search

gangchill / nip-convnet

Convolutional Autoencoder implemented in Tensorflow
Apache License 2.0
2 stars 3 forks source link

Create demo files #11

Open maxkohlbrenner opened 7 years ago

maxkohlbrenner commented 7 years ago

Currently I am getting mixed results for the CAE training again because I was experimenting a lot with the parameter settings. We already had a CNN with ReLUs and a CAE with sigmoid functions. TODO: write demo scripts for both a CNN and a CAE that demonstrate working parameter settings

maxkohlbrenner commented 7 years ago

Setting for the CAE that seems to work is:

in the train_...py : filter_dims = [(5,5), (5,5)] hidden_channels = [8, 16] pooling_type = 'strided_conv' strides = None # other strides should not work yet activation_function = 'sigmoid' batch_size = 100 max_iterations = 1000 chk_iterations = 100 step_size = 0.0001 tie_conv_weights = False

in the autoencoder file: self.weight_init_stddev = 0.00015 self.weight_init_mean = 0.0001 self.initial_bias_value = 0.0001

reaches reconstruction error of it 900 avg_re 0.010880947113 however, seems like it needs more regularization.