-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsmb1Py_runner_profile.py
180 lines (147 loc) · 7.07 KB
/
smb1Py_runner_profile.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
from functools import partial
import random
from game_runner_neat import GameRunner
from runnerConfiguration import RunnerConfig, IOData
from baseGame import EvalGame
from games.smb1Py.py_mario_bros.PythonSuperMario_master.smb_game import SMB1Game
from games.smb1Py.py_mario_bros.PythonSuperMario_master.source.states.segmentGenerator import SegmentGenerator,GenerationOptions
import os
import neat
import multiprocessing
from training_data import TrainingDataManager
try:
import cPickle as pickle
except:
import pickle
from games.smb1Py.py_mario_bros.PythonSuperMario_master.source import tools
from games.smb1Py.py_mario_bros.PythonSuperMario_master.source import constants as c
import run_states as run_states
# import cProfile as profile
steps_threshold = 800;
def task_obstruction_score(obstructions):
return -obstructions[0];
def getFitness(inputs):
obstructions = inputs['task_obstructions'];
return task_obstruction_score(obstructions) + inputs['player_state'] + inputs['task_reached']*50 - inputs['steps']*5;
def getRunning(inputs):
return (not(inputs['done']) and (not inputs['stillness_time'] > steps_threshold));
def generate_data(instructions:list[tuple[GenerationOptions,int]],shuffle=True):
data = []
for options,quantity in instructions:
data += SegmentGenerator.generateBatch(options,quantity);
if shuffle:
random.shuffle(data);
return data;
NAME = "smb1Py";
if __name__ == "__main__":
import gc
multiprocessing.freeze_support();
run_state = run_states.CONTINUE;
currentRun = 10;
override_config = True;
manual_continue_generation = None;
diff_inputs = False;
diff_outputs = False;
output_map = None;
prevData = ['player_state',
IOData('vel','array',array_size=[2]),
IOData('task_position_offset','array',array_size=[2]),
IOData('pos','array',array_size=[2])];
inputOptions = c.COLLISION_GRID;
reRunGeneration = 1523;
# reRunId = 88;
customGenome = None;
##TRAINING_DATA##
set_data = True;
add_data = False;
start_data_index = 0
additional_data_indices = [3];
configs = [
GenerationOptions(num_blocks=0,ground_height=7,valid_task_blocks=c.FLOOR,valid_start_blocks=c.FLOOR), #0
GenerationOptions(num_blocks=0,ground_height=7,valid_task_blocks=c.INNER,valid_start_blocks=c.FLOOR), #1
GenerationOptions(num_blocks=(1,3),ground_height=7,valid_task_blocks=c.INNER,valid_start_blocks=c.FLOOR), #2
GenerationOptions(num_blocks=(0,4),ground_height=7,task_batch_size=(1,4)), #3
GenerationOptions(num_blocks=(0,8),ground_height=(7,8),task_batch_size=(1,4)), #4
GenerationOptions(num_blocks=(0,4),ground_height=7,task_batch_size=(1,4),num_gaps=(1,2),gap_width=(1,2)), #5
GenerationOptions(num_blocks=(0,6),ground_height=7,task_batch_size=(1,4),num_gaps=(1,2),gap_width=(1,4)), #6
GenerationOptions(num_blocks=(0,4),ground_height=7,task_batch_size=(1,4),num_gaps=(1,2),gap_width=(1,3),allow_gap_under_start=True), #7
GenerationOptions(num_blocks=(0,6),ground_height=7,task_batch_size=(1,3),num_enemies={c.ENEMY_TYPE_GOOMBA:1},valid_enemy_positions=c.GROUNDED), #8
GenerationOptions(size=(20,15),inner_size=(14,9),num_blocks=(0,8),ground_height=(7,8),task_batch_size=(1,4)), #9
GenerationOptions(size=(18,14),inner_size=(12,8),num_blocks=(0,6),ground_height=7,task_batch_size=(1,4),num_gaps=(1,3),gap_width=(1,3)), #10
];
orders = [(configs[4],6)];
tdManager = TrainingDataManager(NAME,17,generation_func=partial(generate_data,orders));
if (run_state == run_states.NEW or set_data):
data = generate_data(orders);
tdManager.set_data(data);
if add_data:
for idx in additional_data_indices:
tdManager.add_data(SegmentGenerator.generateBatch(configs[idx],20));
inputData = [
'player_state',
IOData('vel','array',array_size=[2]),
IOData('task_position_offset','array',array_size=[2]),
IOData('pos','array',array_size=[2])];
config_suffix = "-nogrid"
if inputOptions == c.FULL:
inputData += [
IOData('collision_grid','array',[15,15]),
IOData('enemy_grid','array',[15,15]),
IOData('box_grid','array',[15,15]),
IOData('brick_grid','array',[15,15]),
IOData('powerup_grid','array',[15,15])]
config_suffix = "-full"
if inputOptions == c.COLLISION_GRID:
inputData.append(IOData('collision_grid','array',[15,15]))
config_suffix = "-blockgrid"
runConfig = RunnerConfig(
getFitness,
getRunning,
logging=True,
parallel=True,
gameName=NAME,
returnData=inputData,
num_trials=1,
num_generations=None,
training_data=tdManager);
runConfig.reporters = [tdManager];
runConfig.tile_scale = 2;
runConfig.view_distance = 3.75;
runConfig.task_obstruction_score = task_obstruction_score;
runConfig.external_render = False;
runConfig.parallel_processes = 6;
runConfig.chunkFactor = 24;
runConfig.saveFitness = True;
runConfig.logPath = f'logs\\smb1Py\\run-{currentRun}-log.txt';
runConfig.fitness_collection_type='delta_max';
print(runConfig.gameName);
game = EvalGame(SMB1Game);
runner = GameRunner(game,runConfig);
config_path = os.path.join(os.path.dirname(__file__), 'configs','config-pygame-smb1' + config_suffix);
config = neat.Config(neat.DefaultGenome, neat.DefaultReproduction,
neat.DefaultSpeciesSet, neat.DefaultStagnation,
config_path);
config_transfer = None;
if override_config:
config_transfer = (config, runConfig.get_input_transfer(prevData) if diff_inputs else None, output_transfer if diff_outputs else None)
if (run_state == run_states.EVAL_CUSTOM):
customGenome = neat.genome.DefaultGenome(0);
customGenome.configure_new(config.genome_config);
customGenome.add_connection(config.genome_config,-1,2,-1,True);
customGenome.add_connection(config.genome_config,-1,3,1,True);
if (run_state == run_states.CONTINUE):
winner = eval("runner.continue_run('run_' + str(currentRun),manual_generation=manual_continue_generation,manual_config_override=config_transfer)");
print('\nBest genome:\n{!s}'.format(winner));
else:
local_dir = os.path.dirname(__file__)
if (run_state == run_states.NEW):
winner = runner.run(config,'run_' + str(currentRun));
print('\nBest genome:\n{!s}'.format(winner))
if (run_state == run_states.RERUN):
runner.replay_best(reRunGeneration,config,'run_' + str(currentRun),net=True,randomReRoll=True);
if (run_state == run_states.RERUN_ALL):
runner.replay_generation(reRunGeneration,'run_' + str(currentRun));
if (run_state == run_states.RERUN_ID):
runner.render_genome_by_id(reRunId,reRunGeneration,config,'run_' + str(currentRun),net=True);
if (run_state == run_states.EVAL_CUSTOM):
runner.render_custom_genome_object(customGenome,config,net=True)