added results

amp_est.c → amp_est_bi.c

@@ -21,16 +21,16 @@ int main(void) {
     kad_node_t* t;
     kann_t* ann;
     size_t inputs = 1024;
-    size_t outputs = 1;
-    size_t num_layers = 4;
-    size_t neurons = 64;
+    size_t outputs = 4;
+    size_t num_layers = 1;
+    size_t neurons = 8;
     size_t SR = 192000;
 
     /* Create the neural network */
     t = kann_layer_input(inputs);
     for (size_t i = 0; i < num_layers; i++) {
         t = kann_layer_dense(t, neurons);
-        t = kad_relu(t);
+        t = kad_softmax(t);
     }
     t = kann_layer_cost(t, outputs, KANN_C_MSE);
     ann = kann_new(t, 0);
@@ -71,9 +71,10 @@ int main(void) {
     for (size_t j = 0; j < testsize; j++) {
         output = kann_apply1(ann, x->vec_space[j]);
         printf("Target: %.10f; prediction: %.10f; error factor: %.10f\n", 
-            y->vec_space[j][0], *output, y->vec_space[j][0] / *output);
-        fprintf(csv, "%f, %f\n", *output, y->vec_space[j][0] / *output);
+            y->vec_space[j][0], *output, y->vec_space[j][0] - *output);
+        fprintf(csv, "%f, %f\n", y->vec_space[j][0], y->vec_space[j][0] - *output);
     }
+
     sig_free(x);
     sig_free(y);
     fclose(csv);

est_plot.py → amp_est_plot.py

@@ -15,7 +15,7 @@
 
 for filename in glob.glob('*.csv'):
 
-    freq = []
+    amp = []
     error = []
     tmp = []
 
@@ -29,16 +29,15 @@
     l = len(data)
 
     for i in range(l):
-        freq = np.append(freq, tmp[i][0])
+        amp = np.append(amp, tmp[i][0])
         error = np.append(error, tmp[i][1])
 
-    plt.xlabel('Frequency (Hz)')
-    plt.ylabel('Error (target-ANN ratio)')
-
-    plt.axhline(y = 1,linewidth = 1, color = 'r', label = "Best fit")
-    plt.scatter(freq, error, linewidth = .5, color = 'black', label = "ANN output")
-
-    plt.xscale('log')
+    plt.xlabel('Amplitude target')
+    plt.ylabel('Prediction error (target-ANN difference)')
+
+    plt.ylim(-1, 1)
+    plt.axhline(y = 0, linewidth = .25, color = 'r', label = "Best fit")
+    plt.scatter(amp, error, marker = "o", s = .1, linewidth = 1, color = 'black', label = "Prediction error")
 
     plt.title("Amplitude estimation")
 
@@ -49,3 +48,5 @@
     plt.grid(True)
 
     plt.savefig(name)
+
+    plt.clf()