(build-classify-dbn layers classes)
Build a Deep Belief Network using Restricted Boltzmann Machines designed to classify an observation.
See build-dbn
for layers usage. classes is the number of possible
classes the observation could be.
Build a Deep Belief Network using Restricted Boltzmann Machines designed to classify an observation. See `build-dbn` for layers usage. classes is the number of possible classes the observation could be.
(build-dbn layers)
Build a Deep Belief Network composed of Restricted Boltzmann Machines.
layers is a vector of nodes in each layer, starting with the visible layer.
Ex: [784 500 500 2000] -> 784-500 RBM, a 500-500 RBM, and a 500-2000 RBM
Build a Deep Belief Network composed of Restricted Boltzmann Machines. layers is a vector of nodes in each layer, starting with the visible layer. Ex: [784 500 500 2000] -> 784-500 RBM, a 500-500 RBM, and a 500-2000 RBM
(classify-obv dbn obv)
Given a DBN and a single observation, return the model's prediction.
Given a DBN and a single observation, return the model's prediction.
(test-dbn dbn dataset)
Test a classification Deep Belief Network on a given dataset.
The dataset should have the label as the last entry in each observation.
Test a classification Deep Belief Network on a given dataset. The dataset should have the label as the last entry in each observation.
(train-classify-dbn dbn dataset params)
Train a Deep Belief Network designed to classify data vectors.
dataset is a softmax-labeled dataset, in the same format as that produced by deebn.mnist/load-data-with-softmax (the softmax precedes the data vector).
Check train-rbm and train-dbn for more information about parameters.
Train a Deep Belief Network designed to classify data vectors. dataset is a softmax-labeled dataset, in the same format as that produced by deebn.mnist/load-data-with-softmax (the softmax precedes the data vector). Check train-rbm and train-dbn for more information about parameters.
(train-dbn dbn dataset params)
Train a generative Deep Belief Network on a dataset. This trained model doesn't have an inherent value, unless the trained weights are subsequently used to initialize another network, e.g. a simple feedforward neural network.
dataset is an unlabeled dataset used for unsupervised training.
mean-field? is a key in the params map, and is a boolean indicating whether to use the expected value from a hidden layer as the input to the next RBM in the network, or use the sampled binary value. Defaults to true.
query-final? is a key in the params map, and is used to determine if the final RBm trained is queried for the state of its hidden layer. This is only used when training the generative layers of a classification DBN, and changes the return type (both the trained DBN and the final transformed dataset are returned if this is true).
See train-rbm
for details on hyper-parameters passed in the param map.
Train a generative Deep Belief Network on a dataset. This trained model doesn't have an inherent value, unless the trained weights are subsequently used to initialize another network, e.g. a simple feedforward neural network. dataset is an unlabeled dataset used for unsupervised training. mean-field? is a key in the params map, and is a boolean indicating whether to use the expected value from a hidden layer as the input to the next RBM in the network, or use the sampled binary value. Defaults to true. query-final? is a key in the params map, and is used to determine if the final RBm trained is queried for the state of its hidden layer. This is only used when training the generative layers of a classification DBN, and changes the return type (both the trained DBN and the final transformed dataset are returned if this is true). See `train-rbm` for details on hyper-parameters passed in the param map.
cljdoc is a website building & hosting documentation for Clojure/Script libraries
× close