tony/tensorneat-mend
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update hyperneat and related examples

Browse Source
This commit is contained in:
root
2024-07-11 15:08:02 +08:00
parent 9bad577d89
commit 3cb5fbf581
7 changed files with 102 additions and 136 deletions
Download Patch File Download Diff File Expand all files Collapse all files

60
examples/brax/walker.py
View File

@@ -1,60 +0,0 @@
from pipeline import Pipeline
from algorithm.neat import *
from problem.rl_env import BraxEnv
from tensorneat.common import Act
import jax, jax.numpy as jnp
def split_right_left(randkey, forward_func, obs):
right_obs_keys = jnp.array([2, 3, 4, 11, 12, 13])
left_obs_keys = jnp.array([5, 6, 7, 14, 15, 16])
right_action_keys = jnp.array([0, 1, 2])
left_action_keys = jnp.array([3, 4, 5])
right_foot_obs = obs
left_foot_obs = obs
left_foot_obs = left_foot_obs.at[right_obs_keys].set(obs[left_obs_keys])
left_foot_obs = left_foot_obs.at[left_obs_keys].set(obs[right_obs_keys])
right_action, left_action = jax.vmap(forward_func)(jnp.stack([right_foot_obs, left_foot_obs]))
# print(right_action.shape)
# print(left_action.shape)
return jnp.concatenate([right_action, left_action])
if __name__ == "__main__":
pipeline = Pipeline(
algorithm=NEAT(
species=DefaultSpecies(
genome=DefaultGenome(
num_inputs=17,
num_outputs=3,
max_nodes=50,
max_conns=100,
node_gene=DefaultNodeGene(
activation_options=(Act.tanh,),
activation_default=Act.tanh,
),
output_transform=Act.tanh,
),
pop_size=1000,
species_size=10,
),
),
problem=BraxEnv(
env_name="walker2d",
max_step=1000,
action_policy=split_right_left
),
generation_limit=10000,
fitness_target=5000,
)
# initialize state
state = pipeline.setup()
# print(state)
# run until terminate
state, best = pipeline.auto_run(state)

51
examples/brax/walker2d.py Normal file
View File

@@ -0,0 +1,51 @@
from tensorneat.pipeline import Pipeline
from tensorneat.algorithm.neat import NEAT
from tensorneat.genome import DefaultGenome, BiasNode
from tensorneat.problem.rl_env import BraxEnv
from tensorneat.common import Act, Agg
import jax, jax.numpy as jnp
def random_sample_policy(randkey, obs):
return jax.random.uniform(randkey, (6,))
if __name__ == "__main__":
pipeline = Pipeline(
algorithm=NEAT(
pop_size=1000,
species_size=20,
survival_threshold=0.1,
compatibility_threshold=1.0,
genome=DefaultGenome(
max_nodes=100,
max_conns=200,
num_inputs=17,
num_outputs=6,
init_hidden_layers=(),
node_gene=BiasNode(
activation_options=Act.tanh,
aggregation_options=Agg.sum,
),
output_transform=Act.standard_tanh,
),
),
problem=BraxEnv(
env_name="walker2d",
max_step=1000,
obs_normalization=True,
sample_episodes=1000,
sample_policy=random_sample_policy,
),
seed=42,
generation_limit=100,
fitness_target=5000,
)
# initialize state
state = pipeline.setup()
# print(state)
# run until terminate
state, best = pipeline.auto_run(state)

54
examples/func_fit/xor_hyperneat.py
View File

@@ -1,53 +1,33 @@
from pipeline import Pipeline
from algorithm.neat import *
from algorithm.hyperneat import *
from tensorneat.pipeline import Pipeline
from tensorneat.algorithm.neat import NEAT
from tensorneat.algorithm.hyperneat import HyperNEAT, FullSubstrate
from tensorneat.genome import DefaultGenome
from tensorneat.common import Act
from problem.func_fit import XOR3d
from tensorneat.problem.func_fit import XOR3d
if __name__ == "__main__":
pipeline = Pipeline(
algorithm=HyperNEAT(
substrate=FullSubstrate(
input_coors=[(-1, -1), (0.333, -1), (-0.333, -1), (1, -1)], # 3(XOR3d inputs) + 1(bias)
hidden_coors=[
(-1, -0.5), (0.333, -0.5), (-0.333, -0.5),
(1, -0.5),
(-1, 0),
(0.333, 0),
(-0.333, 0),
(1, 0),
(-1, 0.5),
(0.333, 0.5),
(-0.333, 0.5),
(1, 0.5),
],
output_coors=[
(0, 1), # one output
],
input_coors=((-1, -1), (-0.33, -1), (0.33, -1), (1, -1)),
hidden_coors=((-1, 0), (0, 0), (1, 0)),
output_coors=((0, 1),),
),
neat=NEAT(
species=DefaultSpecies(
genome=DefaultGenome(
num_inputs=4, # [*coor1, *coor2]
num_outputs=1, # the weight of connection between two coor1 and coor2
max_nodes=50,
max_conns=100,
node_gene=DefaultNodeGene(
activation_default=Act.tanh,
activation_options=(Act.tanh,),
),
output_transform=Act.tanh, # the activation function for output node in NEAT
),
pop_size=1000,
species_size=10,
compatibility_threshold=2,
survival_threshold=0.03,
pop_size=10000,
species_size=20,
survival_threshold=0.01,
genome=DefaultGenome(
num_inputs=4, # size of query coors
num_outputs=1,
init_hidden_layers=(),
output_transform=Act.standard_tanh,
),
),
activation=Act.tanh,
activate_time=10,
output_transform=Act.sigmoid, # the activation function for output node in HyperNEAT
output_transform=Act.standard_sigmoid,
),
problem=XOR3d(),
generation_limit=300,