as.vector(test$classes),as.data.frame(h2o.predict(object = rf_model,                                                            newdata = test)))finalRf_predictions$accurate <- ifelse(finalRf_predictions$actual ==                                          finalRf_predictions$predict,"yes","no")finalRf_predictions$predict_stringent <- ifelse(finalRf_predictions$benign >0.8,"benign",                                                ifelse(finalRf_predictions$malignant                                                        >0.8,"malignant","uncertain"))finalRf_predictions$accurate_stringent <- ifelse(finalRf_predictions$actual ==                                                    finalRf_predictions$predict_stringent,"yes",                                        ifelse(finalRf_predictions$predict_stringent =="uncertain","na","no"))finalRf_predictions %>%  group_by(actual, predict) %>%  dplyr::summarise(n = n())

## # A tibble: 4 x 3## # Groups:  actual [?]##  actual    predict      n#### 1 benign    benign      64## 2 benign    malignant    3## 3 malignant benign        1## 4 malignant malignant    33

finalRf_predictions %>%

  group_by(actual, predict_stringent) %>%

  dplyr::summarise(n = n())

## # A tibble: 5 x 3## # Groups:  actual [?]##  actual    predict_stringent    n#### 1 benign    benign              62## 2 benign    malignant            2## 3 benign    uncertain            3## 4 malignant malignant            29## 5 malignant uncertain            5

finalRf_predictions %>%  ggplot(aes(x = actual, fill = accurate)) +    geom_bar(position ="dodge") +    scale_fill_brewer(palette ="Set1") +labs(fill ="Were\npredictions\naccurate?",        title ="Default predictions")

finalRf_predictions %>%  subset(accurate_stringent !="na") %>%  ggplot(aes(x = actual, fill = accurate_stringent)) +    geom_bar(position ="dodge") +    scale_fill_brewer(palette ="Set1") +labs(fill ="Were\npredictions\naccurate?",        title ="Stringent predictions")

df <- finalRf_predictions[, c(1,3,4)]thresholds <- seq(from = 0, to = 1, by = 0.1)prop_table <- data.frame(threshold = thresholds, prop_true_b = NA, prop_true_m = NA)for(threshold in thresholds) {  pred <- ifelse(df$benign > threshold,"benign","malignant")  pred_t <- ifelse(pred == df$actual, TRUE, FALSE)    group <- data.frame(df,"pred"= pred_t) %>%  group_by(actual, pred) %>%  dplyr::summarise(n = n())    group_b <- filter(group, actual =="benign")    prop_b <- sum(filter(group_b, pred == TRUE)$n) / sum(group_b$n)  prop_table[prop_table$threshold == threshold,"prop_true_b"] <- prop_b    group_m <- filter(group, actual =="malignant")    prop_m <- sum(filter(group_m, pred == TRUE)$n) / sum(group_m$n)  prop_table[prop_table$threshold == threshold,"prop_true_m"] <- prop_m}prop_table %>%  gather(x,y, prop_true_b:prop_true_m) %>%  ggplot(aes(x= threshold,y=y, color =x)) +    geom_point() +    geom_line() +    scale_color_brewer(palette ="Set1") +    labs(y="proportion of true predictions",        color ="b: benign cases\nm: malignant cases")

If you are interested in more machine learning posts, check out the category listing for machine_learning on my blog - https://shirinsplayground.netlify.com/categories/#posts-list-machine-learning - https://shiring.github.io/categories.html#machine_learning-ref

stopCluster(cl)h2o.shutdown()

## Are you sure you want to shutdown the H2O instance running at http://localhost:54321/ (Y/N)?

sessionInfo()

## R version 3.5.0 (2018-04-23)## Platform: x86_64-apple-darwin15.6.0 (64-bit)## Running under: macOS High Sierra 10.13.5## ## Matrix products: default## BLAS: /Library/Frameworks/R.framework/Versions/3.5/Resources/lib/libRblas.0.dylib## LAPACK: /Library/Frameworks/R.framework/Versions/3.5/Resources/lib/libRlapack.dylib## ## locale:## [1] de_DE.UTF-8/de_DE.UTF-8/de_DE.UTF-8/C/de_DE.UTF-8/de_DE.UTF-8## ## attached base packages:## [1] parallel  stats    graphics  grDevices utils    datasets  methods  ## [8] base    ## ## other attached packages:##  [1] ggrepel_0.8.0    reshape2_1.4.3    h2o_3.20.0.2    ##  [4] corrplot_0.84    caret_6.0-80      doParallel_1.0.11##  [7] iterators_1.0.9  foreach_1.4.4    ellipse_0.4.1    ## [10] igraph_1.2.1      bindrcpp_0.2.2    mice_3.1.0      ## [13] lattice_0.20-35  forcats_0.3.0    stringr_1.3.1    ## [16] dplyr_0.7.5      purrr_0.2.5      readr_1.1.1      ## [19] tidyr_0.8.1      tibble_1.4.2      ggplot2_2.2.1    ## [22] tidyverse_1.2.1  ## ## loaded via a namespace (and not attached):##  [1] minqa_1.2.4        colorspace_1.3-2    class_7.3-14      ##  [4] rprojroot_1.3-2    pls_2.6-0          rstudioapi_0.7    ##  [7] DRR_0.0.3          prodlim_2018.04.18  lubridate_1.7.4    ## [10] xml2_1.2.0          codetools_0.2-15    splines_3.5.0      ## [13] mnormt_1.5-5        robustbase_0.93-1  knitr_1.20        ## [16] RcppRoll_0.3.0      jsonlite_1.5        nloptr_1.0.4      ## [19] broom_0.4.4        ddalpha_1.3.4      kernlab_0.9-26    ## [22] sfsmisc_1.1-2      compiler_3.5.0      httr_1.3.1        ## [25] backports_1.1.2    assertthat_0.2.0    Matrix_1.2-14      ## [28] lazyeval_0.2.1      cli_1.0.0          htmltools_0.3.6    ## [31] tools_3.5.0        gtable_0.2.0        glue_1.2.0        ## [34] Rcpp_0.12.17        cellranger_1.1.0    nlme_3.1-137      ## [37] blogdown_0.6        psych_1.8.4        timeDate_3043.102  ## [40] xfun_0.2            gower_0.1.2        lme4_1.1-17        ## [43] rvest_0.3.2        pan_1.4            DEoptimR_1.0-8    ## [46] MASS_7.3-50        scales_0.5.0        ipred_0.9-6        ## [49] hms_0.4.2          RColorBrewer_1.1-2  yaml_2.1.19        ## [52] rpart_4.1-13        stringi_1.2.3      randomForest_4.6-14## [55] e1071_1.6-8        lava_1.6.1          geometry_0.3-6    ## [58] bitops_1.0-6        rlang_0.2.1        pkgconfig_2.0.1    ## [61] evaluate_0.10.1    bindr_0.1.1        recipes_0.1.3      ## [64] labeling_0.3        CVST_0.2-2          tidyselect_0.2.4  ## [67] plyr_1.8.4          magrittr_1.5        bookdown_0.7      ## [70] R6_2.2.2            mitml_0.3-5        dimRed_0.1.0      ## [73] pillar_1.2.3        haven_1.1.1        foreign_0.8-70    ## [76] withr_2.1.2        RCurl_1.95-4.10    survival_2.42-3    ## [79] abind_1.4-5        nnet_7.3-12        modelr_0.1.2      ## [82] crayon_1.3.4        jomo_2.6-2          xgboost_0.71.2    ## [85] utf8_1.1.4          rmarkdown_1.10      grid_3.5.0        ## [88] readxl_1.1.0        data.table_1.11.4  ModelMetrics_1.1.0 ## [91] digest_0.6.15      stats4_3.5.0        munsell_0.5.0      ## [94] magic_1.5-8

TAGGED IN

R machine learning caret h2o random forest gradient boosting neural nets

 NEXT

PREVIOUS 

© 2018 Dr. Shirin Glander. All Rights Reserved

最后编辑于
©著作权归作者所有,转载或内容合作请联系作者
  • 序言:七十年代末,一起剥皮案震惊了整个滨河市,随后出现的几起案子,更是在滨河造成了极大的恐慌,老刑警刘岩,带你破解...
    沈念sama阅读 206,214评论 6 481
  • 序言:滨河连续发生了三起死亡事件,死亡现场离奇诡异,居然都是意外死亡,警方通过查阅死者的电脑和手机,发现死者居然都...
    沈念sama阅读 88,307评论 2 382
  • 文/潘晓璐 我一进店门,熙熙楼的掌柜王于贵愁眉苦脸地迎上来,“玉大人,你说我怎么就摊上这事。” “怎么了?”我有些...
    开封第一讲书人阅读 152,543评论 0 341
  • 文/不坏的土叔 我叫张陵,是天一观的道长。 经常有香客问我,道长,这世上最难降的妖魔是什么? 我笑而不...
    开封第一讲书人阅读 55,221评论 1 279
  • 正文 为了忘掉前任,我火速办了婚礼,结果婚礼上,老公的妹妹穿的比我还像新娘。我一直安慰自己,他们只是感情好,可当我...
    茶点故事阅读 64,224评论 5 371
  • 文/花漫 我一把揭开白布。 她就那样静静地躺着,像睡着了一般。 火红的嫁衣衬着肌肤如雪。 梳的纹丝不乱的头发上,一...
    开封第一讲书人阅读 49,007评论 1 284
  • 那天,我揣着相机与录音,去河边找鬼。 笑死,一个胖子当着我的面吹牛,可吹牛的内容都是我干的。 我是一名探鬼主播,决...
    沈念sama阅读 38,313评论 3 399
  • 文/苍兰香墨 我猛地睁开眼,长吁一口气:“原来是场噩梦啊……” “哼!你这毒妇竟也来了?” 一声冷哼从身侧响起,我...
    开封第一讲书人阅读 36,956评论 0 259
  • 序言:老挝万荣一对情侣失踪,失踪者是张志新(化名)和其女友刘颖,没想到半个月后,有当地人在树林里发现了一具尸体,经...
    沈念sama阅读 43,441评论 1 300
  • 正文 独居荒郊野岭守林人离奇死亡,尸身上长有42处带血的脓包…… 初始之章·张勋 以下内容为张勋视角 年9月15日...
    茶点故事阅读 35,925评论 2 323
  • 正文 我和宋清朗相恋三年,在试婚纱的时候发现自己被绿了。 大学时的朋友给我发了我未婚夫和他白月光在一起吃饭的照片。...
    茶点故事阅读 38,018评论 1 333
  • 序言:一个原本活蹦乱跳的男人离奇死亡,死状恐怖,灵堂内的尸体忽然破棺而出,到底是诈尸还是另有隐情,我是刑警宁泽,带...
    沈念sama阅读 33,685评论 4 322
  • 正文 年R本政府宣布,位于F岛的核电站,受9级特大地震影响,放射性物质发生泄漏。R本人自食恶果不足惜,却给世界环境...
    茶点故事阅读 39,234评论 3 307
  • 文/蒙蒙 一、第九天 我趴在偏房一处隐蔽的房顶上张望。 院中可真热闹,春花似锦、人声如沸。这庄子的主人今日做“春日...
    开封第一讲书人阅读 30,240评论 0 19
  • 文/苍兰香墨 我抬头看了看天上的太阳。三九已至,却和暖如春,着一层夹袄步出监牢的瞬间,已是汗流浃背。 一阵脚步声响...
    开封第一讲书人阅读 31,464评论 1 261
  • 我被黑心中介骗来泰国打工, 没想到刚下飞机就差点儿被人妖公主榨干…… 1. 我叫王不留,地道东北人。 一个月前我还...
    沈念sama阅读 45,467评论 2 352
  • 正文 我出身青楼,却偏偏与公主长得像,于是被迫代替她去往敌国和亲。 传闻我的和亲对象是个残疾皇子,可洞房花烛夜当晚...
    茶点故事阅读 42,762评论 2 345

推荐阅读更多精彩内容