https://github.com/locuslab/monotonedbm

https://github.com/locuslab/monotonedbm

Last synced: over 1 year ago
JSON representation

• Host: GitHub
• URL: https://github.com/locuslab/monotonedbm
• Owner: locuslab
• Created: 2023-04-07T21:07:41.000Z (about 3 years ago)
• Default Branch: main
• Last Pushed: 2023-04-07T21:33:43.000Z (about 3 years ago)
• Last Synced: 2025-01-07T06:47:28.561Z (over 1 year ago)
• Language: Python
• Size: 9.77 KB
• Stars: 2
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
This folder includes the code for our monotone mean-field DEQ implementation.

-- Required package: torch, numpy, matplotlib, tqdm

-- Structure of the code: 

	-- proxsoftmax.py includes the implementation for our prox_f^alpha for alpha between 0 and 1. Notice that one cannot set alpha=1 here.

	-- multitier_conv.py includes the multi-tier structure of the DEQ, in the form of convolutions (notice that even dense linear layers can be re-written as convolutions)

		The most important arguments are 

		sizes = [(num_classes, 28, 28, 1),

         (40, 14, 14, 10,),

         (80, 7, 7, 20),

         (10, 1)]

		kernels = np.array([[3, 0, 0, 0],

		                    [3, 3, 0, 0],

		                    [3, 3, 3, 0],

		                    [0, 0, 1, 1]])

		Here 'sizes' specifies the (input_channel, input_height, input_width, input_groups) for convolutions, or (input_size, input_groups) for linear layers. 'kernels' specifies the connection and kernel sizes.

	-- deq_model.py includes the implementation for the forward-backward splitting. The most high-level class is 'ConvDeqCrf', which will be our model. 

	-- util.py includes utility functions like data preparation.

	-- train_and_eval.py includes training and evaluation. 

	-- main.py. The main entrance for running this code. One can directly run 'python -i main.py'. The variable 'num_classes' indicates the number of bins we use for discretizing the input intensity. For example, if num_classes=2 and the input 	intensity of MNIST image pixel at position (0,0) is x, then the binned output will be 0 if x<0.5, and 1 otherwise.

		MON_DEFAULTS = {

		    'alpha': 0.125,

		    'tol': 1e-2,

		    'max_iter': 50

		}

		specifies the arguments for the forward-backward splitting. 'alpha' is the alpha used in damped iteration and prox_f^alpha, 'tol' is the relative tolerance at which we stop the splitting, and 'max_iter' is the maximum number of (anderson) iterations.