Hi
I am using jeffdonahue 's lstm right now
I wan to ask if my input layer is like this
input:"data"
input_shape:{dim:10 dim:100 dim:1}
input:"indicator"
input_shape:{dim:10 dim:100}
meaning that I have 100 independent sequenceseach have 10 time steps
and for each time step
there is a corresponding output(label)
and the rest of the net is
layer {
name: "lstm1"
type: "LSTM"
bottom: "data"
bottom: "indicator"
top: "lstm1"
recurrent_param{
num_output: 20
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
value: 1
}
}
}
layer {
name: "predict"
type: "InnerProduct"
bottom: "lstm1"
top: "ip1"
inner_product_param {
num_output: 1
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
value: 1
}
}
}
layer {
name: "loss"
type: "EuclideanLoss"
bottom: "ip1"
bottom: "label"
top: "loss"
#include: { phase: TRAIN }
}
"label" is the target outputin this case
how do I setting the "label" data so each training date(each step in each sequence)
can have its corresponding target output(label)?
thanks!