This is an implementation of the longitudinal data neural network package in R. The package is uploaded in Github.
Regressions coefficients, fitted values, residuals, degrees of freedom, residual variance, variance of the regression coefficients and t-values for each coefficient were calculated using linear algebra equations.
example<-create_model(rnn_inputs = c(20,24,24,24,16,16,16,16,16,15),recurrent_droppout =c(0.1,0.1,0.1,0.1,0.1,0.1,0.1,0.1,0.1,0.1),inputs = 232,layer_dropout = c(0.1,0.1),n_nodes_hidden_layers = c(1024,1024),loss_function = 'mean_squared_error',opt = 'adam',metric = 'mean_absolute_error')
rnn_inputs: The number of inputs (integers) per each LSTM (vector of length 10).
recurrent_droppout: The dropout to be applied in the LSTMs (between 0 and 1).
inputs: The number of inputs (integer) to be concatenated with the output of the LSTMs.
layer_dropout: The dropout to be applied between the hidden layers (between 0 and 1).
n_nodes_hidden_layers: The number of nodes in the hidden layers (2 in total).
loss_function: The loss function to be used.
opt: The optimizer to be used.
metric: The metric to be used.
example<-fit_model(model = model,ver = 0, n_epoch = 1,bsize = 32,X1 = X1, X2 = X2, X3 = X3,X4 = X4,X5 = X5,X6 = X6,X7 = X7,X8 = X8,X9 = X9,X10 = X10,Xif = Xif,y = y)
model: The model object produced by create_model().
ver: ver=0 to show nothing, ver=1 to show animated progress bar, ver=2 to just mention the number of epoch during training.
n_epoch: The number of epochs to train the model.
bsize: The batch size.
X1: Features as inputs of 1st LSTM.
X2: Features as inputs of 2nd LSTM.
X3: Features as inputs of 3rd LSTM.
X4: Features as inputs of 4th LSTM.
X5: Features as inputs of 5th LSTM.
X6: Features as inputs of 6th LSTM.
X7: Features as inputs of 7th LSTM.
X8: Features as inputs of 8th LSTM.
X9: Features as inputs of 9th LSTM.
X10: Features as inputs of 10th LSTM.
Xif: The features to be concatenated with the outputs of the LSTMs.
y: The target variable.
example<-evaluate_model(model = fitted_model,X1_test = X1_test, X2_test = X2_test, X3_test = X3_test,X4_test = X4_test,X5_test = X5_test,X6_test = X6_test,X7_test = X7_test,X8_test = X8_test,X9_test = X9_test,X10_test = X10_test,Xif_test = Xif_test,y_test = y_test,bsize = 32)
model: The fitted model object produced by create_model().
X1_test: Features as inputs of 1st LSTM.
X2_test: Features as inputs of 2nd LSTM.
X3_test: Features as inputs of 3rd LSTM.
X4_test: Features as inputs of 4th LSTM.
X5_test: Features as inputs of 5th LSTM.
X6_test: Features as inputs of 6th LSTM.
X7_test: Features as inputs of 7th LSTM.
X8_test: Features as inputs of 8th LSTM.
X9_test: Features as inputs of 9th LSTM.
X10_test: Features as inputs of 10th LSTM.
Xif_test: The features to be concatenated with the outputs of the LSTMs.
y_test: The target variable.
bsize: The batch size.
After installing the package run the following commands in an R session:
library(LDNN)
set.seed(12345)
#Train dummy data
X1 <- matrix(runif(500*20), nrow=500, ncol=20)
X2 <- matrix(runif(500*24), nrow=500, ncol=24)
X3 <- matrix(runif(500*24), nrow=500, ncol=24)
X4 <- matrix(runif(500*24), nrow=500, ncol=24)
X5 <- matrix(runif(500*16), nrow=500, ncol=16)
X6 <- matrix(runif(500*16), nrow=500, ncol=16)
X7 <- matrix(runif(500*16), nrow=500, ncol=16)
X8 <- matrix(runif(500*16), nrow=500, ncol=16)
X9 <- matrix(runif(500*16), nrow=500, ncol=16)
X10 <- matrix(runif(500*15), nrow=500, ncol=15)
Xif <- matrix(runif(500*232), nrow=500, ncol=232)
y <- matrix(runif(500), nrow=500, ncol=1)
#Test dummy data
X1_test <- matrix(runif(500*20), nrow=500, ncol=20)
X2_test <- matrix(runif(500*24), nrow=500, ncol=24)
X3_test <- matrix(runif(500*24), nrow=500, ncol=24)
X4_test <- matrix(runif(500*24), nrow=500, ncol=24)
X5_test <- matrix(runif(500*16), nrow=500, ncol=16)
X6_test <- matrix(runif(500*16), nrow=500, ncol=16)
X7_test <- matrix(runif(500*16), nrow=500, ncol=16)
X8_test <- matrix(runif(500*16), nrow=500, ncol=16)
X9_test <- matrix(runif(500*16), nrow=500, ncol=16)
X10_test <- matrix(runif(500*15), nrow=500, ncol=15)
Xif_test <- matrix(runif(500*232), nrow=500, ncol=232)
y_test <- matrix(runif(500), nrow=500, ncol=1)
#Create the model
model = create_model(rnn_inputs = c(20,24,24,24,16,16,16,16,16,15),
recurrent_droppout = c(0.1,0.1,0.1,0.1,0.1,0.1,0.1,0.1,0.1,0.1),
inputs = 232,
layer_dropout = c(0.1,0.1),
n_nodes_hidden_layers = c(1024,1024),
loss_function = 'mean_squared_error',
opt = 'adam',
metric = 'mean_absolute_error')
#Fit the model
fitted_model = fit_model(model = model,ver = 0, n_epoch = 1,bsize = 32,X1 = X1, X2 = X2, X3 = X3,X4 = X4,X5 = X5,X6 = X6,X7 = X7,X8 = X8,X9 = X9,X10 = X10,Xif = Xif,y = y)
#Evaluate the model on test data
evaluate_model(model = fitted_model,X1_test = X1_test, X2_test = X2_test, X3_test = X3_test,X4_test = X4_test,X5_test = X5_test,X6_test = X6_test,X7_test = X7_test,X8_test = X8_test,X9_test = X9_test,X10_test = X10_test,Xif_test = Xif_test,y_test = y_test,bsize = 32)