package caisar

You can search for identifiers within the package.

in-package search v0.2.0

A platform for characterizing the safety and robustness of artificial intelligence based software

Install

sha256=1b25c8668d428bcfc83c95147b6e45ff0a3bfa05ecd11369d12e963e29819e2e

sha512=edc7d7c0e96802811de3cb1caa3d14cc3d867ee7310748e8188eca9246a362549545c7816c8037511931dc4b7770b5ccc11b0d03abe8843b7c4db7880bf8e1fd

Module to parse neural networks written in the NNet format https://github.com/sisl/NNet

Sourcetype t = private {

n_layers : int;
(*
Number of layers.
*)
n_inputs : int;
(*
Number of inputs.
*)
n_outputs : int;
(*
Number of outputs.
*)
max_layer_size : int;
(*
Maximum layer size.
*)
layer_sizes : int list;
(*
Size of each layer.
*)
min_input_values : float list option;
(*
Minimum values of inputs.
*)
max_input_values : float list option;
(*
Maximum values of inputs.
*)
mean_values : (float list * float) option;
(*
Mean values of inputs and one value for all outputs.
*)
range_values : (float list * float) option;
(*
Range values of inputs and one value for all outputs.
*)
weights_biases : float list list;
(*
All weights and biases of NNet model.
*)
nir : Nir.Ngraph.t;

}

NNet model metadata.

Sourceval parse : ?permissive:bool -> string -> (t, string) Result.t

Parse an NNet file.

parameter permissive
false by default. When set, parsing does not fail on non available information, which are set to None instead.

Sourceval to_nir : t -> Nir.Ngraph.t

Convert an well-formed NNet into a Nir.