gavindiaz/TR_Agent
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix compile errors: NeuralNet input param rename, Orchestrator refs to pointers, EVT_MaxAbsZ call, SYMBOL_MARGIN_INITIAL, remove unused var

Browse Source
This commit is contained in:
pietro_giacobazzi
2026-06-13 15:07:05 +02:00
parent 67c8485373
commit c70c2f28ce
3 changed files with 80 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files
MQL5/Experts/MultiAgentTest/Core/NeuralNet.mqh
+37 -37
View File
@@ -255,50 +255,50 @@ public:
m_lossCsvFn = "";
}
void ForwardPass(vector &input, vector &output, vector &h1Cache) {
h1Cache.Resize(m_hidden);
for(int i = 0; i < m_hidden; i++) {
double sum = m_b1[i];
for(int j = 0; j < m_inputs; j++)
sum += input[j] * m_W1[j][i];
h1Cache[i] = ReLU(sum);
}
// BatchNorm inferenza: normalizza con running stats (NeuroBook §6.3)
ApplyBN(h1Cache);
// Dropout scaling in inferenza: scale = 1 - dropoutRate (NeuroBook §6.2)
if(m_dropoutRate > 0) {
for(int i = 0; i < h1Cache.Size(); i++)
h1Cache[i] *= (1.0 - m_dropoutRate);
}
void ForwardPass(vector &inp, vector &output, vector &h1Cache) {
h1Cache.Resize(m_hidden);
for(int i = 0; i < m_hidden; i++) {
double sum = m_b1[i];
for(int j = 0; j < m_inputs; j++)
sum += inp[j] * m_W1[j][i];
h1Cache[i] = ReLU(sum);
}
// BatchNorm inferenza: normalizza con running stats (NeuroBook §6.3)
ApplyBN(h1Cache);
// Dropout scaling in inferenza: scale = 1 - dropoutRate (NeuroBook §6.2)
if(m_dropoutRate > 0) {
for(int i = 0; i < h1Cache.Size(); i++)
h1Cache[i] *= (1.0 - m_dropoutRate);
}
output.Resize(m_outputs);
for(int i = 0; i < m_outputs; i++) {
double sum = m_b2[i];
for(int j = 0; j < m_hidden; j++)
sum += h1Cache[j] * m_W2[j][i];
output[i] = sum;
}
Softmax(output);
}
output.Resize(m_outputs);
for(int i = 0; i < m_outputs; i++) {
double sum = m_b2[i];
for(int j = 0; j < m_hidden; j++)
sum += h1Cache[j] * m_W2[j][i];
output[i] = sum;
}
Softmax(output);
}
void Forward(vector &input, vector &output) {
vector h1Cache;
ForwardPass(input, output, h1Cache);
}
void Forward(vector &inp, vector &output) {
vector h1Cache;
ForwardPass(inp, output, h1Cache);
}
double TrainSample(vector &input, vector &target, double lr, double weight = 1.0) {
double TrainSample(vector &inp, vector &target, double lr, double weight = 1.0) {
if(!m_initialized) return -1.0;
// ── Forward ──
// Layer 1: W1*x + b1 → z1 → ReLU → h1_raw → Dropout → h1_drop → BN → h1_norm
vector z1(m_hidden);
vector h1_raw(m_hidden);
for(int i = 0; i < m_hidden; i++) {
z1[i] = m_b1[i];
for(int j = 0; j < m_inputs; j++)
z1[i] += input[j] * m_W1[j][i];
h1_raw[i] = ReLU(z1[i]);
}
for(int i = 0; i < m_hidden; i++) {
z1[i] = m_b1[i];
for(int j = 0; j < m_inputs; j++)
z1[i] += inp[j] * m_W1[j][i];
h1_raw[i] = ReLU(z1[i]);
}
// Dropout (NeuroBook §6.2): salva in m_dropoutMask, applica a h1_drop
vector h1_drop(m_hidden);
@@ -390,11 +390,11 @@ public:
for(int i = 0; i < m_hidden; i++)
dL_dz1[i] = dL_dh1_drop[i] * ReLUDeriv(z1[i]);
// dL/dW1 = input ⊗ dL/dz1
// dL/dW1 = inp ⊗ dL/dz1
matrix dL_dW1(m_inputs, m_hidden);
for(int i = 0; i < m_inputs; i++)
for(int j = 0; j < m_hidden; j++)
dL_dW1[i][j] = input[i] * dL_dz1[j];
dL_dW1[i][j] = inp[i] * dL_dz1[j];
// dL/db1 = dL/dz1
vector dL_db1(m_hidden);