Open TNick opened 9 years ago
Yes, that looks like a good idea, a PR would be welcome. Thanks!
softmax_b
and softmax_W
too?
$>grep -R softmax_b
models/dbm/layer.py: self.b = sharedX( np.zeros((n_classes,)), name = 'softmax_b')
models/mlp.py: name='softmax_b')
scripts/tutorials/convolutional_network/convolutional_network.ipynb: "\tsoftmax_b: 0.00999999977648\n"
scripts/tutorials/jobman_integration.ipynb: "\tsoftmax_b: 0.000205\r\n",
scripts/tutorials/multilayer_perceptron/multilayer_perceptron.ipynb: "\tsoftmax_b: 0.00999999977648\n"
scripts/tutorials/multilayer_perceptron/multilayer_perceptron.ipynb: "\tsoftmax_b: 0.00999999977648\n"
scripts/tutorials/stacked_autoencoders/stacked_autoencoders.ipynb: "\tsoftmax_b: 0.0500000007451\n"
$>grep -R softmax_W
models/dbm/layer.py: self.W = sharedX(W, 'softmax_W' )
models/mlp.py: self.W = sharedX(W, 'softmax_W')
scripts/tutorials/convolutional_network/convolutional_network.ipynb: "\tsoftmax_W: 0.00999999977648\n"
scripts/tutorials/jobman_integration.ipynb: "\tsoftmax_W: 0.000205\r\n"
scripts/tutorials/multilayer_perceptron/multilayer_perceptron.ipynb: "\tsoftmax_W: 0.00999999977648\n"
scripts/tutorials/multilayer_perceptron/multilayer_perceptron.ipynb: "\tsoftmax_W: 0.00999999977648\n"
scripts/tutorials/stacked_autoencoders/stacked_autoencoders.ipynb: "\tsoftmax_W: 0.0500000007451\n"
I think consistency everywhere would be welcome.
A network with three
maxout.MaxoutLocalC01B
, amaxout.Maxout
and amlp.Softmax
printsParameter and initial learning rate summary
like so:This is all but consistent. I like the
layer.name + '_' + W/b
notation. It is already used in most parts of the library, asgrep -R "\.name = "
shows.Would you accept a PR with that? If not, please suggest other naming convention and I would be happy to write a PR.