# RNN package and problems with “Predictr”

I have two questions about how to use R's RNN package, specifically the trainr and predictr functions.

Let's suppose I have a time series of 4000 steps for 5 different variables.

1. How should this be passed on to the RNN, since the input has to be a 3D array? Should it be an array of dim(1, 4000, 5) or dim(4000,1,5) or something completely different?

Moreover, let's suppose I successfully train my RNN and would like to make some predictions using the predictr function.

1. Which dimension will the output have and how can it be interpreted?

Until now I have tried to input the data as 3D array (4000, 1, 5) with the following code:

RecNN <- trainr(TrainR, TrainI,
learningrate = 0.003, momentum = 0.003, hidden_dim = c(100),
network_type = "rnn", batch_size = 50,
numepochs = 100, sigmoid = "logistic", use_bias = TRUE)

print("--- DONE LEARNING ---")


And then used the predictr function with the input (250, 1, 5):

results1 <- predictr(RecNN, TestI, hidden = TRUE, real_output = TRUE)

List of 2
$$: num [1:250, 1, 1:100] 0 0 0 0 0 0 0 0 0 0 ...$$ : num [1:250, 1, 1:2] 1 1 1 1 1 1 1 1 1 1 ...


Thanks a lot!

• What code did you try and what error did you get? It will be helpful if you post your attempt here. – rnso Dec 5 at 13:59
• I have added the code to the question now! – Alex Dec 5 at 14:04
• If there is no error and output is sensible, this should be the correct approach. – rnso Dec 5 at 14:12
• Ok, but still 1. How would be the correct way of passing the input during training? An array of dim(1, 4000, 5) or dim(4000,1,5)? And 2. I am not sure how to interpret the output, why are there two lists of different lengths? – Alex Dec 5 at 14:17
• Check this and other posts on google search: quora.com/… – rnso Dec 5 at 14:23

The manual gives needed dimensions (see help on trainr function):

Dimensions of input X need to be as:

dim 1: samples
dim 2: time
dim 3: variables


and dimensions of output Y should be:

dim 1: samples (must be equal to dim 1 of X)
dim 2: time (must be equal to dim 2 of X)
dim 3: variables (could be 1 or more, if a matrix, will be coerce to array)