def train(config, checkpoint_dir=None):
trainer = PPOTrainer(config=config, env='BomberMan-v0')
init_w = trainer.get_policy('policy_01').get_weights()
trainer.restore(
'C:\\Users\\Florian\\ray_results\\PPO_BomberMan-v0_2021-03-25_08-56-43eo23nmho\\checkpoint_002360\\checkpoint-2360'
)
trainer.workers.foreach_worker(
lambda w: w.get_policy('policy_01').set_weights(init_w))
trainer.restore('.\\kill-policy-0\\checkpoint')
trainer.import_model()
iter = 0
#def update_phase(ev):
# ev.foreach_env(lambda e: e.set_phase(phase))
while True:
iter += 1
result = trainer.train()
if iter % 200 == 0:
if not os.path.exists(f'./model-{iter}'):
#trainer.get_policy('policy_01').export_model(f'./model-{iter}')
trainer.export_policy_model(f'./model-{iter}/main',
'policy_01')
trainer.export_policy_model(f'./model-{iter}/collect',
'policy_collect')
trainer.export_policy_model(f'./model-{iter}/destroy',
'policy_destroy')
trainer.export_policy_model(f'./model-{iter}/kill',
'policy_kill')
else:
print("model already saved")
def train(config, checkpoint_dir=None):
trainer = PPOTrainer(config=config, env='BomberMan-v0')
#trainer.restore('C:\\Users\\Florian\\ray_results\\PPO_BomberMan-v0_2021-03-22_10-57-05mz9533ge\\checkpoint_000140\\checkpoint-140')
iter = 0
#def update_phase(ev):
# ev.foreach_env(lambda e: e.set_phase(phase))
while True:
iter += 1
result = trainer.train()
if iter % 250 == 1:
if not os.path.exists(f'./model-{iter}-ckpt'):
#trainer.export_policy_model(f'./model-{iter}/kill', 'policy_kill')
trainer.export_model('h5',f'./model-{iter}')
else:
trainer.import_model(f'./model-{iter}')
print("model already saved")
