regioncam 0.5.2

Visualize linear regions in neural networks

Python interface for Regioncam

Regioncam is a rust library and python package for visualizing linear regions in a neural network. Regioncam works by tracking the output of all neural network layers in the regions where these outputs are linear. The inputs are in a 1 or 2 dimensional space.

Usage

import regioncam
import numpy as np

# Create a regioncam object, with the region [-1..1]^2
rc = regioncam.Regioncam(size=1)
# Apply a linear layer
rng = np.random.default_rng()
weight = rng.standard_normal((2, 30), dtype=np.float32)
bias = rng.standard_normal((30,), dtype=np.float32)
rc.linear(weight, bias)
# Apply a relu activation function
rc.relu()
# Write to svg
rc.write_svg("example.svg")
# Inspect regions
print(f"Created {rc.num_faces} regions")
for face in rc.faces:
    print(face.vertex_ids)

Produces the following svg file:

Arrays passed to regioncam, such as weights and biases must be numpy arrays or torch tensors with dtype float32.

Regioncam also has limited support for torch nn layers:

net = torch.nn.Sequential(
    torch.nn.Linear(2,30),
    torch.nn.ReLU(),
    torch.nn.Linear(30,30),
    torch.nn.ReLU(),
)
rc.add(net)

See examples/ for an example of visualizing a trained torch network.

The following layer types are supported:

• ReLU
• LeakyReLU
• Linear
• Sequential
• Identity
• Dropout (treated as Identity)
• residual

Installation

The regioncam python package can be installed with pip,

pip install regioncam

To compile locally, use

pip install ./regioncam-python

or

cd regioncam-python
maturin build --release

More information

Regioncam is similar to Splinecam, but the algorithm is different. See the github repository for the details.