Article-24337-Working-With-ONNX-Models-In-MQL5-Part-3/make_model.py

# Copyright 2026, Allan Munene Mutiiria.
# https://t.me/Forex_Algo_Trader

import os
import numpy as np
import onnx
from onnx import helper, TensorProto, numpy_helper

# Numbers going in
INPUTS  = 16
# Width of each branch before they rejoin
BRANCH  = 8
# Width after the two branches are concatenated
MERGED  = 10
# Down, flat, up
OUTPUTS = 3

rng = np.random.default_rng(7)

def xavier(rows, cols):
    # Keeps the signal from dying or exploding
    bound = np.sqrt(6.0 / (rows + cols))
    return rng.uniform(-bound, bound, (rows, cols)).astype(np.float32)

# Stored through float_data rather than raw_data, so both paths are covered
Wa = xavier(INPUTS, BRANCH)
initializers = [
    helper.make_tensor("Wa", TensorProto.FLOAT, [INPUTS, BRANCH],
                       Wa.flatten().tolist(), raw=False),
    numpy_helper.from_array(np.zeros(BRANCH, np.float32), "ba"),
    numpy_helper.from_array(xavier(INPUTS, BRANCH), "Wb"),
    numpy_helper.from_array(np.zeros(BRANCH, np.float32), "bb"),
    numpy_helper.from_array(xavier(BRANCH * 2, MERGED), "Wm"),
    numpy_helper.from_array(np.zeros(MERGED, np.float32), "bm"),
    numpy_helper.from_array(xavier(MERGED, OUTPUTS), "Wo"),
    numpy_helper.from_array(np.zeros(OUTPUTS, np.float32), "bo"),
]

# Stored as double so the parser meets a type other than float
initializers.append(numpy_helper.from_array(np.full(MERGED, 0.85, np.float64), "scale"))

# Stored through int64_data, which is a third way a tensor can carry values
initializers.append(helper.make_tensor("reshape_to", TensorProto.INT64, [2],
                                       [1, BRANCH * 2], raw=False))

nodes = [
    # The input feeds two branches at once, so the graph is not a chain
    helper.make_node("Gemm", ["input", "Wa", "ba"], ["gemm_a"], name="branch_a"),
    helper.make_node("Tanh", ["gemm_a"], ["act_a"], name="tanh_a"),
    helper.make_node("Gemm", ["input", "Wb", "bb"], ["gemm_b"], name="branch_b"),
    helper.make_node("Relu", ["gemm_b"], ["act_b"], name="relu_b"),
    # Two edges arrive here, so a layout cannot assume one parent
    helper.make_node("Concat", ["act_a", "act_b"], ["merged"], axis=1, name="join"),
    # Carries a list attribute, which is a form a scalar never exercises
    helper.make_node("Transpose", ["merged"], ["flipped"], perm=[1, 0], name="flip"),
    # Takes its shape from a tensor rather than an attribute
    helper.make_node("Reshape", ["flipped", "reshape_to"], ["restored"], name="restore"),
    helper.make_node("Gemm", ["restored", "Wm", "bm"], ["gemm_m"], name="mixer"),
    helper.make_node("Sigmoid", ["gemm_m"], ["act_m"], name="sigmoid_m"),
    # A node whose second input is an initializer rather than another node
    helper.make_node("Mul", ["act_m", "scale_f"], ["scaled"], name="rescale"),
    helper.make_node("Gemm", ["scaled", "Wo", "bo"], ["gemm_o"], name="output_layer"),
    helper.make_node("Softmax", ["gemm_o"], ["output"], axis=1, name="softmax",
                     doc_string="Turns the three scores into probabilities"),
]

# The double tensor has to be cast before Mul can use it
nodes.insert(9, helper.make_node("Cast", ["scale"], ["scale_f"],
                                 to=TensorProto.FLOAT, name="cast_scale"))

# Holds text rather than numbers, which the reader stores a different way
initializers.append(helper.make_tensor("labels", TensorProto.STRING, [OUTPUTS],
                                       [b"down", b"flat", b"up"], raw=False))

# Named only where a value passes between layers, so every edge can be labelled
inner = [
    helper.make_tensor_value_info("gemm_a",   TensorProto.FLOAT, [1, BRANCH]),
    helper.make_tensor_value_info("act_a",    TensorProto.FLOAT, [1, BRANCH]),
    helper.make_tensor_value_info("gemm_b",   TensorProto.FLOAT, [1, BRANCH]),
    helper.make_tensor_value_info("act_b",    TensorProto.FLOAT, [1, BRANCH]),
    helper.make_tensor_value_info("merged",   TensorProto.FLOAT, [1, BRANCH * 2]),
    helper.make_tensor_value_info("flipped",  TensorProto.FLOAT, [BRANCH * 2, 1]),
    helper.make_tensor_value_info("restored", TensorProto.FLOAT, [1, BRANCH * 2]),
    helper.make_tensor_value_info("gemm_m",   TensorProto.FLOAT, [1, MERGED]),
    helper.make_tensor_value_info("act_m",    TensorProto.FLOAT, [1, MERGED]),
    helper.make_tensor_value_info("scale_f",  TensorProto.FLOAT, [MERGED]),
    helper.make_tensor_value_info("scaled",   TensorProto.FLOAT, [1, MERGED]),
    helper.make_tensor_value_info("gemm_o",   TensorProto.FLOAT, [1, OUTPUTS]),
]

# Stored as an index and value pair rather than a dense run
sparse = helper.make_sparse_tensor(
    helper.make_tensor("sp_values", TensorProto.FLOAT, [3], [0.5, 1.5, 2.5], raw=False),
    helper.make_tensor("sp_index", TensorProto.INT64, [3], [0, 4, 9], raw=False),
    [MERGED])

graph = helper.make_graph(
    nodes, "viewer_sample",
    [helper.make_tensor_value_info("input",  TensorProto.FLOAT, [1, INPUTS])],
    [helper.make_tensor_value_info("output", TensorProto.FLOAT, [1, OUTPUTS])],
    initializer=initializers,
    value_info=inner,
    sparse_initializer=[sparse],
    doc_string="Sample graph carrying every structure the reader handles",
)

model = helper.make_model(graph, producer_name="onnx-model-viewer",
                          opset_imports=[helper.make_opsetid("", 13)])
model.ir_version = 8
model.producer_version = "1.0"
model.domain = "forex.algo.trader"
model.model_version = 1
model.doc_string = "A viewer sample built to exercise every field"

# Key and value pairs the reader reports under metadata
for key, value in (("author", "Allan Munene Mutiiria"),
                   ("purpose", "ONNX Model Viewer sample"),
                   ("built", "make_model.py")):
    entry = model.metadata_props.add()
    entry.key = key
    entry.value = value
onnx.checker.check_model(model)

# Climb to the terminal's MQL5 root from wherever this script was placed
folder = os.path.dirname(os.path.abspath(__file__))
while os.path.basename(folder) != "MQL5" and os.path.dirname(folder) != folder:
    folder = os.path.dirname(folder)

# MQL5 can only read from Files, so the model has to land there
target = os.path.join(folder, "Files", "ONNX Model Viewer")
os.makedirs(target, exist_ok=True)
path = os.path.join(target, "model.onnx")
onnx.save(model, path)

params = sum(int(np.prod(t.dims)) for t in initializers)
print(f"wrote {path}")
print(f"{len(nodes)} nodes, {params} parameters")