- biaslayer first test for old functionality ok

few_shot_tests.py

@@ -171,9 +171,11 @@ def generate_add_samples(self, phase = 'train'):
 
 class BiasLayer(Layer):
 
-    def __init__(self, proto_num, mult_bias = 1, **kwargs):
+    def __init__(self, proto_num, do_bias, bias_num, **kwargs):
         self.proto_num = proto_num
-        self.mult_bias = mult_bias
+        self.do_bias = do_bias
+        self.bias_num = bias_num
+        print('mult bias',do_bias, proto_num)
         super(BiasLayer, self).__init__(**kwargs)
 
     def build(self, input_shape):
@@ -186,13 +188,16 @@ def build(self, input_shape):
 
     def call(self, x):
         #return tf.expand_dims(self.bias, axis = 0)# let
-        return self.bias * self.mult_bias + (1-self.mult_bias) * x
+        if self.do_bias:
+            return self.bias        + x * 0
+        else:
+            return self.bias * 0    + x
 
     def compute_output_shape(self, input_shape):
         return input_shape
 
     def get_config(self):
-        return {'proto_num': self.proto_num, 'mult_bias' : self.mult_bias}
+        return {'proto_num': self.proto_num, 'do_bias' : self.do_bias,'bias_num' : self.bias_num}
 
 inputs = Input(shape=(None,84,84,3))
 print('the shape', inputs.shape)
@@ -222,11 +227,11 @@ def get_config(self):
 input1 = Input(shape=(None,84,84,3))
 input2 = Input(shape=(None,84,84,3)) #, tensor = K.variable(episode_train_img[0:0]))
 
-input2b = BiasLayer(shots * cathegories, mult_bias = float(args.biaslayer1))(input2)
+input2b = BiasLayer(shots * cathegories, args.biaslayer1, 1)(input2)
 encoded_l = model_img(input1)
 encoded_r = model_img(input2b)
 
-encoded_rb = BiasLayer(shots * cathegories, mult_bias = float(args.biaslayer1))(encoded_r)
+encoded_rb = BiasLayer(shots * cathegories, args.biaslayer2, 2)(encoded_r)
 # Add a customized layer to compute the absolute difference between the encodings
 L1_layer = Lambda(lambda tensors:K.abs(tensors[0] - tensors[1]))
 L1_distance = L1_layer([encoded_l, encoded_rb])
@@ -291,6 +296,13 @@ def all_layers(model):
         p='timedi'
     if args.enable_only_layers_of_list is not None:
         l2.trainable = False
+    if isinstance(l2,BiasLayer):
+        print(l2.bias_num, l2.do_bias,args.biaslayer1,args.biaslayer2)
+        if (l2.bias_num == 1):
+            l2.do_bias = l2.trainable = args.biaslayer1
+        if (l2.bias_num == 2):
+            l2.do_bias = l2.trainable = args.biaslayer2
+
     print('{:10} {:10} {:20} {:10}  {:10}'.format(l, p,l2.name, ("fixed", "trainable")[l2.trainable], l2.count_params()))
 
 if args.enable_only_layers_of_list is not None:
@@ -325,6 +337,8 @@ def all_layers(model):
 lambda_model.fit_generator(keras_gen_train.generate_add_samples(), train_epoch_size, args.epochs, 
                            validation_data=keras_gen_train.generate_add_samples('test'), validation_steps=test_epoch_size, callbacks = [tensorboard], workers = 0) 
 #workers = 0 is a work around to correct the number of calls to the validation_data generator
+for l in range(len(lambda_model_layers)):
+    lambda_model_layers[l].trainable = True
 lambda_model.save(args.final_name+'.hdf5')
 
 # tools for debugging