The new model-wide rng parameter is only propagated to graph.Network when using recurrent.Regressor and feedforward.Regressor. Classifiers and Autoencoders do not pass the parameter in their super-calls. At least it seems like it to me because both models give this error:
net = theanets.Autoencoder([784, args.features \ 2, 784], rng=42)
TypeError: init() got an unexpected keyword argument 'rng'
The new model-wide rng parameter is only propagated to graph.Network when using recurrent.Regressor and feedforward.Regressor. Classifiers and Autoencoders do not pass the parameter in their super-calls. At least it seems like it to me because both models give this error:
net = theanets.Autoencoder([784, args.features \ 2, 784], rng=42) TypeError: init() got an unexpected keyword argument 'rng'