def run(folder):
dataset = cifar100and20()
expertf = Expert.linear
gatef = Gate.one_hidden_layer
all_num_experts = [20, 16, 12, 10, 8, 6, 4, 3, 2, 1]
epochs = 20
z_size = 20
results = experiment_numE_loop(all_num_experts, dataset, epochs, gatef,
expertf, folder)
result2 = experiment_gatef_loop({"predesigned": Gate.predesigned},
dataset, epochs, z_size, expertf,
folder)
plot.special_num_E_accuracy(
results["acc"], result2["acc"]["predesigned"], z_size,
folder + "numE_accuracy_with_additional_predesigned_gate.png")
for number_of_experts, activation in results["class activation"].items(
):
for i in range(number_of_experts):
print("plot activation", i, "/", number_of_experts)
plot.plot_activation_of_expert(
activation, i, folder + " expert %i-%i activation.png" %
(i, number_of_experts))
def run(folder):
dataset = mixture_data()
expertf = Expert.convolutional
gates = {
"no-gate": Gate.no_gate,
"predesigned": Gate.predesigned,
"linear": Gate.linear,
"1-hidden": Gate.one_hidden_layer
}
epochs = 20
num_experts = 2
result = experiment_gatef_loop(gates, dataset, epochs, num_experts,
expertf, folder)
for gname, activation in result["class activation"].items():
for i in [0, 1]:
plot.plot_activation_of_expert(
activation, i,
folder + gname + " expert %i activation.png" % i)