1 / 15

Practical Data Science with R - Choosing and evaluating models

Practical Data Science with R - Choosing and evaluating models. Kim Jeong Rae UOS.DML. 2014.11.3. Contents. Mapping problems to machine learning tasks Evaluating models Evaluating classification models Evaluating scoring models Evaluating probability models Evaluating ranking models

ross-wagner
Download Presentation

Practical Data Science with R - Choosing and evaluating models

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Practical Data Sciencewith R- Choosing and evaluating models Kim Jeong Rae UOS.DML. 2014.11.3.

  2. Contents • Mapping problems to machine learning tasks • Evaluating models • Evaluating classification models • Evaluating scoring models • Evaluating probability models • Evaluating ranking models • Evaluating clustering models • Validating models

  3. Mapping problems to machine learning tasks Some common Classification method

  4. Evaluating models – classification models(1/2) spamD <- read.table('spamD.tsv',header=T,sep='\t') spamTrain <- subset(spamD,spamD$rgroup>=10) # Spliting Test/Train data spamTest <- subset(spamD,spamD$rgroup<10) # Spliting Test/Train data spamVars <- setdiff(colnames(spamD),list('rgroup','spam')) # Deleting selection columns spamFormula <- as.formula(paste('spam=="spam"', paste(spamVars, collapse=' + '), sep=' ~ ')) spamModel <- glm(spamFormula, family=binomial(link='logit'), data=spamTrain) # y={0,1} spamTrain$pred<- predict(spamModel,newdata=spamTrain, type='response') spamTest$pred <- predict(spamModel,newdata=spamTest, type='response') print(with(spamTest,table(y=spam,glmPred=pred>0.5))) sample <- spamTest[c(7,35,224,327),c('spam','pred')] print(sample) Building and applying a logistic regression spam model

  5. Evaluating models – classification models(2/2) cM <- table(truth=spamTest$spam, prediction=spamTest$pred>0.5) print(cM) # Accuracy = (TP+TN)/(TP+FP+TN+FN) (cM[1,1]+cM[2,2])/sum(cM) # Precision = TP /(TP+FP) (cM[2,2])/(cM[2,2]+cM[1,2]) # Recall = TP /(TP+FN) (cM[2,2])/(cM[2,2]+cM[2,1]) # F1 = 2*Precision*Recall/(Precision+Recall) P <- (cM[2,2])/(cM[2,2]+cM[1,2]) R <- (cM[2,2])/(cM[2,2]+cM[2,1]) 2*P*R/(P+R) # Sensitivity(=True positive rate) = Recall # Specificity(=True negative rate) = TN/(TN+FP) (cM[1,1])/(cM[1,1]+cM[1,2]) Accuracy, Precision, Recall etc.

  6. Evaluating models – scoring models(1/2) d <- data.frame(y=(1:10)^2, x=1:10) model <- lm(y~x, data=d) summary(model) d$prediction <- predict(model, newdata=d) #install.packages('ggplot2') library('ggplot2') ggplot(data=d) + geom_point(aes(x=x,y=y)) + geom_line(aes(x=x,y=prediction),color='blue') + geom_segment(aes(x=x,y=prediction,yend=y,xend=x)) + scale_y_continuous('') Plotting residuals

  7. Evaluating models – scoring models(2/2) # RMSE sqrt(mean((d$prediction-d$y)^2)) # R-squared 1-sum((d$prediction-d$y)^2)/sum((mean(d$y)-d$y)^2) # correlation cor(d$prediction, d$y, method = "pearson") cor(d$prediction, d$y, method = "spearman") cor(d$prediction, d$y, method = "kendall") # absolute error (sum(abs(d$prediction-d$y))) # mean absolute error (sum(abs(d$prediction-d$y))/length(d$y)) # relative absolute error (sum(abs(d$prediction-d$y))/sum(abs(d$y))) RMSE, R-squared, correlation, absolute error

  8. Evaluating models – probability models(1/3) ggplot(data=spamTest) + geom_density(aes(x=pred,color=spam,linetype=spam)) Making a double density plot

  9. Evaluating models – probability models(2/3) #install.packages('ROCR') library('ROCR') eval <- prediction(spamTest$pred,spamTest$spam) plot(performance(eval,"tpr","fpr")) print(attributes(performance(eval,'auc'))$y.values[[1]]) Plotting the Receiver Operating Characteristic Curve

  10. Evaluating models – probability models(3/3) #### 3.3 Calculating log likelihood #### sum(ifelse(spamTest$spam=='spam', log(spamTest$pred), log(1-spamTest$pred))) sum(ifelse(spamTest$spam=='spam', log(spamTest$pred), log(1-spamTest$pred)))/dim(spamTest)[[1]] #### 3.4 Computing the null model's log likelihood #### pNull <- sum(ifelse(spamTest$spam=='spam',1,0))/dim(spamTest)[[1]] sum(ifelse(spamTest$spam=='spam',1,0))*log(pNull) + sum(ifelse(spamTest$spam=='spam',0,1))*log(1-pNull) #### 3.5 Calculating entropy and conditional entropy #### entropy <- function(x) { xpos <- x[x>0] scaled <- xpos/sum(xpos) sum(-scaled*log(scaled,2)) } print(entropy(table(spamTest$spam))) conditionalEntropy <- function(t) { (sum(t[,1])*entropy(t[,1]) + sum(t[,2])*entropy(t[,2]))/sum(t) } print(conditionalEntropy(cM)) Log likelihood, Entropy

  11. Evaluating models – clustering models #### 5.1 Clustering random data in the plane #### set.seed(32297) d <- data.frame(x=runif(100),y=runif(100)) clus <- kmeans(d,centers=5) d$cluster <- clus$cluster #### 5.2 Plotting our clusters #### #install.packages("grDevises") library('ggplot2'); library('grDevices') h <- do.call(rbind, lapply(unique(clus$cluster), function(c) { f <- subset(d,cluster==c); f[chull(f),]})) ggplot() + geom_text(data=d,aes(label=cluster,x=x,y=y, color=cluster),size=3) + geom_polygon(data=h,aes(x=x,y=y,group=cluster,fill=as.factor(cluster)), alpha=0.4,linetype=0) + theme(legend.position = "none") Plotting clustering with random data

  12. Evaluating models – clustering models #### 5.3 Calculating the size of each cluster #### table(d$cluster) #### 5.4 Calculating the typical distance between items in every pair of clusters #### #install.packages("reshape2") library('reshape2') n <- dim(d)[[1]] pairs <- data.frame( ca = as.vector(outer(1:n,1:n,function(a,b) d[a,'cluster'])), cb = as.vector(outer(1:n,1:n,function(a,b) d[b,'cluster'])), dist = as.vector(outer(1:n,1:n,function(a,b) sqrt((d[a,'x']-d[b,'x'])^2 + (d[a,'y']-d[b,'y'])^2))) ) dcast(pairs,ca~cb,value.var='dist',mean) Intra-cluster distances versus Cross-cluster distances

  13. Validating models • Common model problem • Overfitting

  14. Validating models • Ensuring model quality • Testing on held-out data • K-fold cross-validation • Significance testing • Confidence intervals • Using statistical terminology

  15. 감사합니다.

More Related