Test porta logica XOR con network di neuroni

src/main.zig

@@ -1,69 +1,39 @@
 const std = @import("std");
-const Neuron = @import("neuron.zig").Neuron;
-const Tensor = @import("tensor.zig").Tensor;
+const SimpleNetwork = @import("network.zig").SimpleNetwork;
 
 pub fn main() !void {
-    var gpa = std.heap.GeneralPurposeAllocator(.{}){};
-    const allocator = gpa.allocator();
-    defer _ = gpa.deinit();
+    var net = SimpleNetwork.init(1234);
 
-    // 1. Inizializziamo il neurone (2 input perché la porta AND ha 2 ingressi)
-    var my_neuron = try Neuron.init(allocator, 2);
-    defer my_neuron.deinit();
-
-    // 2. Prepariamo il Dataset (AND Gate)
-    // Creiamo un tensore riutilizzabile per gli input
-    var input_tensor = try Tensor.init(allocator, &[_]usize{2});
-    defer input_tensor.deinit();
-
-    // I 4 casi possibili (Training Data)
-    const training_data = [_][2]f32{
+    const inputs = [_][2]f32{
         .{ 0.0, 0.0 },
         .{ 0.0, 1.0 },
         .{ 1.0, 0.0 },
         .{ 1.0, 1.0 },
     };
 
-    // Le 4 risposte corrette (Labels)
-    const targets = [_]f32{ 0.0, 0.0, 0.0, 1.0 };
+    const targets = [_]f32{ 0.0, 1.0, 1.0, 0.0 };
 
-    const lr: f32 = 0.1; // Learning rate un po' più aggressivo
+    const lr: f32 = 0.5;
 
-    std.debug.print("--- INIZIO TRAINING (AND GATE) ---\n", .{});
+    std.debug.print("--- TRAINING XOR (2 LAYERS) ---\n", .{});
 
-    // 3. Ciclo di Training
     var epoch: usize = 0;
-    while (epoch < 2000) : (epoch += 1) { // 2000 Epoche
-        var total_error: f32 = 0.0;
+    while (epoch < 10000) : (epoch += 1) {
+        var total_loss: f32 = 0.0;
 
-        // Per ogni epoca, passiamo attraverso TUTTI gli esempi
-        for (training_data, 0..) |data, index| {
-            // Carichiamo i dati nel tensore
-            input_tensor.data[0] = data[0];
-            input_tensor.data[1] = data[1];
-
-            // Train su questo specifico esempio
-            const loss = my_neuron.train(input_tensor, targets[index], lr);
-            total_error += loss;
+        for (inputs, 0..) |inp, i| {
+            total_loss += net.train(inp, targets[i], lr);
         }
 
-        // Stampiamo ogni 200 epoche
-        if (epoch % 200 == 0) {
-            std.debug.print("Epoca {d}: Errore Medio = {d:.6}\n", .{ epoch, total_error / 4.0 });
+        if (epoch % 1000 == 0) {
+            std.debug.print("Epoca {d}: Loss = {d:.6}\n", .{ epoch, total_loss / 4.0 });
         }
     }
 
-    std.debug.print("\n--- TEST FINALE ---\n", .{});
-
-    // Verifichiamo cosa ha imparato
-    for (training_data) |data| {
-        input_tensor.data[0] = data[0];
-        input_tensor.data[1] = data[1];
-        const prediction = my_neuron.forward(input_tensor);
-
-        // Arrotondiamo visivamente per capire se è 0 o 1
-        const result_bool: u8 = if (prediction > 0.5) 1 else 0;
-
-        std.debug.print("Input: {d:.1}, {d:.1} -> Predizione: {d:.4} (Interpretato: {d})\n", .{ data[0], data[1], prediction, result_bool });
+    std.debug.print("\n--- TEST XOR ---\n", .{});
+    for (inputs) |inp| {
+        const out = net.forward(inp);
+        const bit: u8 = if (out > 0.5) 1 else 0;
+        std.debug.print("In: {d:.0},{d:.0} -> Out: {d:.4} -> {d}\n", .{ inp[0], inp[1], out, bit });
     }
 }

src/network.zig

@@ -0,0 +1,83 @@
+const std = @import("std");
+
+pub const SimpleNetwork = struct {
+    w_hidden: [2][2]f32,
+    b_hidden: [2]f32,
+
+    w_output: [2]f32,
+    b_output: f32,
+
+    hidden_outputs: [2]f32,
+
+    pub fn init(seed: u64) SimpleNetwork {
+        var prng = std.Random.DefaultPrng.init(seed);
+        const rand = prng.random();
+
+        var net = SimpleNetwork{
+            .w_hidden = undefined,
+            .b_hidden = undefined,
+            .w_output = undefined,
+            .b_output = 0.0,
+            .hidden_outputs = undefined,
+        };
+
+        for (&net.w_hidden) |*row| {
+            row[0] = rand.float(f32) * 2.0 - 1.0;
+            row[1] = rand.float(f32) * 2.0 - 1.0;
+        }
+        for (&net.b_hidden) |*b| b.* = 0.0;
+
+        for (&net.w_output) |*w| w.* = rand.float(f32) * 2.0 - 1.0;
+
+        return net;
+    }
+
+    fn sigmoid(x: f32) f32 {
+        return 1.0 / (1.0 + std.math.exp(-x));
+    }
+
+    fn sigmoid_derivative(x: f32) f32 {
+        return x * (1.0 - x);
+    }
+
+    pub fn forward(self: *SimpleNetwork, input: [2]f32) f32 {
+        for (0..2) |i| {
+            const sum = (input[0] * self.w_hidden[i][0]) +
+                (input[1] * self.w_hidden[i][1]) +
+                self.b_hidden[i];
+            self.hidden_outputs[i] = sigmoid(sum);
+        }
+
+        const sum_out = (self.hidden_outputs[0] * self.w_output[0]) +
+            (self.hidden_outputs[1] * self.w_output[1]) +
+            self.b_output;
+
+        return sigmoid(sum_out);
+    }
+
+    pub fn train(self: *SimpleNetwork, input: [2]f32, target: f32, lr: f32) f32 {
+        const prediction = self.forward(input);
+
+        const output_error = target - prediction;
+        const output_delta = output_error * sigmoid_derivative(prediction);
+
+        var hidden_deltas: [2]f32 = undefined;
+        for (0..2) |i| {
+            const error_contrib = output_delta * self.w_output[i];
+            hidden_deltas[i] = error_contrib * sigmoid_derivative(self.hidden_outputs[i]);
+        }
+
+        for (0..2) |i| {
+            self.w_output[i] += lr * output_delta * self.hidden_outputs[i];
+        }
+        self.b_output += lr * output_delta;
+
+        for (0..2) |i| {
+            self.w_hidden[i][0] += lr * hidden_deltas[i] * input[0];
+            self.w_hidden[i][1] += lr * hidden_deltas[i] * input[1];
+            self.b_hidden[i] += lr * hidden_deltas[i];
+        }
+
+        return output_error * output_error;
+    }
+};