szcf-weiya/ESL-CN

SRBCT microarray data

szcf-weiya opened this issue · 4 comments

szcf-weiya commented

SRBCT microarray data

SBRCT gene expression data. 2318 genes, 63 training samples,
25 test samples.

One gene per row, one sample per column

Cancer classes are labelled 1,2,3,4 for c("EWS","RMS","NB","BL")

Files

  • Training set gene expression: khan.xtrain.txt
  • Training set class labels: khan.ytrain.txt
  • Test set gene expression: khan.xtest.txt
  • Test set class labels : khan.xtest.txt
szcf-weiya commented

diagonal LDA

p652 or ESL CN
image
The original text claims that

Here the diagonal LDA classifier yielded five misclassification errors for the 20 test samples.

As you can see in the above frequency table, there are 7 misclassification errors among 20 test samples (the NA samples are excluded), roughly the same performance.

szcf-weiya commented

Regularized Diagonal LDA (with Delta = 2.0)

image
Zero training error and only 1 misclassification errors among 20 test samples (exclude NA samples), roughly agree with the top panel of Fig. 18.4

szcf-weiya commented

remove NA

image

szcf-weiya commented

Error curves (Fig. 18.4 top)

error_curves
Roughly reproduce the original figure, the cv error might be different since the division of folds.

Tips related to the plot. Cannot find the twiny command in plot.jl, although it does exist a twinx command, which is a bonus-feature and is not described in the docs. JuliaPlots/Plots.jl#337
Then I resorted to the pyplot package.