glm problem

Hello,

I have a weird problem here.  What I want to do is that I need to draw 1000
samples from a matrix, and use glm on them.

when I used this command, it runs without the problem

qdata.glm = glm(X258 ~ ., family = binomial, data =

q2data[sample(dim(q2data)[1], 1000), ])


but if I drew the sample first and run glm() on that sample, it gave a
warning.

qdata.sample = q2data[sample(dim(qdata)[1], 1000), ]

qdata.glm = glm(X258 ~ ., family = binomial, qdata.sample)

Warning message:
In glm.fit(x = X, y = Y, weights = weights, start = start, etastart =
etastart,  :
  fitted probabilities numerically 0 or 1 occurred

Because I need to use that same sample later, I have to save it in
qdata.sample. therefore the first command does not work in my case because I
cannot actually get the same 1000 sample after that.

anyone knows how to solve this?

Thanks!

	[[alternative HTML version deleted]]

Thanks for the reply, that was a typo only in the post. as i was trying to
remove some 'sensitive names'.  In my real work,  i made sure that

1) this command runs without problem
qdata.glm = glm(X258 ~ ., family = binomial, data = qdata[
sample(dim(qdata)[1], 1000), ] )



2) these two gave a warning
qdata.sample = qdata[sample(dim(qdata)[1], 1000), ]
qdata.glm = glm(X258 ~ ., family = binomial, qdata.sample)

Warning message:
In glm.fit(x = X, y = Y, weights = weights, start = start, etastart =
etastart,  :
 fitted probabilities numerically 0 or 1 occurred

Can you explain why this is?

Thanks!




On Sat, Feb 14, 2009 at 8:09 AM, Duncan Murdoch <murdoch at stats.uwo.ca>wrote:

On 13/02/2009 11:59 PM, Roger wrote:

Hello,

I have a weird problem here.  What I want to do is that I need to draw
1000
samples from a matrix, and use glm on them.

when I used this command, it runs without the problem

qdata.glm = glm(X258 ~ ., family = binomial, data =

q2data[sample(dim(q2data)[1], 1000), ])


but if I drew the sample first and run glm() on that sample, it gave a
warning.

qdata.sample = q2data[sample(dim(qdata)[1], 1000), ]

This looks like a typo:  you took the dim of the wrong thing here.

Duncan Murdoch

 qdata.glm = glm(X258 ~ ., family = binomial, qdata.sample)

Warning message:
In glm.fit(x = X, y = Y, weights = weights, start = start, etastart =
etastart,  :
 fitted probabilities numerically 0 or 1 occurred

Because I need to use that same sample later, I have to save it in
qdata.sample. therefore the first command does not work in my case because
I
cannot actually get the same 1000 sample after that.

anyone knows how to solve this?

Thanks!

       [[alternative HTML version deleted]]

	[[alternative HTML version deleted]]