Skip to content

Usage.md

Latest commit

 

History

History
59 lines (45 loc) · 2.12 KB

Usage.md

File metadata and controls

59 lines (45 loc) · 2.12 KB
Raw

Usage

After compilation, the dlinear binary will be located in the bazel-bin/dlinear directory. There are several options to run it:

  • run ./bazel-bin/dlinear/dlinear in the terminal
  • add the absolute path of the directory bazel-bin/dlinear to the PATH environment variable
  • move the binary from ./bazel-bin/dlinear/dlinear to a directory already in the PATH, like /usr/local/bin
  • create a symbolic link to the executable
  • create an alias for the executable

From now on, the snippets will assume that the dlinear binary has been added to the PATH.

# Invoke dlinear help
dlinear -h

File mode

By default, dlinear expects the path to the problem file as the only positional argument. The file can be in the SMT2 or MPS format. If left unspecified, the program will look at the file extension to determine the format.

# Invoke dlinear with a problem in SMT2 format
dlinear path/to/problem.smt2
# Invoke dlinear with a problem in MPS format
dlinear path/to/problem.mps
# Invoke dlinear with a problem assuming it is in SMT2 format
dlinear path/to/problem --format smt2

Stdin mode

dlinear can be used in stdin mode, where the user can input is received from the standard input. Note that in this case the format must be specified explicitly.

# Invoke dlinear in stdin mode (SMT2 format)
dlinear --in --format smt2
# Invoke dlinear in stdin mode (MPS format)
dlinear --in --format mps

When the stdin mode is used, the problem must be typed directly in the terminal. To signal dlinear the end of the input, press Ctrl+D two times. This will start the solver.

Neural network verification

dlinear can be used to verify the robustness of a neural network. The network must be in the ONNX format, while the property must be in the VNN-LIB format. To run the verification, use the following command:

# Invoke dlinear with a neural network and a property
dlinear --onnx path/to/network.onnx path/to/property.vnnlib