def __init__(self,
subphase,
num_samples,
batch_size=32,
deterministic=False):
"""Dataset for random actions.
Arguments:
subphase (str): Attack subphase to collect. TODO FIXME Cover more than just attacks
num_samples (int): Approximate number of samples to gather.
deterministic(bool): Make each iteration produce the same results.
"""
super(RandomActionDataset).__init__()
self.subphase = subphase
self.num_samples = num_samples
self.batch_size = batch_size
self.deterministic = deterministic
# Variables for data generation
self.randagent = RandomAgent()
keys, ship_templates = parseShips('data/test_ships.csv')
training_ships = [
"All Defense Tokens", "All Defense Tokens",
"Imperial II-class Star Destroyer", "MC80 Command Cruiser",
"Assault Frigate Mark II A", "No Shield Ship", "One Shield Ship",
"Mega Die Ship"
]
self.defenders = []
self.attackers = []
for name in training_ships:
self.attackers.append(
Ship(name=name,
template=ship_templates[name],
upgrades=[],
player_number=1,
device='cpu'))
for name in training_ships:
self.defenders.append(
Ship(name=name,
template=ship_templates[name],
upgrades=[],
player_number=2,
device='cpu'))
import random
import torch
import utility
from armada_encodings import (Encodings)
from game_constants import (ArmadaPhases, ArmadaTypes)
from learning_agent import (LearningAgent)
from learning_components import (collect_attack_batches, get_n_examples)
from model import (SeparatePhaseModel)
from random_action_dataset import (RandomActionDataset)
from random_agent import (RandomAgent)
from ship import (Ship)
from world_state import (AttackState, WorldState)
# Initialize ships from the test ship list
keys, ship_templates = utility.parseShips('data/test_ships.csv')
# Test the defense tokens by comparing the results of the test ships with and without those tokens
def update_lifetime_network(lifenet, batch, labels, optimizer, eval_only=False):
"""Do a forward and backward pass through the given lifetime network.
Args:
lifenet (torch.nn.Module): Trainable torch model
batch (torch.tensor) : Training batch
labels (torch.tensor) : Training labels
optimizer (torch.nn.Optimizer) : Optimizer for lifenet parameters.
eval_only (bool) : Only evaluate, don't update parameters.
Returns:
batch error : Average absolute error for this batch
import math
import PyGnuplot as gp
from game_constants import ArmadaDimensions
from utility import get_corners, parseShips, ruler_distance
from ship import Ship
keys, ship_templates = parseShips('data/test_ships.csv')
# Make two ships
alice = Ship(name="Alice",
template=ship_templates["Attacker"],
upgrades=[],
player_number=1)
bob = Ship(name="Bob",
template=ship_templates["Attacker"],
upgrades=[],
player_number=2)
# Put them somewhere
alice.set(name="location", value=[1.5, 1.5])
alice.set(name="heading", value=math.pi / 2.)
print(f"Alice location is {alice.get_range('location')}")
bob.set(name="location", value=[1.5, 0.8])
bob.set(name="heading", value=math.pi / 4.)
print(f"Bob location is {bob.get_range('location')}")
# Get the distance
distance, path = ruler_distance(alice, bob)
# Print out the ship edges and the shortest path
print(f"Distance is {distance}")
required=True,
help='Name of a ship or "all"')
parser.add_argument('--ship2',
type=str,
required=True,
help='Name of a ship or "all"')
parser.add_argument('--ranges',
type=str,
nargs='+',
required=True,
help='Ranges (short, medium, long)')
# TODO Allow specification of hull zones
args = parser.parse_args()
keys, ship_templates = utility.parseShips('data/armada-ship-stats.csv')
#print("keys are", keys)
#print("ships are ", ship_templates)
first_ship_names = []
if 'all' == args.ship1:
first_ship_names = [name for name in ship_templates.keys()]
else:
first_ship_names = [args.ship1]
if args.ship1 not in ship_templates.keys():
print("Unrecognized ship name {}".format(args.ship1))
print("Recognized ship names are:\n")
for name in ship_templates.keys():
print("\t{}".format(name))
exit(1)