I got a warning "A batch with no more than one sample has been given to F.batch_normalization. F.batch_normalization will always output a zero tensor for such batches" when trying to predict a single a image.
I am using Resnet50 model (binary classification output) with Chainer, when training, the accuracy of both train and test set is > 98%. But when i tried to predict a single image, the accuracy is only about 50%.
I guess I have to predict images in batch (because Batch Normalization) to avoid warning and get high accuracy, right? Is it possible to predict single image?
P/s : I tried chainer.using_config('train', False) in prediction script but it didn't help.
Do you set chainer.config.train = False correctly? Note that chainer.using_config is a context manager.
with chainer.using_config('train', False):
do_something()
is (almost) equivalent to:
chainer.config.train = False
try:
do_something()
finally:
chainer.config.train = True
See also: https://docs.chainer.org/en/stable/reference/configuration.html#changing-configuration
Could you please upload the full stack-trace?
I'm very interested in where from the error was raised.
If it is from "chainer/links/normalization/batch_normalization.py", line 271~273, the error can be attributed to chainer.config.train == True.
In other cases, something wrong is happening.
Or, you can try
print("chainer.config.train")
just before your call of L.BatchNormalization.
Related
I'm trying to make a basic markov model using the package msm and things were working fine until I've suddenly started receiving the following error code. I don't know why it's suddenly started throwing this as it was working fine earlier, and I don't think I've changed anything. The error code seems to be pointing to the linear algebra library but I don't know what to do with it exactly ...
Error in balance(baP$z, "S") :
BLAS/LAPACK routine 'DGEBAL' gave error code -3
The code is as follows:
statesDistMatrix2 <- matrix(c(.1,0,0,.1), nrow = 2, ncol = 2)
msm1 <- msm(error ~ stop_datetime, subject = TRIP_ID, data = train_245_mk,
qmatrix = statesDistMatrix2, control=list(fnscale=5000,maxit=500))
From this document http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.230.5929&rep=rep1&type=pdf about LAPACK
It seems that (section Error Indicators and Warnings)
"Errors or warnings detected by the routine: INFO < 0
If INFO = -i, the ith parameter had an illegal value."
Still from this document, the third parameter which seems to be the problem in your case has to be real. Chances are that some complex value appears at some point in the process. Make sure that this parameter only has real values. Sometimes, some operations can lead to results such as 1 + 0.000000001j and this is problematic, try to call the function by taking only the real part of the argument you pass in.
Hope this helps.
I am using MXnet for training a CNN (in R) and I can train the model without any error with the following code:
model <- mx.model.FeedForward.create(symbol=network,
X=train.iter,
ctx=mx.gpu(0),
num.round=20,
array.batch.size=batch.size,
learning.rate=0.1,
momentum=0.1,
eval.metric=mx.metric.accuracy,
wd=0.001,
batch.end.callback=mx.callback.log.speedometer(batch.size, frequency = 100)
)
But as this process is time-consuming, I run it on a server during the night and I want to save the model for the purpose of using it after finishing the training.
I used:
save(list = ls(), file="mymodel.RData")
and
mx.model.save("mymodel", 10)
But none of them can save the model! for example when I load the "mymodel.RData", I can not predict the labels for the test set!
Another example is when I load the "mymodel.RData" and try to plot it with the following code:
graph.viz(model$symbol$as.json())
I get the following error:
Error in model$symbol$as.json() : external pointer is not valid
Can anybody give me a solution for saving and then loading this model for future use?
Thanks
You can save the model by
model <- mx.model.FeedForward.create(symbol=network,
X=train.iter,
ctx=mx.gpu(0),
num.round=20,
array.batch.size=batch.size,
learning.rate=0.1,
momentum=0.1,
eval.metric=mx.metric.accuracy,
wd=0.001,
epoch.end.callback=mx.callback.save.checkpoint("model_prefix")
batch.end.callback=mx.callback.log.speedometer(batch.size, frequency = 100)
)
A mxnet model is an R list, but its first component is not an R object but a C++ pointer and can't be saved and reloaded as an R object. Therefore, the model needs to be serialized to behave as an actual R object. The serialized object is also a list, but its first object is a text string containing model information.
To save a model:
modelR <- mx.serialize(model)
save(modelR, file="~/model1.RData")
To retrieve it and use it again:
load("~/model1.RData", verbose=TRUE)
model <- mx.unserialize(modelR)
The best practice for saving a snapshot of your training progress is to use save_snapshot (http://mxnet.io/api/python/module.html#mxnet.module.Module.save_checkpoint) as part of the callback after every epoch training. In R the equivalent command is probably mx.callback.save.checkpoint, but I'm not using R and not sure about it usage.
Using these snapshots can also allow you to take advantage of the low cost option of using AWS Spot market (https://aws.amazon.com/ec2/spot/pricing/ ), which for example now offers and instance with 16 K80 GPUs for $3.8/hour compare to the on-demand price of $14.4. Such 80%-90% discount is common in the spot market and can optimize the speed and cost of your training, as long as you use these snapshots correctly.
At first, let's create some sample categorical data with 3 levels.
y<-sample(c("A","B","C"),50,replace=TRUE)
I'm trying to formulate a Bayesian statistical model in which the y variable follows categorical distribution with parameters theta1,theta2,theta3. These parameters describe the probability a single y[i] belongs to the corresponding category. In the bayesian perspective, these parameters are also random variables and we use to assign a dirichlet prior to them with hyper-parameters alpha1,alpha2,alpha3.
I'm having some problems with the syntax as it seems.
CODE
model<-function(){
#likelihood
for( i in 1:N){
y[i]~ dcat(theta[])
}
#prior
theta[1:3]~ ddirch(alpha[])
}
library(R2OpenBUGS)
model.file <- file.path(tempdir(),"model.txt")
write.model(model, model.file)
y<-sample(c("A","B","C"),50,replace=TRUE)
N<-50
alpha<-c(1,1,1)
data<-list('y','N','alpha')
params<-c('theta')
inits<-function(){theta=c(1/3,1/3,1/3)}
We call OpenBUGS through R, with the bugs function
out<-bugs(data,inits,params,model.file,n.chains = 2
,n.iter=6000,codaPkg = TRUE,n.burnin = 1000,DIC = TRUE)
I've tried different ways to syntactically formulate the above code, dribbling through the errors and getting familiar with the log.txt file (that is the file that keeps the OpenBUGS output) until this code gave me a log.txt with no errors while R still has problems.
R output
Error in bugs.run(n.burnin, OpenBUGS.pgm, debug = debug, WINE = WINE, :
Look at the log file in /tmp/Rtmpofdk0t and
try again with 'debug=TRUE' to figure out what went wrong within OpenBUGS.
In addition: Warning message:
In FUN(X[[i]], ...) : class of 'x' was discarded
log.txt
OpenBUGS version 3.2.3 rev 1012
model is syntactically correct
data loaded
model compiled
initial values generated, model initialized
1000 updates took 0 s
monitor set
monitor set
monitor set
monitor set
deviance set
Thanks in advance for your help
I think you should rename theta1, theta2, theta3 with alpha1, alpha2, alpha3, because you use the alpha1,... in the function ddirch, but you never declare them. Instead you declare theta1 and so on, but never use them.
If there are any other issues, you might have a look at the log file, like the compiler suggests.
After numerous experiments, i figured out that for some reason OpenBUGS cant accept factor variables given as usual. So i changed my data ( format "A","B","C") to numeric (format 1,2,3) with the as.numeric R function and everything ran smoothly!
I am running a script that loops over a list of stock pair combinations... occasionally the script stops running due to an error generated by differing data lengths between pair combo and I simply remove the mismatched stock from consideration):
Error in model.frame.default(formula = stckY ~ stckX + 0, drop.unused.levels = TRUE) :
variable lengths differ (found for 'stckX')
Is there any way I can make R / Rstudio play a sound when the error message occurs so that I can be alerted without having to keep my eyes on the screen while the script is looping along?
I can generate sounds linearly using:
beep <- function(n = 3){
for(i in seq(n)){
system("rundll32 user32.dll,MessageBeep -1")
Sys.sleep(.5)
}
}
beep()
but how can I do this conditional on an error message?
Building on #frankc answer and #hrbrmstr comment, a way to do this:
install.packages("beepr")
library(beepr)
options(error = beep)
try options(error = beep)
you would still need to define beep before you do this. Haven't verified this works but it should per ?options:
'error': either a function or an expression governing the handling
of non-catastrophic errors such as those generated by 'stop'
as well as by signals and internally detected errors. If the
option is a function, a call to that function, with no
arguments, is generated as the expression. The default value
is 'NULL': see 'stop' for the behaviour in that case. The
functions 'dump.frames' and 'recover' provide alternatives
that allow post-mortem debugging. Note that these need to
specified as e.g. 'options(error = utils::recover)' in
startup files such as '.Rprofile'.
I am trying to use bestglm on a logistic regression model fitting a smaller sample of my data (100x25) matrix (originaly 250kx40).
I run it in BATCH mode. and I wish to write the bestglm models to a file.
dput(test2,file="test2.txt",control=c("keepNA","keepInteger","showAttributes"))
t2<-dget(file="test2.txt")
Error in UseMethod("family") :
no applicable method for 'family' applied to an object of class "NULL"
Any ideas as to how to solve this problem?
There is something in the dput'ing of the bestglm that isn't right. If you simply dput to the console, copy+paste it back in to a new object, it will fail with the same error.
It may be a bug and you may want to reach out to the package maintainer(s).
In the meantime, try using saveRDS and readRDS:
saveRDS(test2, file="test2.rds")
t2 <- readRDS("test2.rds")