gastonstat / discriminer Goto Github PK

There is an error in linDA() when priors are specified and validation = "crossval" is specified.

set.seed(1)
num_class <- length(unique(iris$Species))
# Works
DiscriMiner::linDA(variables = iris[1:4], group = iris$Species, prior = rep(1/num_class, num_class))
# Error
DiscriMiner::linDA(variables = iris[1:4], group = iris$Species, prior = rep(1/num_class, num_class), validation = "crossval")
# Error: object 'props' not found

In looking at the source code for linDA() the problem seems to be that in the validation == "crossval" section of the code it references the object props (line 154), but if priors are specified (lines 101-111) the object props is not created, unlike if prior is no specified (lines 111-115). I did some testing and adding
props = prior at line 111 seems to resolve the issue. See my pull request to resolve the issue.

The commands written in the installation section didn't work:

• Package has been removed from CRAN
• install_github('DiscriMiner', username='gastonstat') gave an error "Invalid git repo specification: 'DiscriMiner'"

The command that did work was:

install_github('gastonstat/DiscriMiner')

On R 4.2.3 with devtools 2.4.5

Dear Dr. Gaston Sanchez:
Do you know the discriminant Q2 (DQ2) which is taken as an improvement for the Q2 value used in the validation of PLSDA models (http://link.springer.com/article/10.1007%2Fs11306-008-0126-2). Can this method be added into package DiscriMiner?
Best regards!
Bo

set.seed(3)
data(Sonar)
data = Sonar
class.col = 61
trainIds = sample(seq_row(data), size=0.66*nrow(data))
model = plsDA(variables=data[trainIds,-class.col], group=data[trainIds,class.col])
#Error in Wh[, i] : incorrect number of dimensions.

Depending on the seed it sometimes also works without problems.

Dear Gaston,

I'm trying to use the function "plsDA" of your package, using the following command :
model <- plsDA(data[,-1],data[, 1],autosel = TRUE, cv = "LKO", k=10)
where the first column of "data" contains the groups (two groups).
But I get the following error message :
Error in if (sum(w.dif^2) < 1e-06 || iter == 100) break :
missing value where TRUE/FALSE needed

Could you help me to find out what the problem is ?

Kind regards,
Arnaud.

quaDA(..., prob = TRUE) doesn't seem to have an influence on the result of classify

Hi,
When I run the following example code in DiscriMiner package:
data(iris)
my_pls1 = plsDA(iris[,1:4], iris$Species, autosel=FALSE, comps=2)

The Q2 output is below:
my_pls1$Q2
Q2.setosa Q2.versicolor Q2.virginica Q2.global
t1 0.8594805 0.03936045 0.5009850 0.46660863
t2 0.2133430 0.16897758 0.1028906 -0.01269048

1. Could you please tell me how to explain the result?
2. The value of Q2.global for t2 is -0.01269048, what's the meaning?
   Best regards!
   Bo

When I try to fit a model using only one variable I receive an error message. This happens for linDA, quaDA, geoDA, plsDA.

set.seed(1)
DiscriMiner::linDA(variables = iris[1], group = iris$Species)
# Error in matrix(0, ng, ncol(X)) : non-numeric matrix extent

DiscriMiner::quaDA(variables = iris[1], group = iris$Species)
# Error in matrix(0, ng, ncol(X)) : non-numeric matrix extent

DiscriMiner::geoDA(variables = iris[1], group = iris$Species)
# Error in matrix(0, ng, ncol(X)) : non-numeric matrix extent

DiscriMiner::plsDA(variables = iris[1], group = iris$Species, autosel = F, comps = 2)
# Error in if (nc == n) nc = n - 1 : argument is of length zero

In geoDA.R, lines 99-102:

    if (any(learn) <= 0 || any(learn) > n)
      stop("\nsubscript out of bounds in 'learn' set")
    if (any(test) <= 0 || any(test) > n)
      stop("\nsubscript out of bounds in 'test' set")

It should be any(learn <= 0) || any(learn > n) and similarly any(test <= 0) || any(test > n). The parentheses are wrong.