# Load the terminals symbols of the grammar
with open("terminals.txt", "r") as file:
unique_draw = file.readlines()
for index, e in enumerate(unique_draw):
unique_draw[index] = e[0:-1]
# RNN decoder
imitate_net = ImitateJoint(hd_sz=config.hidden_size,
input_size=config.input_size,
encoder=encoder_net,
mode=config.mode,
num_draws=len(unique_draw),
canvas_shape=config.canvas_shape)
imitate_net.cuda()
imitate_net.epsilon = config.eps
if config.preload_model:
print("pre loading model")
pretrained_dict = torch.load(config.pretrain_modelpath)
imitate_net_dict = imitate_net.state_dict()
pretrained_dict = {
k: v
for k, v in pretrained_dict.items() if k in imitate_net_dict
}
imitate_net_dict.update(pretrained_dict)
imitate_net.load_state_dict(imitate_net_dict)
for param in imitate_net.parameters():
param.requires_grad = True
# Load the terminals symbols of the grammar
with open("terminals.txt", "r") as file:
unique_draw = file.readlines()
for index, e in enumerate(unique_draw):
unique_draw[index] = e[0:-1]
imitate_net = ImitateJoint(hd_sz=config.hidden_size,
input_size=config.input_size,
encoder=encoder_net,
mode=config.mode,
num_draws=len(unique_draw),
canvas_shape=config.canvas_shape)
imitate_net.cuda()
imitate_net.epsilon = 0
test_size = 3000
# This is to find top-1 performance.
paths = [config.pretrain_modelpath]
save_viz = False
for p in paths:
print(p, flush=True)
config.pretrain_modelpath = p
image_path = "data/cad/predicted_images/{}/top_1_prediction/images/".format(
p.split("/")[-1])
expressions_path = "data/cad/predicted_images/{}/top_1_prediction/expressions/".format(
p.split("/")[-1])
results_path = "data/cad/predicted_images/{}/top_1_prediction/".format(