remove useless sigma initialisation in fc_nn + add random features to...

remove useless sigma initialisation in fc_nn + add random features to neural_networks (to be splitted and renamed as tensorflow utils) + add comment to the time_fct

skluc/examples/fc_nn.py

@@ -44,9 +44,6 @@ Y_test = Y_test.astype(np.float32)
 
 
 def main():
-    SIGMA = 5.0
-    print("Sigma = {}".format(SIGMA))
-
     with tf.Graph().as_default():
         input_dim, output_dim = X_train.shape[1], Y_train.shape[1]
 

skluc/neural_networks.py

@@ -5,6 +5,7 @@ import numpy as np
 
 # todo peutetre renommer ce module en "tensorflow utils"
 
+
 def weight_variable(shape, trainable=True):
     # initial = tf.truncated_normal(shape, stddev=0.1)
     # return tf.Variable(initial, name="weights", trainable=trainable)
@@ -160,3 +161,31 @@ def inference_mnist(x_image, output_dim):
     out_fc = fc_mnist(h_conv)
     y_out, keep_prob = classification_mnist(out_fc, output_dim)
     return y_out, keep_prob
+
+
+# todo verifier que ces random_variable/random_biases sont utilisables
+# --- Random Fourier Features --- #
+
+def random_variable(shape, sigma):
+    W = np.random.normal(size=shape, scale=sigma).astype(np.float32)
+    return tf.Variable(W, name="random_Weights", trainable=False)
+
+
+def random_biases(shape):
+    b = np.random.uniform(0, 2 * np.pi, size=shape).astype(np.float32)
+    return tf.Variable(b, name="random_biase", trainable=False)
+
+
+# --- Representation Layer --- #
+
+def random_features(conv_out, sigma):
+    with tf.name_scope("random_features"):
+        init_dim = np.prod([s.value for s in conv_out.shape if s.value is not None])
+        conv_out2 = tf.reshape(conv_out, [-1, init_dim])
+
+        W = random_variable((init_dim, init_dim), sigma)
+        b = random_biases(init_dim)
+        h1 = tf.matmul(conv_out2, W, name="Wx") + b
+        h1_cos = tf.cos(h1)
+        h1_final = tf.scalar_mul(np.sqrt(2.0 / init_dim).astype(np.float32), h1_cos)
+        return h1_final
\ No newline at end of file

skluc/utils.py

@@ -2,6 +2,15 @@ import time as t
 
 
 def time_fct(fct, *args, n_iter=100, **kwargs):
+    """
+    Return the average time spent by the function.
+
+    :param fct: the actual function to time
+    :param args: the positional arguments of the function
+    :param n_iter: number of runs of the function
+    :param kwargs: the keyword arguments of the function
+    :return: the average time of execution of the function
+    """
     time_sum = 0
     for _ in range(n_iter):
         start = t.time()