print(class_names) print(dataset_sizes) return dataloaders, dataset_sizes if __name__ == "__main__": parser = argparse.ArgumentParser(description='test') parser.add_argument("model", type=str, help="test_model") args = parser.parse_args() device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") model = Net() model.load_state_dict(torch.load('../donemodel/' + args.model)) print("test model is ", args.model) model.eval() batch_size = 8 dataloaders, dataset_sizes = data_process(batch_size) # model.to(device) preprocessing = dict(mean=[0, 0, 0], std=[1, 1, 1], axis=-3) fmodel = foolbox.models.PyTorchModel(model.eval(), bounds=(0, 1), num_classes=16, preprocessing=preprocessing) correct = 0 total = 0 kk = 0 torch.manual_seed(12345) ten = 0
all_loss = nn.CrossEntropyLoss(reduction='none')(model(X + delta), y) max_delta[all_loss >= max_loss] = delta.detach()[all_loss >= max_loss] max_loss = torch.max(max_loss, all_loss) return max_delta if __name__ == "__main__": parser = argparse.ArgumentParser(description='test') parser.add_argument("model", type=str, help="test_model") args = parser.parse_args() model = Net() model.load_state_dict(torch.load('../donemodel/' + args.model)) print("test model is ", args.model) model.eval() batch_size = 1 dataloaders, dataset_sizes = data_process_lisa(batch_size) device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") model.to(device) eps = [0.5, 1, 1.5, 2, 2.5, 3] # eps is epsilon of the l_2 bound alpha = [0.05, 0.1, 0.15, 0.2, 0.25, 0.3] # alpha is learning rate itera = [20, 20, 20, 20, 20, 20] # iterations to find optimal restart = [ 1, 1, 1, 1, 1, 1 ] # restart times, since we just do some standard check of our model, # we do not use mutliple restarts, but you can change that if you want # delete some hyperparmeters could speed up for i in range(len(eps)):
class AnnModel(UtilityModel): def __init__(self, path): super().__init__('ann') self.pot_rewards = { 11: 300, # Insure the agent values the goal suit. Handle seeing 11 cards elsewhere 10: 10 * 10 + 100, # [cards] * 10 + [10 pot] 9: 9 * 10 + 100, # [cards] * 10 + [10 pot] 8: 8 * 10 + .33 * 120 + .67 * 100, # [cards] * 10 + [prob. 8 goal] * [8 pot] + [prob. 10 goal] * [10 pot] 7: 7 * 10 + .33 * 120 + .67 * 100, # [cards] * 10 + [prob. 8 goal] * [8 pot] + [prob. 10 goal] * [10 pot] 6: 6 * 10 + .33 * 120 + .67 * 100, # [cards] * 10 + [prob. 8 goal] * [8 pot] + [prob. 10 goal] * [10 pot] # [other hands 10] = C(n+r-1,r-1)) where n = 5 and r = 3 so [other hands 10] = 21 # [cards] * 10 + [prob 8 goal] * [8 pot] + [prob 10 goal] * (3/[other hands 10] * [10 pot]/2 + ([other hands 10] - 3)/[other hands 10] * [10 pot]]) 5: 5 * 10 + .33 * 120 + .67 * (3/21 * 100/2 + 18/21 * 100), # [other hands 10] = C(n+r-1,r-1)) where n = 6 and r = 3 so [other hands 10] = 28 # [other hands 8] = C(n+r-1,r-1)) where n = 4 and r = 3 so [other hands 8] = 15 # [cards] * 10 + [prob 8 goal] * (3/[other hands 8] * [8 pot] / 2 + ([other hands 8] - 3)/[other hands 8]) + [prob 10 goal] * (3/[other hands] * [10 pot]/2 * ([other hands 10] - 6)/[other hands 10] * [10 pot]) 4: 4 * 10 + .33 * (3/15 * 120/2 + 12/15 * 120) + .67 * (3/28 * 100/2 + 22/28 * 100), # [other hands 10] = C(n+r-1,r-1)) where n = 7 and r = 3 so [other hands 10] = 36 # [other hands 8] = C(n+r-1,r-1)) where n = 5 and r = 3 so [other hands 8] = 21 # [cards] * 10 + [prob 8 goal] * (3/[other hands 8] * [8 pot] / 3) + [prob 10 goal] * (3/[other hands 10] * [10 pot] / 2) 3: 3 * 10 + .33 * (3/21 * 120/3) + .67 * (3/36 * 100 / 2), # [other hands 8] = C(n+r-1,r-1)) where n = 6 and r = 3 so [other hands 8] = 28 # [cards] * 10 + [prob 8 goal] * (1/[other hands 8] * [8 pot] / 4) 2: 2 * 10 + .33 * (1/28 * 120/4), 1: 10, # 1 * 10 0: 0, # 0 * 10 } self.model = Net(16, 32, 64) # TODO don't hard code these self.model.load_state_dict(torch.load(path)) self.model.eval() def get_card_values(self, figgie: Figgie, index: int) -> np.ndarray: hand = figgie.cards[index] result = np.zeros(4, dtype=int) for s in SUITS: if hand[s.value] > 10: goal_suit = s.opposite() result[goal_suit.value] = 100 return result for s in SUITS: result[s.value] = (self.pot_rewards[hand[s.value] + 1] - self.pot_rewards[hand[s.value]]) input = np.array([ [cards for cards in figgie.cards[figgie.active_player]], [market.buying_price if market.buying_price is not None else 0 for market in figgie.markets], [market.selling_price if market.selling_price is not None else 0 for market in figgie.markets], [market.last_price if market.last_price is not None else 0 for market in figgie.markets], [market.operations for market in figgie.markets], [market.transactions for market in figgie.markets], ], dtype=np.float32).flatten() input = torch.from_numpy(input).view(-1, 24) percents = self.model(input) percents = torch.exp(percents) percents = percents.detach().numpy().flatten() return (percents * result).astype(int)