cbergmeir / rsnns Goto Github PK

Hi. I am making use of elman neural networks.

I want to use RELU FUNCTIONs on networks... but, there's no RELU Function on RSNNS

Which options could I use, if I wnat to use RELU Function?

If it is impossible, can I make a pull requests about RELU Function? (Because, I don't know how to use/test my c++ RELU code, and also dependencies...)

Please Let me know..! I really want your help.

krui_getSiteName chokes when using sites within the network and at least RadialBasisLearning due to incomplete link handling. Solution for learn_f.cpp:3724 as follows:

if (NetInitialize || LearnFuncHasChanged) {
  FOR_ALL_UNITS(unit_ptr) {
    if UNIT_HAS_SITES( unit_ptr ){
      FOR_ALL_SITES_AND_LINKS(unit_ptr, site_ptr, link_ptr) {
        link_ptr->value_a = 0.0;
      }
    }else{
      FOR_ALL_LINKS(unit_ptr, link_ptr) {
        link_ptr->value_a = 0.0;
      }
    }
  }
}

Also, the following pointer operations in kr_mem.cpp look kind of suspicious. Should post increments be used instead?

tmp_ptr = ++link_array;
tmp_ptr = ++site_array;

Hi,

Can you look into the following issue?

When I run mlp() I get the error 'terminate called after throwing an instance of 'Rcpp: not compatible' followed by a crash of R.

Thx.

Hello,

I would like to know why the results using Windows or Linux are different.

library(RSNNS)

set.seed(1)

data = matrix(rnorm(300), ncol = 3)
model = mlp(x = data[1:95, 1:2], y = data[1:95, 3], size = 5)
predict = predict(model, data5[96:100, 1:2])
predict
# Results using Windows:
#           [,1]
# 96  0.02506411
# 97  0.01911207
# 98  0.02082419
# 99  0.01711116
# 100 0.01743222

# Results using Linux:
#           [,1]
# 96  0.02582934
# 97  0.01976289
# 98  0.02150856
# 99  0.01771399
# 100 0.01804006

The R versions and the versions of package RSNNS are the same.

Best regards,
Marius

I'm using SNNS function loadNet(netfile) to recover network state. I can't find any equivalent in RSNNS.

Using data <-readPatFile(filename) function is too long. For example file with 45000 records using SNNS loadPattern(filename) function takes 1-2 sec. instead RSNNS takes 30min. Why?

Is it possible to use the RSNNS library to perform RBF interpolation when the target output is longer than the input?

There are 4 columns, "fitted" which stands for a value for each surf_x, surf_y, and surf_z value. surf_x etc. and all values of "outputs" all represent a coordinate in 3D space.

The goal is to use RBF interpolation to find the "fitted" value corresponding to the coordinates found in "outputs" where all values of surf_x in "inputs" can be found in column [,1] of "outputs".

library(RSNNS)

> dim(inputs)
[1] 2312    4

> head(inputs)
      fitted    surf_x    surf_y   surf_z
1 -2.0260418 -64.20550 -84.79310 219.4273
2  1.2358981 -67.00667 -79.65610 227.9487
3 -1.3667838 -62.88993 -86.83528 218.1500
4  0.1172459 -66.91177 -79.44738 227.0095
5  1.1278194 -67.30763 -80.56429 226.5518
6 -2.7412495 -62.88353 -85.40290 218.0784

> dim(outputs)
[1] 6971    3

> head(outputs)
        [,1]    [,2]    [,3]
[1,] -38.000 -41.794 180.858
[2,] -39.542 -50.167 175.664
[3,] -33.440 -56.959 173.340
[4,] -17.804 -60.643 181.561
[5,]  -8.156 -50.190 201.004
[6,] -23.019 -36.171 196.698

rbf_res <- RSNNS::rbf(x = inputs,
               y = outputs)

Error in checkInput(x, y) : nrows of 'x' and 'y' must match

Hello cbergmeri, When I installed package RSNNS, error occurs. The error is "make: *** [RSNNS.so] Error 1 ERROR: compilation failed for package 'RSNNS' ". I find RSNNS is compiled by clang. But I use gcc to compile it. Is this an reason for this question? I am looking forward to your reply. Thanks！！！

Dear Christoph,
I would like suggest a feature to add in your great package, that is cruelly missing in R version.
It would be nice to propose an initialization method copying the weights of an already trained network, that would allow to iteratively train a network with new examples.
Or even better: an input parameter in high level functions (mlp, elman ...) to pass an already existing network to update.
Thank you for telling me if i am missing this functionality. I actually didn't find a way to achieve this browsing R documentation of RSNNS.

Best Regards,
Nicolas

I noticed that biases are not applied in a conventional manner in a Multilayer Perceptron model. It is not clear how the feedforward step applies bies between the layer and their neurons.

Please, could you explain how they work?

Is it possible to assign new learning algorithm to the package, which is not already included like Tabu search or Simulated Annealing?

Hi,
I use the RSNNS package and today I noticed something strange. It always worked, however, today I used it and the forecast is very strange.

Do you know something? I am writing an article with this package.

require(RSNNS)
a<-jordan(scale(mtcars[,5]),mtcars[,1], size=8, learnFuncParams = c(0.1), maxit=1000,linOut = TRUE)
b<-predict(a,data.frame(scale(mtcars[,5])))
plot(mtcars[,1],type="l",col="blue",ylim = c(1,40))
lines(b,type="l",col="black")
summary(a)
weightMatrix(a)
a<-mlp(mtcars[,5],mtcars[,1], size=10, learnFuncParams = c(0.1), maxit=1000,linOut = TRUE)
b<-predict(a,data.frame(mtcars[,5]))
plot(mtcars[,1],type="l",col="blue",ylim = c(1,40))
lines(b,type="l",col="black")

Hello,
I am trying to install a package in sparkR and getting the error
unable to access index for repository http://mran.revolutionanalytics.com/snapshot/2015-05-01/src/contrib

The version of R i'm using is 3.2.0 and its throwing error for every new package installation.
I did try to download the package zip file and load it from there but that also not worked for me getting this as an error
‘randomForest’ is not a valid installed package.
Is there any solution?

when using, e.g.: size=c(2,2), the function exits with a seg fault, but should better report that Jordan networks with more than one hidden layer don't make much sense, as their feedback comes directly from the output layer.

library(RSNNS)

data(snnsData)
inputs <- snnsData$laser_1000.pat[,inputColumns(snnsData$laser_1000.pat)]
outputs <- snnsData$laser_1000.pat[,outputColumns(snnsData$laser_1000.pat)]

patterns <- splitForTrainingAndTest(inputs, outputs, ratio=0.15)

modelJordan <- jordan(patterns$inputsTrain, patterns$targetsTrain,
size=c(2,2), learnFuncParams=c(0.1), maxit=100,
inputsTest=patterns$inputsTest,
targetsTest=patterns$targetsTest, linOut=FALSE)

Dear developer,

How can I predict the value of hidden nodes in a mlp model?

For keras I can short the model upto a layer call 'dense_1' and make my predictions on 'ph.test':
layer_name <- 'dense_1';
chk_model <- keras_model(inputs=model$input, outputs=get_layer(model, layer_name)$output);
chko <- predict(chk_model, ph.test);

Thanks in advance,

Ad Denissen.

Hi again,

I had a quick question about the predict method for pruned neural networks. I'm getting some weird results when I predict values with new data from a pruned mlp network. Here's some code to illustrate the problem.

A plot of the pruned network...

library(RSNNS)
# devtools::install_github('fawda123/NeuralNetTools')
library(NeuralNetTools)
x <- neuraldat[, c('X1', 'X2', 'X3')]
y <- neuraldat[, 'Y1']

# pruned model using code from RSSNS pruning demo
pruneFuncParams <- list(max_pr_error_increase = 10.0, pr_accepted_error = 1.0, 
  no_of_pr_retrain_cycles = 1000, min_error_to_stop = 0.01, init_matrix_value = 1e-6, 
  input_pruning = TRUE, hidden_pruning = TRUE)
mod <- mlp(x, y, size = 5, pruneFunc = "OptimalBrainSurgeon", 
  pruneFuncParams = pruneFuncParams)

# plot shows that only X3 has weighted connections
plotnet(mod)

Now, when I give the pruned model some new data with a constant value for X3, why would the predicted response change? I would expect changing values for X1 and X2 to have no effect on model predictions because the connection weights are zero, whereas changing values for X3 would produce a response for Y because the connection weights are non-zero. Am I thinking about this correctly?

# ranges
# apply(x, 2, range)

# no obs for prediction data
n <- 10

# holding X3 constant gets a response
newdat <- data.frame(X1 = seq(0, 1, length = n), X2 = seq(0, 1, length = n), X3 = 0.5)
predict(mod, newdata = newdat)

# holding all but X3 constant returns a single value for the response

newdat <- data.frame(X1 = 0.5, X2 = 0.5, X3 = seq(0, 1, length = n))
predict(mod, newdata = newdat)

Any idea why this is happening?

Thanks,

Marcus

Hi there,

I'm the author of the NeuralNetTools package. I noticed your todo list for the next version of RSNNS links to an old gist of mine for plotting nnet objects. The CRAN release for NeuralNetTools has a plotnet method for mlp models and I just pushed a feature to the devel version to plot pruned mlp models. Thought you might be interested... I'll probably submit the latest release to CRAN pretty soon, so you could include it as a dependency in a future version of RSNNS.

Keep up the great work!

-Marcus

Why learnFuncParams=c(10, 0.1) in mlp command with BackpropBatch argument for learnFunc.
please reply asap.

I do not see where time-delay neural network is documented? Can you please be so kind and gracious enough to let me know where the tdnn is documented in your package documentation. If not, could you be so kind and gracious enough to briefly include an example of how to do this with the SNSS code please?

I have read an article about SNNS package from 2012. The authors original article is here. They mention on page 2 that several network architectures are implemented in SNSS, including time-delay neural networks. I believe the source code is located here. Unfortunately, when I look at the R package documentation I do not see this mentioned in the package api.