def __init__(self, params):
super(KerasGraphFF3, self).__init__(params)
# declare graph_params and update from dict --graph_params
self.graph_params["ff_hidden_1"] = 128
self.graph_params["ff_hidden_2"] = 128
self.graph_params["ff_hidden_3"] = 128
self.graph_params = update_params(self.graph_params,
self._flags.graph_params, "graph")
# initilize keras layer
self._tracked_layers["add_layer"] = tf.keras.layers.Add()
self._tracked_layers["flatten_1"] = tf.keras.layers.Flatten()
self._tracked_layers["ff_layer_1"] = tf.keras.layers.Dense(
self.graph_params["ff_hidden_1"],
activation=tf.nn.leaky_relu,
name="ff_layer_1")
self._tracked_layers["ff_layer_2"] = tf.keras.layers.Dense(
self.graph_params["ff_hidden_2"],
activation=tf.nn.leaky_relu,
name="ff_layer_2")
self._tracked_layers["ff_layer_3"] = tf.keras.layers.Dense(
self.graph_params["ff_hidden_3"],
activation=tf.nn.leaky_relu,
name="ff_layer_3")
self._tracked_layers["last_layer"] = tf.keras.layers.Dense(
6, activation=None, name="last_layer")
self._tracked_layers["flatten_2"] = tf.keras.layers.Flatten()
def __init__(self, params):
super(GraphConv2MultiFF, self).__init__(params)
# v0.1
self.graph_params["mid_layer_activation"] = "leaky_relu"
self.graph_params["conv_layer_activation"] = "leaky_relu"
self.graph_params["input_dropout"] = 0.0
self.graph_params["batch_norm"] = False
self.graph_params["dense_layers"] = [256, 128]
self.graph_params = update_params(self.graph_params,
self._flags.graph_params, "graph")
def __init__(self, params):
super(GraphConv2MultiFFTriangle, self).__init__(params)
# v0.2 # raise version if adjusting default values
self.graph_params["dense_layers"] = [512, 256, 128, 64, 32]
self.graph_params["nhidden_dense_final"] = 6
self.graph_params["mid_layer_activation"] = "leaky_relu"
self.graph_params["conv_layer_activation"] = "leaky_relu"
self.graph_params["batch_norm"] = False
self.graph_params["edge_classifier"] = False
self.graph_params["nhidden_max_edges"] = 6
self.graph_params = update_params(self.graph_params,
self._flags.graph_params, "graph")
def __init__(self, params):
super(GraphMultiFF, self).__init__(params)
# v0.2
self.graph_params["mid_layer_activation"] = "leaky_relu"
self.graph_params["batch_norm"] = False
self.graph_params["dense_layers"] = [512, 256, 128, 64]
self.graph_params["dense_dropout"] = [
] # [0.0, 0.0] dropout after each dense layer
self.graph_params["input_dropout"] = 0.01
self.graph_params["abs_as_input"] = False
self.graph_params = update_params(self.graph_params,
self._flags.graph_params, "graph")
def __init__(self, params):
super(GraphMultiFF, self).__init__(params)
# v0.3
if not self._flags.complex_phi:
self.fc_size_0 = 3
else:
self.fc_size_0 = 4
self.graph_params["dense_layers"] = [512, 256, 128, 64, 32]
self.graph_params["input_dropout"] = 0.0
self.graph_params["ff_dropout"] = 0.0
self.graph_params["uniform_noise"] = 0.0
self.graph_params["normal_noise"] = 0.0
self.graph_params["nhidden_dense_final"] = 6
self.graph_params["edge_classifier"] = False
self.graph_params["batch_norm"] = False
self.graph_params["nhidden_max_edges"] = 6
self.graph_params = update_params(self.graph_params,
self._flags.graph_params, "graph")
def print_params(self):
self.graph_params = update_params(self.graph_params,
self._flags.graph_params, "graph")
print("{}_params:".format("graph"))
for i, j in enumerate(self.graph_params):
print(" {}: {}".format(j, self.graph_params[j]))
def update_params(self):
"""Updating of the default params if provided via flags as a dict"""
self._optimizer_params = update_params(self._optimizer_params, self._flags.optimizer_params, "Optimizer")