RBM_DBN

Implement RBM (Restricted Boltzmann machine), DBN (Deep belief network) and DNN (Deep neural network) from scratch.

Gained a lot for the whole process of peoject: construct the model; train, save and deploy the model; manage the logging information; write the config files.

How to use

• folder logs saves the logging files;
• models contains the pretrained DNN models;
• the images generated by RBM and DBN are saved in the folder outputs;
• environement.yml provides the required Python package dependencies;
• logging.conf provides the logging configurations;
• there are some useful functions in utils.py;
• principal_RBM_alpha.py, principal_DBN_alpha.py, principal_DNN_MNIST.py are three main scripts. You can use them in this way (you can also find them in commands.txt):

python principal_RBM_alpha.py --n_characters 12 --n_neurons 200 300

python principal_DBN_alpha.py --n_characters 1 --n_neurons 300 --n_layers 2 3

python principal_DNN_MNIST.py --n_samples 1000 --epochs 2 --neurons_per_layer 200 --n_layers 2 --no-pretraining

You can find more details through the argument -h. For example: python principal_DNN_MNIST.py -h.

• automatic_call.py allows you to automatically call three above scripts multiple times (in case of doing tests).