NNParamsPrinter.jl

NNParamsPrinter.jl is a simple package for printing neural network parameters in a readable format. Currently only supported for Lux.

Installation

using Pkg
Pkg.add("NNParamsPrinter")

Usage

using Lux
using NNParamsPrinter
using StableRNGs

U = Chain(
    Dense(1, 3, rbf),
    Dense(3, 3, rbf),
    Dense(3, 3, rbf),
    Dense(3, 1)
)

nn_init_params, snn = Lux.setup(rng, U);
printWeightsBiases(U, nn_init_params, print_values = true)

Layer 1 :  Dense(1 => 3, rbf) : 
        weights (shape: (3, 1)):
                Float32[0.70705664; -1.2754807; 1.0824884;;]
        bias (shape: (3,)):
                Float32[0.18551421, 0.4326675, 0.024800062]
Layer 2 :  Dense(3 => 3, rbf) : 
        weights (shape: (3, 3)):
                Float32[-0.7039337 0.65329766 0.73901486; -0.8899543 0.31681418 -0.7408178; -0.4000666 0.48276234 -0.5379133]
        bias (shape: (3,)):
                Float32[0.42945153, 0.3800699, -0.18156144]
Layer 3 :  Dense(3 => 3, rbf) : 
        weights (shape: (3, 3)):
                Float32[-0.022448301 -0.13235784 -0.32542706; 0.59363365 -0.1478889 0.9222369; -0.6551368 0.78240037 -0.21426916]
        bias (shape: (3,)):
                Float32[-0.18754774, 0.27221495, 0.07845076]
Layer 4 :  Dense(3 => 1) : 
        weights (shape: (1, 3)):
                Float32[-0.8237293 -0.6526296 0.12768888]
        bias (shape: (1,)):
                Float32[0.1146311]

Tested and Supported:

• Conv
• BatchNorm
• MaxPool
• Dropout
• FlattenLayer
• Dense
• LSTMCell
• RNNCell
• GRUCell
• AdaptiveMaxPool
• AdaptiveMeanPool
• AlphaDropout
• Bilinear
• BranchLayer
• Chain
• ConvTranspose
• CrossCor
• Embedding
• GlobalMaxPool
• GlobalMeanPool
• GroupNorm
• InstanceNorm
• LayerNorm
• Maxout
• MeanPool
• NoOpLayer
• PairwiseFusion
• Parallel
• Recurrence
• ReshapeLayer
• Scale
• SelectDim
• SkipConnection
• StatefulRecurrentCell
• Upsample
• VariationalHiddenDropout
• WeightNorm
• WrappedFunction