This repository was archived by the owner on Jan 7, 2025. It is now read-only.
This repository was archived by the owner on Jan 7, 2025. It is now read-only.
Training an autoencoder #117
Description
Hi,
I am trying to pretrain an auto-encoder (similar to the one specified in caffe mnist). I have tthe last 3 layers as follows:
layer {
name: "loss"
type: "SigmoidCrossEntropyLoss"
bottom: "decode1"
bottom: "flatdata"
top: "cross_entropy_loss"
loss_weight: 1
}
layer {
name: "decode1neuron"
type: "Sigmoid"
bottom: "decode1"
top: "decode1neuron"
}
layer {
name: "loss"
type: "EuclideanLoss"
bottom: "decode1neuron"
bottom: "flatdata"
top: "l2_error"
loss_weight: 0
}
Although these two loss layers are specified, when I start training, it immediately fails with this error message:
Traceback (most recent call last):
File "/home/mainak/digits/digits/scheduler.py", line 394, in task_thread
task.run(**options)
File "/home/mainak/digits/digits/task.py", line 161, in run
self.before_run()
File "/home/mainak/digits/digits/model/tasks/caffe_train.py", line 98, in before_run
self.save_prototxt_files()
File "/home/mainak/digits/digits/model/tasks/caffe_train.py", line 158, in save_prototxt_files
assert len(loss_layers) > 0, 'must specify a loss layer'
AssertionError: must specify a loss layer
Any idea where I might be going wrong? Also, I believe DIGITS is not whining about a loss layer after removing it himself? Also, is it possible at all to pre-train an auto-encoder in DIGITS?
Thanks in advance!