nn.JoinTable() in new dimension?

[Johnny ho]

Let say I have a table of 3 tenor.

{

1: double Tensor of 100x20x30

2:double Tensor of 100x20x30

3:double Tensor of 100x20x30

}

I want to join these tensors and make a tensor of 3x100x20x30. I know nn.JoinTable can do something like 300x20x30 (in dim 1), but it cannot create a new dimension, any ideas on how to do this?

[alban desmaison]

You should use view to add one dimension to your tensor to get 1x100x20x30 so you can use the nn.JoinTable module.
In this case you can either use the view function on your tensor before feeding them to the network https://github.com/torch/torch7/blob/master/doc/tensor.md#result-viewresult-tensor-sizes

Or you can use the nn.View module in your network to add a dimension https://github.com/torch/nn/blob/master/doc/simple.md#nn.View

[Johnny ho]

Thanks! I will try that.

[liangjz]

what if i have a 2 dims tensor T[m,n], T want to turn it into T[1,m,n], but i don't know what m,n  are, can i finish this operate by applying a simple nn layer ?

在 2015年11月3日星期二 UTC+8下午5:06:16，alban desmaison写道：

[abc123]

You can use nn.Replicate() if you don't know the dimensions apriori.
nn.Replicate(1,1) will work.