diff --git a/model.ml b/model.ml
index e13dbcc0257adafeedac656e599fffb068209f78..8fc7eb5e6e5fa2e3e858c2a4198d5974f56a2920 100644
--- a/model.ml
+++ b/model.ml
@@ -4,21 +4,191 @@
(* *)
(**************************************************************************)
-type format = Onnx | Nnet [@@deriving show { with_path = false}]
+open Base
+module Format = Caml.Format
+module Sys = Caml.Sys
+module Filename = Caml.Filename
+
+type nnet = {
+ n_layers : int;
+ n_inputs: int;
+ n_outputs: int;
+ max_layer_size: int;
+ layer_sizes: int list;
+ min_input_values: float list;
+ max_input_values: float list;
+ mean_values: float list * float;
+ range_values: float list * float;
+} [@@deriving show { with_path = false }]
+
+type format = Onnx | Nnet of nnet [@@deriving show { with_path = false}]
type t = {
format: format;
filename: string;
}
+
+(* Nnet format handling. *)
+
+let nnet_format_error s =
+ Error (Format.sprintf "Nnet format error: %s condition not satisfied." s)
+
+let nnet_delimiter = Str.regexp ","
+
+(* Parse a single Nnet format line: split line using [nnet_delimiter] as
+ delimiter, and convert each string into a number by means of converter [f]. *)
+let handle_nnet_line ~f line =
+ List.map
+ ~f:(fun s -> f (String.strip s))
+ (Str.split nnet_delimiter line)
+
+(* Skip the header part, ie comments, of the Nnet format. *)
+let handle_nnet_header filename in_channel =
+ let exception End_of_header in
+ let pos_in = ref (Stdlib.pos_in in_channel) in
+ try
+ while true do
+ let line = Stdlib.input_line in_channel in
+ if not (Str.string_match (Str.regexp "//") line 0)
+ then raise End_of_header
+ else pos_in := Stdlib.pos_in in_channel
+ done;
+ assert false
+ with
+ | End_of_header ->
+ (* At this point we have read one line past the header part: seek back. *)
+ Stdlib.seek_in in_channel !pos_in;
+ Ok ()
+ | End_of_file ->
+ Error (Format.sprintf "Nnet model not found in file `%s'." filename)
+
+(* Retrieve number of layers, inputs, outputs and maximum layer size. *)
+let handle_nnet_basic_info in_channel =
+ try
+ let line = Stdlib.input_line in_channel in
+ match handle_nnet_line ~f:Stdlib.int_of_string line with
+ | [ n_layers; n_inputs; n_outputs; max_layer_size ] ->
+ Ok (n_layers, n_inputs, n_outputs, max_layer_size)
+ | _ ->
+ nnet_format_error "second"
+ with End_of_file ->
+ nnet_format_error "second"
+
+(* Retrieve size of each layer, including inputs and outputs. *)
+let handle_nnet_layer_sizes n_layers in_channel =
+ try
+ let line = Stdlib.input_line in_channel in
+ let layer_sizes = handle_nnet_line ~f:Stdlib.int_of_string line in
+ if List.length layer_sizes = (n_layers + 1)
+ then Ok layer_sizes
+ else nnet_format_error "third"
+ with End_of_file ->
+ nnet_format_error "third"
+
+(* Skip unused flag. *)
+let handle_nnet_unused_flag in_channel =
+ try
+ let _ = Stdlib.input_line in_channel in
+ Ok ()
+ with End_of_file ->
+ nnet_format_error "forth"
+
+(* Retrive minimum values of inputs. *)
+let handle_nnet_min_input_values n_inputs in_channel =
+ try
+ let line = Stdlib.input_line in_channel in
+ let min_input_values = handle_nnet_line ~f:Stdlib.float_of_string line in
+ if List.length min_input_values = n_inputs
+ then Ok min_input_values
+ else nnet_format_error "fifth"
+ with End_of_file ->
+ nnet_format_error "fifth"
+
+(* Retrive maximum values of inputs. *)
+let handle_nnet_max_input_values n_inputs in_channel =
+ try
+ let line = Stdlib.input_line in_channel in
+ let max_input_values = handle_nnet_line ~f:Stdlib.float_of_string line in
+ if List.length max_input_values = n_inputs
+ then Ok max_input_values
+ else nnet_format_error "sixth"
+ with End_of_file ->
+ nnet_format_error "sixth"
+
+(* Retrieve mean values of inputs and one value for all outputs. *)
+let handle_nnet_mean_values n_inputs in_channel =
+ try
+ let line = Stdlib.input_line in_channel in
+ let mean_values = handle_nnet_line ~f:Stdlib.float_of_string line in
+ if List.length mean_values = (n_inputs + 1)
+ then
+ let mean_input_values, mean_output_value =
+ List.split_n mean_values n_inputs
+ in
+ Ok (mean_input_values, List.hd_exn mean_output_value)
+ else
+ nnet_format_error "seventh"
+ with End_of_file ->
+ nnet_format_error "seventh"
+
+(* Retrieve range values of inputs and one value for all outputs. *)
+let handle_nnet_range_values n_inputs in_channel =
+ try
+ let line = Stdlib.input_line in_channel in
+ let range_values = handle_nnet_line ~f:Stdlib.float_of_string line in
+ if List.length range_values = (n_inputs + 1)
+ then
+ let range_input_values, range_output_value =
+ List.split_n range_values n_inputs
+ in
+ Ok (range_input_values, List.hd_exn range_output_value)
+ else
+ nnet_format_error "eighth"
+ with End_of_file ->
+ nnet_format_error "eighth"
+
+(* Retrieves [filename] Nnet model metadata wrt Nnet format specification (see
+ https://github.com/sisl/NNet for details.) *)
+let retrieve_nnet_metadata filename =
+ let open Result in
+ let in_channel = Stdlib.open_in filename in
+ try
+ handle_nnet_header filename in_channel >>= fun () ->
+ handle_nnet_basic_info in_channel >>= fun (n_ls, n_ins, n_outs, max_l_size) ->
+ handle_nnet_layer_sizes n_ls in_channel >>= fun layer_sizes ->
+ handle_nnet_unused_flag in_channel >>= fun () ->
+ handle_nnet_min_input_values n_ins in_channel >>= fun min_input_values ->
+ handle_nnet_max_input_values n_ins in_channel >>= fun max_input_values ->
+ handle_nnet_mean_values n_ins in_channel >>= fun mean_values ->
+ handle_nnet_range_values n_ins in_channel >>= fun range_values ->
+ Stdlib.close_in in_channel;
+ Ok
+ { n_layers = n_ls;
+ n_inputs = n_ins;
+ n_outputs = n_outs;
+ max_layer_size = max_l_size;
+ layer_sizes;
+ min_input_values;
+ max_input_values;
+ mean_values;
+ range_values; }
+ with Failure msg ->
+ Error (Format.sprintf "Unexpected error: %s." msg)
+
+(* Generic model. *)
+
let build ~filename =
+ let open Result in
if Sys.file_exists filename
then
- let format =
- if Filename.check_suffix filename "nnet"
- then Nnet
- else Onnx
- in
+ begin
+ if Filename.check_suffix filename "onnx"
+ then Ok Onnx
+ else
+ retrieve_nnet_metadata filename >>= fun nnet ->
+ Ok (Nnet nnet)
+ end >>= fun format ->
Ok { format; filename }
else
Error (Format.sprintf "No such file `%s'." filename)
diff --git a/model.mli b/model.mli
index 45687d863e14af274565c307a74ac073145de2b1..fdc61c7d7a6c0c7ea8194760e1aad4ea5987f16a 100644
--- a/model.mli
+++ b/model.mli
@@ -4,15 +4,36 @@
(* *)
(**************************************************************************)
-type format = Onnx | Nnet [@@deriving show { with_path = false }]
+(** Nnet model metadata. *)
+type nnet = private {
+ (** Number of layers. *)
+ n_layers : int;
+ (** Number of inputs. *)
+ n_inputs: int;
+ (** Number of outputs. *)
+ n_outputs: int;
+ (** Maximum layer size. *)
+ max_layer_size: int;
+ (** Size of each layer. *)
+ layer_sizes: int list;
+ (** Minimum values of inputs. *)
+ min_input_values: float list;
+ (** Maximum values of inputs. *)
+ max_input_values: float list;
+ (** Mean values of inputs and one value for all outputs. *)
+ mean_values: float list * float;
+ (** Range values of inputs and one value for all outputs. *)
+ range_values: float list * float;
+} [@@deriving show { with_path = false }]
+
+type format = Onnx | Nnet of nnet [@@deriving show { with_path = false }]
type t = private {
format: format;
filename: string;
}
-(** Builds a model out of the given [filename], if possible.
-
- The model is inferred from the [filename] extension.
-*)
+(** Builds a model out of the given [filename], if possible. The model is
+ inferred from the [filename] extension for the Onnx case, while Nnet models
+ are parsed for metadata retrieval and conformity checks. *)
val build: filename:string -> (t, string) Result.t