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

change project structure and using .ini as config file

Browse Source
This commit is contained in:
wls2002
2023-06-15 11:05:26 +08:00
parent 47fb0151f4
commit acedd67617
30 changed files with 97 additions and 301 deletions
Download Patch File Download Diff File Expand all files Collapse all files

0
algorithms/__init__.py
View File

0
algorithms/hyper_neat/__init__.py
View File

1
algorithms/neat/__init__.py
View File

@@ -1 +0,0 @@
from .pipeline import Pipeline

323
algorithms/neat/function_factory.py
View File

@@ -1,323 +0,0 @@
"""
Lowers, compiles, and creates functions used in the NEAT pipeline.
"""
from functools import partial
import time
import numpy as np
from jax import jit, vmap
from .genome import act_name2key, agg_name2key, initialize_genomes
from .genome import topological_sort, forward_single, unflatten_connections
from .population import create_next_generation_then_speciate
class FunctionFactory:
def __init__(self, config):
self.config = config
self.expand_coe = config.basic.expands_coe
self.precompile_times = config.basic.pre_compile_times
self.compiled_function = {}
self.compile_time = 0
self.load_config_vals(config)
self.create_topological_sort_with_args()
self.create_single_forward_with_args()
self.create_update_speciate_with_args()
def load_config_vals(self, config):
self.compatibility_threshold = self.config.neat.species.compatibility_threshold
self.problem_batch = config.basic.problem_batch
self.pop_size = config.neat.population.pop_size
self.disjoint_coe = config.neat.genome.compatibility_disjoint_coefficient
self.compatibility_coe = config.neat.genome.compatibility_weight_coefficient
self.num_inputs = config.basic.num_inputs
self.num_outputs = config.basic.num_outputs
self.input_idx = np.arange(self.num_inputs)
self.output_idx = np.arange(self.num_inputs, self.num_inputs + self.num_outputs)
bias = config.neat.gene.bias
self.bias_mean = bias.init_mean
self.bias_std = bias.init_stdev
self.bias_mutate_strength = bias.mutate_power
self.bias_mutate_rate = bias.mutate_rate
self.bias_replace_rate = bias.replace_rate
response = config.neat.gene.response
self.response_mean = response.init_mean
self.response_std = response.init_stdev
self.response_mutate_strength = response.mutate_power
self.response_mutate_rate = response.mutate_rate
self.response_replace_rate = response.replace_rate
weight = config.neat.gene.weight
self.weight_mean = weight.init_mean
self.weight_std = weight.init_stdev
self.weight_mutate_strength = weight.mutate_power
self.weight_mutate_rate = weight.mutate_rate
self.weight_replace_rate = weight.replace_rate
activation = config.neat.gene.activation
self.act_default = act_name2key[activation.default]
self.act_list = np.array([act_name2key[name] for name in activation.options])
self.act_replace_rate = activation.mutate_rate
aggregation = config.neat.gene.aggregation
self.agg_default = agg_name2key[aggregation.default]
self.agg_list = np.array([agg_name2key[name] for name in aggregation.options])
self.agg_replace_rate = aggregation.mutate_rate
enabled = config.neat.gene.enabled
self.enabled_reverse_rate = enabled.mutate_rate
genome = config.neat.genome
self.add_node_rate = genome.node_add_prob
self.delete_node_rate = genome.node_delete_prob
self.add_connection_rate = genome.conn_add_prob
self.delete_connection_rate = genome.conn_delete_prob
self.single_structure_mutate = genome.single_structural_mutation
def create_initialize(self, N, C):
func = partial(
initialize_genomes,
pop_size=self.pop_size,
N=N,
C=C,
num_inputs=self.num_inputs,
num_outputs=self.num_outputs,
default_bias=self.bias_mean,
default_response=self.response_mean,
default_act=self.act_default,
default_agg=self.agg_default,
default_weight=self.weight_mean
)
return func
def create_update_speciate_with_args(self):
species_kwargs = {
"disjoint_coe": self.disjoint_coe,
"compatibility_coe": self.compatibility_coe,
"compatibility_threshold": self.compatibility_threshold
}
mutate_kwargs = {
"input_idx": self.input_idx,
"output_idx": self.output_idx,
"bias_mean": self.bias_mean,
"bias_std": self.bias_std,
"bias_mutate_strength": self.bias_mutate_strength,
"bias_mutate_rate": self.bias_mutate_rate,
"bias_replace_rate": self.bias_replace_rate,
"response_mean": self.response_mean,
"response_std": self.response_std,
"response_mutate_strength": self.response_mutate_strength,
"response_mutate_rate": self.response_mutate_rate,
"response_replace_rate": self.response_replace_rate,
"weight_mean": self.weight_mean,
"weight_std": self.weight_std,
"weight_mutate_strength": self.weight_mutate_strength,
"weight_mutate_rate": self.weight_mutate_rate,
"weight_replace_rate": self.weight_replace_rate,
"act_default": self.act_default,
"act_list": self.act_list,
"act_replace_rate": self.act_replace_rate,
"agg_default": self.agg_default,
"agg_list": self.agg_list,
"agg_replace_rate": self.agg_replace_rate,
"enabled_reverse_rate": self.enabled_reverse_rate,
"add_node_rate": self.add_node_rate,
"delete_node_rate": self.delete_node_rate,
"add_connection_rate": self.add_connection_rate,
"delete_connection_rate": self.delete_connection_rate,
}
self.update_speciate_with_args = partial(
create_next_generation_then_speciate,
species_kwargs=species_kwargs,
mutate_kwargs=mutate_kwargs
)
def create_update_speciate(self, N, C, S):
key = ("update_speciate", N, C, S)
if key not in self.compiled_function:
self.compile_update_speciate(N, C, S)
return self.compiled_function[key]
def compile_update_speciate(self, N, C, S):
s = time.time()
func = self.update_speciate_with_args
randkey_lower = np.zeros((2,), dtype=np.uint32)
pop_nodes_lower = np.zeros((self.pop_size, N, 5))
pop_cons_lower = np.zeros((self.pop_size, C, 4))
winner_part_lower = np.zeros((self.pop_size,), dtype=np.int32)
loser_part_lower = np.zeros((self.pop_size,), dtype=np.int32)
elite_mask_lower = np.zeros((self.pop_size,), dtype=bool)
new_node_keys_start_lower = np.zeros((self.pop_size,), dtype=np.int32)
pre_spe_center_nodes_lower = np.zeros((S, N, 5))
pre_spe_center_cons_lower = np.zeros((S, C, 4))
species_keys = np.zeros((S,), dtype=np.int32)
new_species_keys_lower = 0
compiled_func = jit(func).lower(
randkey_lower,
pop_nodes_lower,
pop_cons_lower,
winner_part_lower,
loser_part_lower,
elite_mask_lower,
new_node_keys_start_lower,
pre_spe_center_nodes_lower,
pre_spe_center_cons_lower,
species_keys,
new_species_keys_lower,
).compile()
self.compiled_function[("update_speciate", N, C, S)] = compiled_func
self.compile_time += time.time() - s
def create_topological_sort_with_args(self):
self.topological_sort_with_args = topological_sort
def compile_topological_sort(self, n):
s = time.time()
func = self.topological_sort_with_args
nodes_lower = np.zeros((n, 5))
connections_lower = np.zeros((2, n, n))
func = jit(func).lower(nodes_lower, connections_lower).compile()
self.compiled_function[('topological_sort', n)] = func
self.compile_time += time.time() - s
def create_topological_sort(self, n):
key = ('topological_sort', n)
if key not in self.compiled_function:
self.compile_topological_sort(n)
return self.compiled_function[key]
def compile_topological_sort_batch(self, n):
s = time.time()
func = self.topological_sort_with_args
func = vmap(func)
nodes_lower = np.zeros((self.pop_size, n, 5))
connections_lower = np.zeros((self.pop_size, 2, n, n))
func = jit(func).lower(nodes_lower, connections_lower).compile()
self.compiled_function[('topological_sort_batch', n)] = func
self.compile_time += time.time() - s
def create_topological_sort_batch(self, n):
key = ('topological_sort_batch', n)
if key not in self.compiled_function:
self.compile_topological_sort_batch(n)
return self.compiled_function[key]
def create_single_forward_with_args(self):
func = partial(
forward_single,
input_idx=self.input_idx,
output_idx=self.output_idx
)
self.single_forward_with_args = func
def compile_batch_forward(self, n):
s = time.time()
func = self.single_forward_with_args
func = vmap(func, in_axes=(0, None, None, None))
inputs_lower = np.zeros((self.problem_batch, self.num_inputs))
cal_seqs_lower = np.zeros((n,), dtype=np.int32)
nodes_lower = np.zeros((n, 5))
connections_lower = np.zeros((2, n, n))
func = jit(func).lower(inputs_lower, cal_seqs_lower, nodes_lower, connections_lower).compile()
self.compiled_function[('batch_forward', n)] = func
self.compile_time += time.time() - s
def create_batch_forward(self, n):
key = ('batch_forward', n)
if key not in self.compiled_function:
self.compile_batch_forward(n)
return self.compiled_function[key]
def compile_pop_batch_forward(self, n):
s = time.time()
func = self.single_forward_with_args
func = vmap(func, in_axes=(0, None, None, None)) # batch_forward
func = vmap(func, in_axes=(None, 0, 0, 0)) # pop_batch_forward
inputs_lower = np.zeros((self.problem_batch, self.num_inputs))
cal_seqs_lower = np.zeros((self.pop_size, n), dtype=np.int32)
nodes_lower = np.zeros((self.pop_size, n, 5))
connections_lower = np.zeros((self.pop_size, 2, n, n))
func = jit(func).lower(inputs_lower, cal_seqs_lower, nodes_lower, connections_lower).compile()
self.compiled_function[('pop_batch_forward', n)] = func
self.compile_time += time.time() - s
def create_pop_batch_forward(self, n):
key = ('pop_batch_forward', n)
if key not in self.compiled_function:
self.compile_pop_batch_forward(n)
return self.compiled_function[key]
def ask_pop_batch_forward(self, pop_nodes, pop_cons):
n, c = pop_nodes.shape[1], pop_cons.shape[1]
batch_unflatten_func = self.create_batch_unflatten_connections(n, c)
pop_cons = batch_unflatten_func(pop_nodes, pop_cons)
ts = self.create_topological_sort_batch(n)
# for connections with enabled is false, set weight to 0)
pop_cal_seqs = ts(pop_nodes, pop_cons)
# print(pop_cal_seqs)
forward_func = self.create_pop_batch_forward(n)
def debug_forward(inputs):
return forward_func(inputs, pop_cal_seqs, pop_nodes, pop_cons)
return debug_forward
def ask_batch_forward(self, nodes, connections):
n = nodes.shape[0]
ts = self.create_topological_sort(n)
cal_seqs = ts(nodes, connections)
forward_func = self.create_batch_forward(n)
def debug_forward(inputs):
return forward_func(inputs, cal_seqs, nodes, connections)
return debug_forward
def compile_batch_unflatten_connections(self, n, c):
s = time.time()
func = unflatten_connections
func = vmap(func)
pop_nodes_lower = np.zeros((self.pop_size, n, 5))
pop_connections_lower = np.zeros((self.pop_size, c, 4))
func = jit(func).lower(pop_nodes_lower, pop_connections_lower).compile()
self.compiled_function[('batch_unflatten_connections', n, c)] = func
self.compile_time += time.time() - s
def create_batch_unflatten_connections(self, n, c):
key = ('batch_unflatten_connections', n, c)
if key not in self.compiled_function:
self.compile_batch_unflatten_connections(n, c)
return self.compiled_function[key]

9
algorithms/neat/genome/__init__.py
View File

@@ -1,9 +0,0 @@
from .genome import expand, expand_single, initialize_genomes
from .forward import forward_single
from .activations import act_name2key
from .aggregations import agg_name2key
from .crossover import crossover
from .mutate import mutate
from .distance import distance
from .graph import topological_sort
from .utils import unflatten_connections

140
algorithms/neat/genome/activations.py
View File

@@ -1,140 +0,0 @@
import jax
import jax.numpy as jnp
from jax import jit
@jit
def sigmoid_act(z):
z = jnp.clip(z * 5, -60, 60)
return 1 / (1 + jnp.exp(-z))
@jit
def tanh_act(z):
z = jnp.clip(z * 2.5, -60, 60)
return jnp.tanh(z)
@jit
def sin_act(z):
z = jnp.clip(z * 5, -60, 60)
return jnp.sin(z)
@jit
def gauss_act(z):
z = jnp.clip(z * 5, -3.4, 3.4)
return jnp.exp(-z ** 2)
@jit
def relu_act(z):
return jnp.maximum(z, 0)
@jit
def elu_act(z):
return jnp.where(z > 0, z, jnp.exp(z) - 1)
@jit
def lelu_act(z):
leaky = 0.005
return jnp.where(z > 0, z, leaky * z)
@jit
def selu_act(z):
lam = 1.0507009873554804934193349852946
alpha = 1.6732632423543772848170429916717
return jnp.where(z > 0, lam * z, lam * alpha * (jnp.exp(z) - 1))
@jit
def softplus_act(z):
z = jnp.clip(z * 5, -60, 60)
return 0.2 * jnp.log(1 + jnp.exp(z))
@jit
def identity_act(z):
return z
@jit
def clamped_act(z):
return jnp.clip(z, -1, 1)
@jit
def inv_act(z):
z = jnp.maximum(z, 1e-7)
return 1 / z
@jit
def log_act(z):
z = jnp.maximum(z, 1e-7)
return jnp.log(z)
@jit
def exp_act(z):
z = jnp.clip(z, -60, 60)
return jnp.exp(z)
@jit
def abs_act(z):
return jnp.abs(z)
@jit
def hat_act(z):
return jnp.maximum(0, 1 - jnp.abs(z))
@jit
def square_act(z):
return z ** 2
@jit
def cube_act(z):
return z ** 3
ACT_TOTAL_LIST = [sigmoid_act, tanh_act, sin_act, gauss_act, relu_act, elu_act, lelu_act, selu_act, softplus_act,
identity_act, clamped_act, inv_act, log_act, exp_act, abs_act, hat_act, square_act, cube_act]
act_name2key = {
'sigmoid': 0,
'tanh': 1,
'sin': 2,
'gauss': 3,
'relu': 4,
'elu': 5,
'lelu': 6,
'selu': 7,
'softplus': 8,
'identity': 9,
'clamped': 10,
'inv': 11,
'log': 12,
'exp': 13,
'abs': 14,
'hat': 15,
'square': 16,
'cube': 17,
}
@jit
def act(idx, z):
idx = jnp.asarray(idx, dtype=jnp.int32)
# change idx from float to int
res = jax.lax.switch(idx, ACT_TOTAL_LIST, z)
return jnp.where(jnp.isnan(res), jnp.nan, res)
# return jax.lax.switch(idx, ACT_TOTAL_LIST, z)

109
algorithms/neat/genome/aggregations.py
View File

@@ -1,109 +0,0 @@
"""
aggregations, two special case need to consider:
1. extra 0s
2. full of 0s
"""
import jax
import jax.numpy as jnp
import numpy as np
from jax import jit
@jit
def sum_agg(z):
z = jnp.where(jnp.isnan(z), 0, z)
return jnp.sum(z, axis=0)
@jit
def product_agg(z):
z = jnp.where(jnp.isnan(z), 1, z)
return jnp.prod(z, axis=0)
@jit
def max_agg(z):
z = jnp.where(jnp.isnan(z), -jnp.inf, z)
return jnp.max(z, axis=0)
@jit
def min_agg(z):
z = jnp.where(jnp.isnan(z), jnp.inf, z)
return jnp.min(z, axis=0)
@jit
def maxabs_agg(z):
z = jnp.where(jnp.isnan(z), 0, z)
abs_z = jnp.abs(z)
max_abs_index = jnp.argmax(abs_z)
return z[max_abs_index]
@jit
def median_agg(z):
non_zero_mask = ~jnp.isnan(z)
n = jnp.sum(non_zero_mask, axis=0)
z = jnp.where(jnp.isnan(z), jnp.inf, z)
sorted_valid_values = jnp.sort(z)
def _even_case():
return (sorted_valid_values[n // 2 - 1] + sorted_valid_values[n // 2]) / 2
def _odd_case():
return sorted_valid_values[n // 2]
median = jax.lax.cond(n % 2 == 0, _even_case, _odd_case)
return median
@jit
def mean_agg(z):
non_zero_mask = ~jnp.isnan(z)
valid_values_sum = sum_agg(z)
valid_values_count = jnp.sum(non_zero_mask, axis=0)
mean_without_zeros = valid_values_sum / valid_values_count
return mean_without_zeros
AGG_TOTAL_LIST = [sum_agg, product_agg, max_agg, min_agg, maxabs_agg, median_agg, mean_agg]
agg_name2key = {
'sum': 0,
'product': 1,
'max': 2,
'min': 3,
'maxabs': 4,
'median': 5,
'mean': 6,
}
@jit
def agg(idx, z):
idx = jnp.asarray(idx, dtype=jnp.int32)
def full_nan():
return 0.
def not_full_nan():
return jax.lax.switch(idx, AGG_TOTAL_LIST, z)
return jax.lax.cond(jnp.all(jnp.isnan(z)), full_nan, not_full_nan)
vectorized_agg = jax.vmap(agg, in_axes=(0, 0))
if __name__ == '__main__':
array = jnp.asarray([1, 2, np.nan, np.nan, 3, 4, 5, np.nan, np.nan, np.nan, np.nan], dtype=jnp.float32)
for names in agg_name2key.keys():
print(names, agg(agg_name2key[names], array))
array2 = jnp.asarray([0, 0, 0, 0], dtype=jnp.float32)
for names in agg_name2key.keys():
print(names, agg(agg_name2key[names], array2))

76
algorithms/neat/genome/crossover.py
View File

@@ -1,76 +0,0 @@
from functools import partial
from typing import Tuple
import jax
from jax import jit, vmap, Array
from jax import numpy as jnp
@jit
def crossover(randkey: Array, nodes1: Array, cons1: Array, nodes2: Array, cons2: Array) \
-> Tuple[Array, Array]:
"""
use genome1 and genome2 to generate a new genome
notice that genome1 should have higher fitness than genome2 (genome1 is winner!)
:param randkey:
:param nodes1:
:param cons1:
:param nodes2:
:param cons2:
:return:
"""
randkey_1, randkey_2 = jax.random.split(randkey)
# crossover nodes
keys1, keys2 = nodes1[:, 0], nodes2[:, 0]
nodes2 = align_array(keys1, keys2, nodes2, 'node')
new_nodes = jnp.where(jnp.isnan(nodes1) | jnp.isnan(nodes2), nodes1, crossover_gene(randkey_1, nodes1, nodes2))
# crossover connections
con_keys1, con_keys2 = cons1[:, :2], cons2[:, :2]
cons2 = align_array(con_keys1, con_keys2, cons2, 'connection')
new_cons = jnp.where(jnp.isnan(cons1) | jnp.isnan(cons2), cons1, crossover_gene(randkey_2, cons1, cons2))
return new_nodes, new_cons
# @partial(jit, static_argnames=['gene_type'])
def align_array(seq1: Array, seq2: Array, ar2: Array, gene_type: str) -> Array:
"""
After I review this code, I found that it is the most difficult part of the code. Please never change it!
make ar2 align with ar1.
:param seq1:
:param seq2:
:param ar2:
:param gene_type:
:return:
align means to intersect part of ar2 will be at the same position as ar1,
non-intersect part of ar2 will be set to Nan
"""
seq1, seq2 = seq1[:, jnp.newaxis], seq2[jnp.newaxis, :]
mask = (seq1 == seq2) & (~jnp.isnan(seq1))
if gene_type == 'connection':
mask = jnp.all(mask, axis=2)
intersect_mask = mask.any(axis=1)
idx = jnp.arange(0, len(seq1))
idx_fixed = jnp.dot(mask, idx)
refactor_ar2 = jnp.where(intersect_mask[:, jnp.newaxis], ar2[idx_fixed], jnp.nan)
return refactor_ar2
# @jit
def crossover_gene(rand_key: Array, g1: Array, g2: Array) -> Array:
"""
crossover two genes
:param rand_key:
:param g1:
:param g2:
:return:
only gene with the same key will be crossover, thus don't need to consider change key
"""
r = jax.random.uniform(rand_key, shape=g1.shape)
return jnp.where(r > 0.5, g1, g2)

0
algorithms/neat/genome/debug/__init__.py
View File

88
algorithms/neat/genome/debug/tools.py
View File

@@ -1,88 +0,0 @@
from collections import defaultdict
import numpy as np
def check_array_valid(nodes, cons, input_keys, output_keys):
nodes_dict, cons_dict = array2object(nodes, cons, input_keys, output_keys)
# assert is_DAG(cons_dict.keys()), "The genome is not a DAG!"
def array2object(nodes, cons, input_keys, output_keys):
"""
Convert a genome from array to dict.
:param nodes: (N, 5)
:param cons: (C, 4)
:param output_keys:
:param input_keys:
:return: nodes_dict[key: (bias, response, act, agg)], cons_dict[(i_key, o_key): (weight, enabled)]
"""
# update nodes_dict
nodes_dict = {}
for i, node in enumerate(nodes):
if np.isnan(node[0]):
continue
key = int(node[0])
assert key not in nodes_dict, f"Duplicate node key: {key}!"
if key in input_keys:
assert np.all(np.isnan(node[1:])), f"Input node {key} must has None bias, response, act, or agg!"
nodes_dict[key] = (None,) * 4
else:
assert np.all(~np.isnan(node[1:])), f"Normal node {key} must has non-None bias, response, act, or agg!"
bias = node[1]
response = node[2]
act = node[3]
agg = node[4]
nodes_dict[key] = (bias, response, act, agg)
# check nodes_dict
for i in input_keys:
assert i in nodes_dict, f"Input node {i} not found in nodes_dict!"
for o in output_keys:
assert o in nodes_dict, f"Output node {o} not found in nodes_dict!"
# update connections
cons_dict = {}
for i, con in enumerate(cons):
if np.all(np.isnan(con)):
pass
elif np.all(~np.isnan(con)):
i_key = int(con[0])
o_key = int(con[1])
if (i_key, o_key) in cons_dict:
assert False, f"Duplicate connection: {(i_key, o_key)}!"
assert i_key in nodes_dict, f"Input node {i_key} not found in nodes_dict!"
assert o_key in nodes_dict, f"Output node {o_key} not found in nodes_dict!"
weight = con[2]
enabled = (con[3] == 1)
cons_dict[(i_key, o_key)] = (weight, enabled)
else:
assert False, f"Connection {i} must has all None or all non-None!"
return nodes_dict, cons_dict
def is_DAG(edges):
all_nodes = set()
for a, b in edges:
if a == b: # cycle
return False
all_nodes.union({a, b})
for node in all_nodes:
visited = {n: False for n in all_nodes}
def dfs(n):
if visited[n]:
return False
visited[n] = True
for a, b in edges:
if a == n:
if not dfs(b):
return False
return True
if not dfs(node):
return False
return True

97
algorithms/neat/genome/distance.py
View File

@@ -1,97 +0,0 @@
from jax import jit, vmap, Array
from jax import numpy as jnp
from .utils import EMPTY_NODE, EMPTY_CON
@jit
def distance(nodes1: Array, cons1: Array, nodes2: Array, cons2: Array, disjoint_coe: float = 1.,
compatibility_coe: float = 0.5) -> Array:
"""
Calculate the distance between two genomes.
nodes are a 2-d array with shape (N, 5), its columns are [key, bias, response, act, agg]
connections are a 3-d array with shape (2, N, N), axis 0 means [weights, enable]
"""
nd = node_distance(nodes1, nodes2, disjoint_coe, compatibility_coe) # node distance
cd = connection_distance(cons1, cons2, disjoint_coe, compatibility_coe) # connection distance
return nd + cd
@jit
def node_distance(nodes1, nodes2, disjoint_coe=1., compatibility_coe=0.5):
node_cnt1 = jnp.sum(~jnp.isnan(nodes1[:, 0]))
node_cnt2 = jnp.sum(~jnp.isnan(nodes2[:, 0]))
max_cnt = jnp.maximum(node_cnt1, node_cnt2)
nodes = jnp.concatenate((nodes1, nodes2), axis=0)
keys = nodes[:, 0]
sorted_indices = jnp.argsort(keys, axis=0)
nodes = nodes[sorted_indices]
nodes = jnp.concatenate([nodes, EMPTY_NODE], axis=0) # add a nan row to the end
fr, sr = nodes[:-1], nodes[1:] # first row, second row
intersect_mask = (fr[:, 0] == sr[:, 0]) & ~jnp.isnan(nodes[:-1, 0])
non_homologous_cnt = node_cnt1 + node_cnt2 - 2 * jnp.sum(intersect_mask)
nd = batch_homologous_node_distance(fr, sr)
nd = jnp.where(jnp.isnan(nd), 0, nd)
homologous_distance = jnp.sum(nd * intersect_mask)
val = non_homologous_cnt * disjoint_coe + homologous_distance * compatibility_coe
return jnp.where(max_cnt == 0, 0, val / max_cnt)
@jit
def connection_distance(cons1, cons2, disjoint_coe=1., compatibility_coe=0.5):
con_cnt1 = jnp.sum(~jnp.isnan(cons1[:, 0]))
con_cnt2 = jnp.sum(~jnp.isnan(cons2[:, 0]))
max_cnt = jnp.maximum(con_cnt1, con_cnt2)
cons = jnp.concatenate((cons1, cons2), axis=0)
keys = cons[:, :2]
sorted_indices = jnp.lexsort(keys.T[::-1])
cons = cons[sorted_indices]
cons = jnp.concatenate([cons, EMPTY_CON], axis=0) # add a nan row to the end
fr, sr = cons[:-1], cons[1:] # first row, second row
# both genome has such connection
intersect_mask = jnp.all(fr[:, :2] == sr[:, :2], axis=1) & ~jnp.isnan(fr[:, 0])
non_homologous_cnt = con_cnt1 + con_cnt2 - 2 * jnp.sum(intersect_mask)
cd = batch_homologous_connection_distance(fr, sr)
cd = jnp.where(jnp.isnan(cd), 0, cd)
homologous_distance = jnp.sum(cd * intersect_mask)
val = non_homologous_cnt * disjoint_coe + homologous_distance * compatibility_coe
return jnp.where(max_cnt == 0, 0, val / max_cnt)
@vmap
def batch_homologous_node_distance(b_n1, b_n2):
return homologous_node_distance(b_n1, b_n2)
@vmap
def batch_homologous_connection_distance(b_c1, b_c2):
return homologous_connection_distance(b_c1, b_c2)
@jit
def homologous_node_distance(n1, n2):
d = 0
d += jnp.abs(n1[1] - n2[1]) # bias
d += jnp.abs(n1[2] - n2[2]) # response
d += n1[3] != n2[3] # activation
d += n1[4] != n2[4]
return d
@jit
def homologous_connection_distance(c1, c2):
d = 0
d += jnp.abs(c1[2] - c2[2]) # weight
d += c1[3] != c2[3] # enable
return d

47
algorithms/neat/genome/forward.py
View File

@@ -1,47 +0,0 @@
import jax
from jax import Array, numpy as jnp
from jax import jit, vmap
from .aggregations import agg
from .activations import act
from .utils import I_INT
# TODO: enabled information doesn't influence forward. That is wrong!
@jit
def forward_single(inputs: Array, cal_seqs: Array, nodes: Array, connections: Array,
input_idx: Array, output_idx: Array) -> Array:
"""
jax forward for single input shaped (input_num, )
nodes, connections are single genome
:argument inputs: (input_num, )
:argument input_idx: (input_num, )
:argument output_idx: (output_num, )
:argument cal_seqs: (N, )
:argument nodes: (N, 5)
:argument connections: (2, N, N)
:return (output_num, )
"""
N = nodes.shape[0]
ini_vals = jnp.full((N,), jnp.nan)
ini_vals = ini_vals.at[input_idx].set(inputs)
def scan_body(carry, i):
def hit():
ins = carry * connections[0, :, i]
z = agg(nodes[i, 4], ins)
z = z * nodes[i, 2] + nodes[i, 1]
z = act(nodes[i, 3], z)
new_vals = carry.at[i].set(z)
return new_vals
def miss():
return carry
return jax.lax.cond((i == I_INT) | (jnp.isin(i, input_idx)), miss, hit), None
vals, _ = jax.lax.scan(scan_body, ini_vals, cal_seqs)
return vals[output_idx]

201
algorithms/neat/genome/genome.py
View File

@@ -1,201 +0,0 @@
"""
Vectorization of genome representation.
Utilizes Tuple[nodes: Array, connections: Array] to encode the genome, where:
1. N, C are pre-set values that determines the maximum number of nodes and connections in the network, and will increase if the genome becomes
too large to be represented by the current value of N and C.
2. nodes is an array of shape (N, 5), dtype=float, with columns corresponding to: key, bias, response, activation function
(act), and aggregation function (agg).
3. connections is an array of shape (C, 4), dtype=float, with columns corresponding to: i_key, o_key, weight, enabled.
Empty nodes or connections are represented using np.nan.
"""
from typing import Tuple, Dict
import jax
import numpy as np
from numpy.typing import NDArray
from jax import numpy as jnp
from jax import jit
from jax import Array
from .utils import fetch_first
def initialize_genomes(pop_size: int,
N: int,
C: int,
num_inputs: int,
num_outputs: int,
default_bias: float = 0.0,
default_response: float = 1.0,
default_act: int = 0,
default_agg: int = 0,
default_weight: float = 0.0) \
-> Tuple[NDArray, NDArray, NDArray, NDArray]:
"""
Initialize genomes with default values.
Args:
pop_size (int): Number of genomes to initialize.
N (int): Maximum number of nodes in the network.
C (int): Maximum number of connections in the network.
num_inputs (int): Number of input nodes.
num_outputs (int): Number of output nodes.
default_bias (float, optional): Default bias value for output nodes. Defaults to 0.0.
default_response (float, optional): Default response value for output nodes. Defaults to 1.0.
default_act (int, optional): Default activation function index for output nodes. Defaults to 1.
default_agg (int, optional): Default aggregation function index for output nodes. Defaults to 0.
default_weight (float, optional): Default weight value for connections. Defaults to 0.0.
Raises:
AssertionError: If the sum of num_inputs, num_outputs, and 1 is greater than N.
Returns:
Tuple[NDArray, NDArray, NDArray, NDArray]: pop_nodes, pop_connections, input_idx, and output_idx arrays.
"""
# Reserve one row for potential mutation adding an extra node
assert num_inputs + num_outputs + 1 <= N, f"Too small N: {N} for input_size: " \
f"{num_inputs} and output_size: {num_outputs}!"
assert num_inputs * num_outputs + 1 <= C, f"Too small C: {C} for input_size: " \
f"{num_inputs} and output_size: {num_outputs}!"
pop_nodes = np.full((pop_size, N, 5), np.nan)
pop_cons = np.full((pop_size, C, 4), np.nan)
input_idx = np.arange(num_inputs)
output_idx = np.arange(num_inputs, num_inputs + num_outputs)
pop_nodes[:, input_idx, 0] = input_idx
pop_nodes[:, output_idx, 0] = output_idx
pop_nodes[:, output_idx, 1] = default_bias
pop_nodes[:, output_idx, 2] = default_response
pop_nodes[:, output_idx, 3] = default_act
pop_nodes[:, output_idx, 4] = default_agg
grid_a, grid_b = np.meshgrid(input_idx, output_idx)
grid_a, grid_b = grid_a.flatten(), grid_b.flatten()
pop_cons[:, :num_inputs * num_outputs, 0] = grid_a
pop_cons[:, :num_inputs * num_outputs, 1] = grid_b
pop_cons[:, :num_inputs * num_outputs, 2] = default_weight
pop_cons[:, :num_inputs * num_outputs, 3] = 1
return pop_nodes, pop_cons, input_idx, output_idx
def expand(pop_nodes: NDArray, pop_cons: NDArray, new_N: int, new_C: int) -> Tuple[NDArray, NDArray]:
"""
Expand the genome to accommodate more nodes.
:param pop_nodes: (pop_size, N, 5)
:param pop_cons: (pop_size, C, 4)
:param new_N:
:param new_C:
:return:
"""
pop_size, old_N, old_C = pop_nodes.shape[0], pop_nodes.shape[1], pop_cons.shape[1]
new_pop_nodes = np.full((pop_size, new_N, 5), np.nan)
new_pop_nodes[:, :old_N, :] = pop_nodes
new_pop_cons = np.full((pop_size, new_C, 4), np.nan)
new_pop_cons[:, :old_C, :] = pop_cons
return new_pop_nodes, new_pop_cons
def expand_single(nodes: NDArray, cons: NDArray, new_N: int, new_C: int) -> Tuple[NDArray, NDArray]:
"""
Expand a single genome to accommodate more nodes.
:param nodes: (N, 5)
:param cons: (2, N, N)
:param new_N:
:param new_C:
:return:
"""
old_N, old_C = nodes.shape[0], cons.shape[0]
new_nodes = np.full((new_N, 5), np.nan)
new_nodes[:old_N, :] = nodes
new_cons = np.full((new_C, 4), np.nan)
new_cons[:old_C, :] = cons
return new_nodes, new_cons
@jit
def count(nodes, cons):
node_cnt = jnp.sum(~jnp.isnan(nodes[:, 0]))
cons_cnt = jnp.sum(~jnp.isnan(cons[:, 0]))
return node_cnt, cons_cnt
@jit
def add_node(nodes: Array, cons: Array, new_key: int,
bias: float = 0.0, response: float = 1.0, act: int = 0, agg: int = 0) -> Tuple[Array, Array]:
"""
add a new node to the genome.
"""
exist_keys = nodes[:, 0]
idx = fetch_first(jnp.isnan(exist_keys))
nodes = nodes.at[idx].set(jnp.array([new_key, bias, response, act, agg]))
return nodes, cons
@jit
def delete_node(nodes: Array, cons: Array, node_key: int) -> Tuple[Array, Array]:
"""
delete a node from the genome. only delete the node, regardless of connections.
"""
node_keys = nodes[:, 0]
idx = fetch_first(node_keys == node_key)
return delete_node_by_idx(nodes, cons, idx)
@jit
def delete_node_by_idx(nodes: Array, cons: Array, idx: int) -> Tuple[Array, Array]:
"""
use idx to delete a node from the genome. only delete the node, regardless of connections.
"""
nodes = nodes.at[idx].set(np.nan)
return nodes, cons
@jit
def add_connection(nodes: Array, cons: Array, i_key: int, o_key: int,
weight: float = 1.0, enabled: bool = True) -> Tuple[Array, Array]:
"""
add a new connection to the genome.
"""
con_keys = cons[:, 0]
idx = fetch_first(jnp.isnan(con_keys))
return add_connection_by_idx(nodes, cons, idx, i_key, o_key, weight, enabled)
@jit
def add_connection_by_idx(nodes: Array, cons: Array, idx: int, i_key: int, o_key: int,
weight: float = 0.0, enabled: bool = True) -> Tuple[Array, Array]:
"""
use idx to add a new connection to the genome.
"""
cons = cons.at[idx].set(jnp.array([i_key, o_key, weight, enabled]))
return nodes, cons
@jit
def delete_connection(nodes: Array, cons: Array, i_key: int, o_key: int) -> Tuple[Array, Array]:
"""
delete a connection from the genome.
"""
idx = fetch_first((cons[:, 0] == i_key) & (cons[:, 1] == o_key))
return delete_connection_by_idx(nodes, cons, idx)
@jit
def delete_connection_by_idx(nodes: Array, cons: Array, idx: int) -> Tuple[Array, Array]:
"""
use idx to delete a connection from the genome.
"""
cons = cons.at[idx].set(np.nan)
return nodes, cons

179
algorithms/neat/genome/graph.py
View File

@@ -1,179 +0,0 @@
"""
Some graph algorithms implemented in jax.
Only used in feed-forward networks.
"""
import jax
from jax import jit, vmap, Array
from jax import numpy as jnp
# from .utils import fetch_first, I_INT
from algorithms.neat.genome.utils import fetch_first, I_INT
@jit
def topological_sort(nodes: Array, connections: Array) -> Array:
"""
a jit-able version of topological_sort! that's crazy!
:param nodes: nodes array
:param connections: connections array
:return: topological sorted sequence
Example:
nodes = jnp.array([
[0],
[1],
[2],
[3]
])
connections = jnp.array([
[
[0, 0, 1, 0],
[0, 0, 1, 1],
[0, 0, 0, 1],
[0, 0, 0, 0]
],
[
[0, 0, 1, 0],
[0, 0, 1, 1],
[0, 0, 0, 1],
[0, 0, 0, 0]
]
])
topological_sort(nodes, connections) -> [0, 1, 2, 3]
"""
connections_enable = connections[1, :, :] == 1
in_degree = jnp.where(jnp.isnan(nodes[:, 0]), jnp.nan, jnp.sum(connections_enable, axis=0))
res = jnp.full(in_degree.shape, I_INT)
idx = 0
def scan_body(carry, _):
res_, idx_, in_degree_ = carry
i = fetch_first(in_degree_ == 0.)
def hit():
# add to res and flag it is already in it
new_res = res_.at[idx_].set(i)
new_idx = idx_ + 1
new_in_degree = in_degree_.at[i].set(-1)
# decrease in_degree of all its children
children = connections_enable[i, :]
new_in_degree = jnp.where(children, new_in_degree - 1, new_in_degree)
return new_res, new_idx, new_in_degree
def miss():
return res_, idx_, in_degree_
return jax.lax.cond(i == I_INT, miss, hit), None
scan_res, _ = jax.lax.scan(scan_body, (res, idx, in_degree), None, length=in_degree.shape[0])
res, _, _ = scan_res
return res
@jit
@vmap
def batch_topological_sort(pop_nodes: Array, pop_connections: Array) -> Array:
"""
batch version of topological_sort
:param pop_nodes:
:param pop_connections:
:return:
"""
return topological_sort(pop_nodes, pop_connections)
@jit
def check_cycles(nodes: Array, connections: Array, from_idx: Array, to_idx: Array) -> Array:
"""
Check whether a new connection (from_idx -> to_idx) will cause a cycle.
:param nodes: JAX array
The array of nodes.
:param connections: JAX array
The array of connections.
:param from_idx: int
The index of the starting node.
:param to_idx: int
The index of the ending node.
:return: JAX array
An array indicating if there is a cycle caused by the new connection.
Example:
nodes = jnp.array([
[0],
[1],
[2],
[3]
])
connections = jnp.array([
[
[0, 0, 1, 0],
[0, 0, 1, 1],
[0, 0, 0, 1],
[0, 0, 0, 0]
],
[
[0, 0, 1, 0],
[0, 0, 1, 1],
[0, 0, 0, 1],
[0, 0, 0, 0]
]
])
check_cycles(nodes, connections, 3, 2) -> True
check_cycles(nodes, connections, 2, 3) -> False
check_cycles(nodes, connections, 0, 3) -> False
check_cycles(nodes, connections, 1, 0) -> False
"""
connections_enable = ~jnp.isnan(connections[0, :, :])
connections_enable = connections_enable.at[from_idx, to_idx].set(True)
nodes_visited = jnp.full(nodes.shape[0], False)
nodes_visited = nodes_visited.at[to_idx].set(True)
def scan_body(visited, _):
new_visited = jnp.dot(visited, connections_enable)
new_visited = jnp.logical_or(visited, new_visited)
return new_visited, None
nodes_visited, _ = jax.lax.scan(scan_body, nodes_visited, None, length=nodes_visited.shape[0])
return nodes_visited[from_idx]
if __name__ == '__main__':
nodes = jnp.array([
[0],
[1],
[2],
[3],
[jnp.nan]
])
connections = jnp.array([
[
[jnp.nan, jnp.nan, 1, jnp.nan, jnp.nan],
[jnp.nan, jnp.nan, 1, 1, jnp.nan],
[jnp.nan, jnp.nan, jnp.nan, 1, jnp.nan],
[jnp.nan, jnp.nan, jnp.nan, jnp.nan, jnp.nan],
[jnp.nan, jnp.nan, jnp.nan, jnp.nan, jnp.nan]
],
[
[jnp.nan, jnp.nan, 1, jnp.nan, jnp.nan],
[jnp.nan, jnp.nan, 1, 1, jnp.nan],
[jnp.nan, jnp.nan, jnp.nan, 1, jnp.nan],
[jnp.nan, jnp.nan, jnp.nan, jnp.nan, jnp.nan],
[jnp.nan, jnp.nan, jnp.nan, jnp.nan, jnp.nan]
]
]
)
print(topological_sort(nodes, connections))
print(check_cycles(nodes, connections, 3, 2))
print(check_cycles(nodes, connections, 2, 3))
print(check_cycles(nodes, connections, 0, 3))
print(check_cycles(nodes, connections, 1, 0))

457
algorithms/neat/genome/mutate.py
View File

@@ -1,457 +0,0 @@
from typing import Tuple
from functools import partial
import jax
import numpy as np
from jax import numpy as jnp
from jax import jit, vmap, Array
from .utils import fetch_random, fetch_first, I_INT, unflatten_connections
from .genome import add_node, delete_node_by_idx, delete_connection_by_idx, add_connection
from .graph import check_cycles
# TODO: Temporally delete single_structural_mutation, for i need to run it as soon as possible.
@jit
def mutate(rand_key: Array,
nodes: Array,
connections: Array,
new_node_key: int,
input_idx: Array,
output_idx: Array,
bias_mean: float = 0,
bias_std: float = 1,
bias_mutate_strength: float = 0.5,
bias_mutate_rate: float = 0.7,
bias_replace_rate: float = 0.1,
response_mean: float = 1.,
response_std: float = 0.,
response_mutate_strength: float = 0.,
response_mutate_rate: float = 0.,
response_replace_rate: float = 0.,
weight_mean: float = 0.,
weight_std: float = 1.,
weight_mutate_strength: float = 0.5,
weight_mutate_rate: float = 0.7,
weight_replace_rate: float = 0.1,
act_default: int = 0,
act_list: Array = None,
act_replace_rate: float = 0.1,
agg_default: int = 0,
agg_list: Array = None,
agg_replace_rate: float = 0.1,
enabled_reverse_rate: float = 0.1,
add_node_rate: float = 0.2,
delete_node_rate: float = 0.2,
add_connection_rate: float = 0.4,
delete_connection_rate: float = 0.4,
):
"""
:param output_idx:
:param input_idx:
:param agg_default:
:param act_default:
:param rand_key:
:param nodes: (N, 5)
:param connections: (2, N, N)
:param new_node_key:
:param bias_mean:
:param bias_std:
:param bias_mutate_strength:
:param bias_mutate_rate:
:param bias_replace_rate:
:param response_mean:
:param response_std:
:param response_mutate_strength:
:param response_mutate_rate:
:param response_replace_rate:
:param weight_mean:
:param weight_std:
:param weight_mutate_strength:
:param weight_mutate_rate:
:param weight_replace_rate:
:param act_list:
:param act_replace_rate:
:param agg_list:
:param agg_replace_rate:
:param enabled_reverse_rate:
:param add_node_rate:
:param delete_node_rate:
:param add_connection_rate:
:param delete_connection_rate:
:return:
"""
def m_add_node(rk, n, c):
return mutate_add_node(rk, n, c, new_node_key, bias_mean, response_mean, act_default, agg_default)
def m_add_connection(rk, n, c):
return mutate_add_connection(rk, n, c, input_idx, output_idx)
def m_delete_node(rk, n, c):
return mutate_delete_node(rk, n, c, input_idx, output_idx)
def m_delete_connection(rk, n, c):
return mutate_delete_connection(rk, n, c)
r1, r2, r3, r4, rand_key = jax.random.split(rand_key, 5)
# mutate add node
aux_nodes, aux_connections = m_add_node(r1, nodes, connections)
nodes = jnp.where(rand(r1) < add_node_rate, aux_nodes, nodes)
connections = jnp.where(rand(r1) < add_node_rate, aux_connections, connections)
# mutate add connection
aux_nodes, aux_connections = m_add_connection(r3, nodes, connections)
nodes = jnp.where(rand(r3) < add_connection_rate, aux_nodes, nodes)
connections = jnp.where(rand(r3) < add_connection_rate, aux_connections, connections)
# mutate delete node
aux_nodes, aux_connections = m_delete_node(r2, nodes, connections)
nodes = jnp.where(rand(r2) < delete_node_rate, aux_nodes, nodes)
connections = jnp.where(rand(r2) < delete_node_rate, aux_connections, connections)
# mutate delete connection
aux_nodes, aux_connections = m_delete_connection(r4, nodes, connections)
nodes = jnp.where(rand(r4) < delete_connection_rate, aux_nodes, nodes)
connections = jnp.where(rand(r4) < delete_connection_rate, aux_connections, connections)
nodes, connections = mutate_values(rand_key, nodes, connections, bias_mean, bias_std, bias_mutate_strength,
bias_mutate_rate, bias_replace_rate, response_mean, response_std,
response_mutate_strength, response_mutate_rate, response_replace_rate,
weight_mean, weight_std, weight_mutate_strength,
weight_mutate_rate, weight_replace_rate, act_list, act_replace_rate, agg_list,
agg_replace_rate, enabled_reverse_rate)
return nodes, connections
@jit
def mutate_values(rand_key: Array,
nodes: Array,
cons: Array,
bias_mean: float = 0,
bias_std: float = 1,
bias_mutate_strength: float = 0.5,
bias_mutate_rate: float = 0.7,
bias_replace_rate: float = 0.1,
response_mean: float = 1.,
response_std: float = 0.,
response_mutate_strength: float = 0.,
response_mutate_rate: float = 0.,
response_replace_rate: float = 0.,
weight_mean: float = 0.,
weight_std: float = 1.,
weight_mutate_strength: float = 0.5,
weight_mutate_rate: float = 0.7,
weight_replace_rate: float = 0.1,
act_list: Array = None,
act_replace_rate: float = 0.1,
agg_list: Array = None,
agg_replace_rate: float = 0.1,
enabled_reverse_rate: float = 0.1) -> Tuple[Array, Array]:
"""
Mutate values of nodes and connections.
Args:
rand_key: A random key for generating random values.
nodes: A 2D array representing nodes.
cons: A 3D array representing connections.
bias_mean: Mean of the bias values.
bias_std: Standard deviation of the bias values.
bias_mutate_strength: Strength of the bias mutation.
bias_mutate_rate: Rate of the bias mutation.
bias_replace_rate: Rate of the bias replacement.
response_mean: Mean of the response values.
response_std: Standard deviation of the response values.
response_mutate_strength: Strength of the response mutation.
response_mutate_rate: Rate of the response mutation.
response_replace_rate: Rate of the response replacement.
weight_mean: Mean of the weight values.
weight_std: Standard deviation of the weight values.
weight_mutate_strength: Strength of the weight mutation.
weight_mutate_rate: Rate of the weight mutation.
weight_replace_rate: Rate of the weight replacement.
act_list: List of the activation function values.
act_replace_rate: Rate of the activation function replacement.
agg_list: List of the aggregation function values.
agg_replace_rate: Rate of the aggregation function replacement.
enabled_reverse_rate: Rate of reversing enabled state of connections.
Returns:
A tuple containing mutated nodes and connections.
"""
k1, k2, k3, k4, k5, rand_key = jax.random.split(rand_key, num=6)
bias_new = mutate_float_values(k1, nodes[:, 1], bias_mean, bias_std,
bias_mutate_strength, bias_mutate_rate, bias_replace_rate)
response_new = mutate_float_values(k2, nodes[:, 2], response_mean, response_std,
response_mutate_strength, response_mutate_rate, response_replace_rate)
weight_new = mutate_float_values(k3, cons[:, 2], weight_mean, weight_std,
weight_mutate_strength, weight_mutate_rate, weight_replace_rate)
act_new = mutate_int_values(k4, nodes[:, 3], act_list, act_replace_rate)
agg_new = mutate_int_values(k5, nodes[:, 4], agg_list, agg_replace_rate)
# mutate enabled
r = jax.random.uniform(rand_key, cons[:, 3].shape)
enabled_new = jnp.where(r < enabled_reverse_rate, 1 - cons[:, 3], cons[:, 3])
enabled_new = jnp.where(~jnp.isnan(cons[:, 3]), enabled_new, jnp.nan)
nodes = nodes.at[:, 1].set(bias_new)
nodes = nodes.at[:, 2].set(response_new)
nodes = nodes.at[:, 3].set(act_new)
nodes = nodes.at[:, 4].set(agg_new)
cons = cons.at[:, 2].set(weight_new)
cons = cons.at[:, 3].set(enabled_new)
return nodes, cons
@jit
def mutate_float_values(rand_key: Array, old_vals: Array, mean: float, std: float,
mutate_strength: float, mutate_rate: float, replace_rate: float) -> Array:
"""
Mutate float values of a given array.
Args:
rand_key: A random key for generating random values.
old_vals: A 1D array of float values to be mutated.
mean: Mean of the values.
std: Standard deviation of the values.
mutate_strength: Strength of the mutation.
mutate_rate: Rate of the mutation.
replace_rate: Rate of the replacement.
Returns:
A mutated 1D array of float values.
"""
k1, k2, k3, rand_key = jax.random.split(rand_key, num=4)
noise = jax.random.normal(k1, old_vals.shape) * mutate_strength
replace = jax.random.normal(k2, old_vals.shape) * std + mean
r = jax.random.uniform(k3, old_vals.shape)
new_vals = old_vals
new_vals = jnp.where(r < mutate_rate, new_vals + noise, new_vals)
new_vals = jnp.where(
jnp.logical_and(mutate_rate < r, r < mutate_rate + replace_rate),
replace,
new_vals
)
new_vals = jnp.where(~jnp.isnan(old_vals), new_vals, jnp.nan)
return new_vals
@jit
def mutate_int_values(rand_key: Array, old_vals: Array, val_list: Array, replace_rate: float) -> Array:
"""
Mutate integer values (act, agg) of a given array.
Args:
rand_key: A random key for generating random values.
old_vals: A 1D array of integer values to be mutated.
val_list: List of the integer values.
replace_rate: Rate of the replacement.
Returns:
A mutated 1D array of integer values.
"""
k1, k2, rand_key = jax.random.split(rand_key, num=3)
replace_val = jax.random.choice(k1, val_list, old_vals.shape)
r = jax.random.uniform(k2, old_vals.shape)
new_vals = old_vals
new_vals = jnp.where(r < replace_rate, replace_val, new_vals)
new_vals = jnp.where(~jnp.isnan(old_vals), new_vals, jnp.nan)
return new_vals
@jit
def mutate_add_node(rand_key: Array, nodes: Array, cons: Array, new_node_key: int,
default_bias: float = 0, default_response: float = 1,
default_act: int = 0, default_agg: int = 0) -> Tuple[Array, Array]:
"""
Randomly add a new node from splitting a connection.
:param rand_key:
:param new_node_key:
:param nodes:
:param cons:
:param default_bias:
:param default_response:
:param default_act:
:param default_agg:
:return:
"""
# randomly choose a connection
i_key, o_key, idx = choice_connection_key(rand_key, nodes, cons)
def nothing(): # there is no connection to split
return nodes, cons
def successful_add_node():
# disable the connection
new_nodes, new_cons = nodes, cons
new_cons = new_cons.at[idx, 3].set(False)
# add a new node
new_nodes, new_cons = \
add_node(new_nodes, new_cons, new_node_key,
bias=default_bias, response=default_response, act=default_act, agg=default_agg)
# add two new connections
w = new_cons[idx, 2]
new_nodes, new_cons = add_connection(new_nodes, new_cons, i_key, new_node_key, weight=1, enabled=True)
new_nodes, new_cons = add_connection(new_nodes, new_cons, new_node_key, o_key, weight=w, enabled=True)
return new_nodes, new_cons
# if from_idx == I_INT, that means no connection exist, do nothing
nodes, cons = jax.lax.cond(idx == I_INT, nothing, successful_add_node)
return nodes, cons
# TODO: Need we really need to delete a node?
@jit
def mutate_delete_node(rand_key: Array, nodes: Array, cons: Array,
input_keys: Array, output_keys: Array) -> Tuple[Array, Array]:
"""
Randomly delete a node. Input and output nodes are not allowed to be deleted.
:param rand_key:
:param nodes:
:param cons:
:param input_keys:
:param output_keys:
:return:
"""
# randomly choose a node
node_key, node_idx = choice_node_key(rand_key, nodes, input_keys, output_keys,
allow_input_keys=False, allow_output_keys=False)
def nothing():
return nodes, cons
def successful_delete_node():
# delete the node
aux_nodes, aux_cons = delete_node_by_idx(nodes, cons, node_idx)
# delete all connections
aux_cons = jnp.where(((aux_cons[:, 0] == node_key) | (aux_cons[:, 1] == node_key))[:, jnp.newaxis],
jnp.nan, aux_cons)
return aux_nodes, aux_cons
nodes, cons = jax.lax.cond(node_idx == I_INT, nothing, successful_delete_node)
return nodes, cons
@jit
def mutate_add_connection(rand_key: Array, nodes: Array, cons: Array,
input_keys: Array, output_keys: Array) -> Tuple[Array, Array]:
"""
Randomly add a new connection. The output node is not allowed to be an input node. If in feedforward networks,
cycles are not allowed.
:param rand_key:
:param nodes:
:param cons:
:param input_keys:
:param output_keys:
:return:
"""
# randomly choose two nodes
k1, k2 = jax.random.split(rand_key, num=2)
i_key, from_idx = choice_node_key(k1, nodes, input_keys, output_keys,
allow_input_keys=True, allow_output_keys=True)
o_key, to_idx = choice_node_key(k2, nodes, input_keys, output_keys,
allow_input_keys=False, allow_output_keys=True)
con_idx = fetch_first((cons[:, 0] == i_key) & (cons[:, 1] == o_key))
def successful():
new_nodes, new_cons = add_connection(nodes, cons, i_key, o_key, weight=1, enabled=True)
return new_nodes, new_cons
def already_exist():
new_cons = cons.at[con_idx, 3].set(True)
return nodes, new_cons
def cycle():
return nodes, cons
is_already_exist = con_idx != I_INT
unflattened = unflatten_connections(nodes, cons)
is_cycle = check_cycles(nodes, unflattened, from_idx, to_idx)
choice = jnp.where(is_already_exist, 0, jnp.where(is_cycle, 1, 2))
nodes, cons = jax.lax.switch(choice, [already_exist, cycle, successful])
return nodes, cons
@jit
def mutate_delete_connection(rand_key: Array, nodes: Array, cons: Array):
"""
Randomly delete a connection.
:param rand_key:
:param nodes:
:param cons:
:return:
"""
# randomly choose a connection
i_key, o_key, idx = choice_connection_key(rand_key, nodes, cons)
def nothing():
return nodes, cons
def successfully_delete_connection():
return delete_connection_by_idx(nodes, cons, idx)
nodes, cons = jax.lax.cond(idx == I_INT, nothing, successfully_delete_connection)
return nodes, cons
@partial(jit, static_argnames=('allow_input_keys', 'allow_output_keys'))
def choice_node_key(rand_key: Array, nodes: Array,
input_keys: Array, output_keys: Array,
allow_input_keys: bool = False, allow_output_keys: bool = False) -> Tuple[Array, Array]:
"""
Randomly choose a node key from the given nodes. It guarantees that the chosen node not be the input or output node.
:param rand_key:
:param nodes:
:param input_keys:
:param output_keys:
:param allow_input_keys:
:param allow_output_keys:
:return: return its key and position(idx)
"""
node_keys = nodes[:, 0]
mask = ~jnp.isnan(node_keys)
if not allow_input_keys:
mask = jnp.logical_and(mask, ~jnp.isin(node_keys, input_keys))
if not allow_output_keys:
mask = jnp.logical_and(mask, ~jnp.isin(node_keys, output_keys))
idx = fetch_random(rand_key, mask)
key = jnp.where(idx != I_INT, nodes[idx, 0], jnp.nan)
return key, idx
@jit
def choice_connection_key(rand_key: Array, nodes: Array, cons: Array) -> Tuple[Array, Array, Array]:
"""
Randomly choose a connection key from the given connections.
:param rand_key:
:param nodes:
:param cons:
:return: i_key, o_key, idx
"""
idx = fetch_random(rand_key, ~jnp.isnan(cons[:, 0]))
i_key = jnp.where(idx != I_INT, cons[idx, 0], jnp.nan)
o_key = jnp.where(idx != I_INT, cons[idx, 1], jnp.nan)
return i_key, o_key, idx
@jit
def rand(rand_key):
return jax.random.uniform(rand_key, ())

106
algorithms/neat/genome/utils.py
View File

@@ -1,106 +0,0 @@
from functools import partial
from typing import Tuple
import jax
from jax import numpy as jnp, Array
from jax import jit, vmap
I_INT = jnp.iinfo(jnp.int32).max # infinite int
EMPTY_NODE = jnp.full((1, 5), jnp.nan)
EMPTY_CON = jnp.full((1, 4), jnp.nan)
@jit
def unflatten_connections(nodes, cons):
"""
transform the (C, 4) connections to (2, N, N)
this function is only used for transform a genome to the forward function, so here we set the weight of un=enabled
connections to nan, that means we dont consider such connection when forward;
:param cons:
:param nodes:
:return:
"""
N = nodes.shape[0]
node_keys = nodes[:, 0]
i_keys, o_keys = cons[:, 0], cons[:, 1]
i_idxs = key_to_indices(i_keys, node_keys)
o_idxs = key_to_indices(o_keys, node_keys)
res = jnp.full((2, N, N), jnp.nan)
# Is interesting that jax use clip when attach data in array
# however, it will do nothing set values in an array
res = res.at[0, i_idxs, o_idxs].set(cons[:, 2])
res = res.at[1, i_idxs, o_idxs].set(cons[:, 3])
# (2, N, N), (2, N, N), (2, N, N)
# res = jnp.where(res[1, :, :] == 0, jnp.nan, res)
return res
@partial(vmap, in_axes=(0, None))
def key_to_indices(key, keys):
return fetch_first(key == keys)
@jit
def fetch_first(mask, default=I_INT) -> Array:
"""
fetch the first True index
:param mask: array of bool
:param default: the default value if no element satisfying the condition
:return: the index of the first element satisfying the condition. if no element satisfying the condition, return I_INT
example:
>>> a = jnp.array([1, 2, 3, 4, 5])
>>> fetch_first(a > 3)
3
>>> fetch_first(a > 30)
I_INT
"""
idx = jnp.argmax(mask)
return jnp.where(mask[idx], idx, default)
@jit
def fetch_last(mask, default=I_INT) -> Array:
"""
similar to fetch_first, but fetch the last True index
"""
reversed_idx = fetch_first(mask[::-1], default)
return jnp.where(reversed_idx == -1, -1, mask.shape[0] - reversed_idx - 1)
@jit
def fetch_random(rand_key, mask, default=I_INT) -> Array:
"""
similar to fetch_first, but fetch a random True index
"""
true_cnt = jnp.sum(mask)
cumsum = jnp.cumsum(mask)
target = jax.random.randint(rand_key, shape=(), minval=1, maxval=true_cnt + 1)
mask = jnp.where(true_cnt == 0, False, cumsum >= target)
return fetch_first(mask, default)
@jit
def argmin_with_mask(arr: Array, mask: Array) -> Array:
masked_arr = jnp.where(mask, arr, jnp.inf)
min_idx = jnp.argmin(masked_arr)
return min_idx
if __name__ == '__main__':
a = jnp.array([1, 2, 3, 4, 5])
print(fetch_first(a > 3))
print(fetch_first(a > 30))
print(fetch_last(a > 3))
print(fetch_last(a > 30))
rand_key = jax.random.PRNGKey(0)
for t in [-1, 0, 1, 2, 3, 4, 5]:
for _ in range(10):
rand_key, _ = jax.random.split(rand_key)
print(jax.random.randint(rand_key, shape=(), minval=1, maxval=2))
print(t, fetch_random(rand_key, a > t))

158
algorithms/neat/pipeline.py
View File

@@ -1,158 +0,0 @@
from typing import List, Union, Tuple, Callable
import time
import jax
import numpy as np
from .species import SpeciesController
from .genome import expand, expand_single
from .function_factory import FunctionFactory
from .population import *
class Pipeline:
"""
Neat algorithm pipeline.
"""
def __init__(self, config, function_factory, seed=42):
self.time_dict = {}
self.function_factory = function_factory
self.randkey = jax.random.PRNGKey(seed)
np.random.seed(seed)
self.config = config
self.N = config.basic.init_maximum_nodes
self.C = config.basic.init_maximum_connections
self.S = config.basic.init_maximum_species
self.expand_coe = config.basic.expands_coe
self.pop_size = config.neat.population.pop_size
self.species_controller = SpeciesController(config)
self.initialize_func = self.function_factory.create_initialize(self.N, self.C)
self.pop_nodes, self.pop_cons, self.input_idx, self.output_idx = self.initialize_func()
self.create_and_speciate = self.function_factory.create_update_speciate(self.N, self.C, self.S)
self.generation = 0
self.generation_time_list = []
self.species_controller.init_speciate(self.pop_nodes, self.pop_cons)
self.best_fitness = float('-inf')
self.best_genome = None
self.generation_timestamp = time.time()
self.evaluate_time = 0
def ask(self):
"""
Create a forward function for the population.
:return:
Algorithm gives the population a forward function, then environment gives back the fitnesses.
"""
return self.function_factory.ask_pop_batch_forward(self.pop_nodes, self.pop_cons)
def tell(self, fitnesses):
self.generation += 1
winner_part, loser_part, elite_mask, pre_spe_center_nodes, pre_spe_center_cons, pre_species_keys, new_species_key_start = self.species_controller.ask(
fitnesses,
self.generation,
self.S, self.N, self.C)
new_node_keys = np.arange(self.generation * self.pop_size, self.generation * self.pop_size + self.pop_size)
self.pop_nodes, self.pop_cons, idx2specie, new_center_nodes, new_center_cons, new_species_keys = self.create_and_speciate(
self.randkey, self.pop_nodes, self.pop_cons, winner_part, loser_part, elite_mask,
new_node_keys,
pre_spe_center_nodes, pre_spe_center_cons, pre_species_keys, new_species_key_start)
self.pop_nodes, self.pop_cons, idx2specie, new_center_nodes, new_center_cons, new_species_keys = \
jax.device_get([self.pop_nodes, self.pop_cons, idx2specie, new_center_nodes, new_center_cons, new_species_keys])
self.species_controller.tell(idx2specie, new_center_nodes, new_center_cons, new_species_keys, self.generation)
self.expand()
def auto_run(self, fitness_func, analysis: Union[Callable, str] = "default"):
for _ in range(self.config.neat.population.generation_limit):
forward_func = self.ask()
tic = time.time()
fitnesses = fitness_func(forward_func)
self.evaluate_time += time.time() - tic
assert np.all(~np.isnan(fitnesses)), "fitnesses should not be nan!"
if analysis is not None:
if analysis == "default":
self.default_analysis(fitnesses)
else:
assert callable(analysis), f"What the fuck you passed in? A {analysis}?"
analysis(fitnesses)
if max(fitnesses) >= self.config.neat.population.fitness_threshold:
print("Fitness limit reached!")
return self.best_genome
self.tell(fitnesses)
print("Generation limit reached!")
return self.best_genome
def expand(self):
"""
Expand the population if needed.
:return:
when the maximum node number of the population >= N
the population will expand
"""
pop_node_keys = self.pop_nodes[:, :, 0]
pop_node_sizes = np.sum(~np.isnan(pop_node_keys), axis=1)
max_node_size = np.max(pop_node_sizes)
if max_node_size >= self.N:
self.N = int(self.N * self.expand_coe)
# self.C = int(self.C * self.expand_coe)
print(f"node expand to {self.N}!")
self.pop_nodes, self.pop_cons = expand(self.pop_nodes, self.pop_cons, self.N, self.C)
# don't forget to expand representation genome in species
for s in self.species_controller.species.values():
s.representative = expand_single(*s.representative, self.N, self.C)
pop_con_keys = self.pop_cons[:, :, 0]
pop_node_sizes = np.sum(~np.isnan(pop_con_keys), axis=1)
max_con_size = np.max(pop_node_sizes)
if max_con_size >= self.C:
# self.N = int(self.N * self.expand_coe)
self.C = int(self.C * self.expand_coe)
print(f"connections expand to {self.C}!")
self.pop_nodes, self.pop_cons = expand(self.pop_nodes, self.pop_cons, self.N, self.C)
# don't forget to expand representation genome in species
for s in self.species_controller.species.values():
s.representative = expand_single(*s.representative, self.N, self.C)
self.create_and_speciate = self.function_factory.create_update_speciate(self.N, self.C, self.S)
def default_analysis(self, fitnesses):
max_f, min_f, mean_f, std_f = max(fitnesses), min(fitnesses), np.mean(fitnesses), np.std(fitnesses)
species_sizes = [len(s.members) for s in self.species_controller.species.values()]
new_timestamp = time.time()
cost_time = new_timestamp - self.generation_timestamp
self.generation_time_list.append(cost_time)
self.generation_timestamp = new_timestamp
max_idx = np.argmax(fitnesses)
if fitnesses[max_idx] > self.best_fitness:
self.best_fitness = fitnesses[max_idx]
self.best_genome = (self.pop_nodes[max_idx], self.pop_cons[max_idx])
print(f"Generation: {self.generation}",
f"fitness: {max_f}, {min_f}, {mean_f}, {std_f}, Species sizes: {species_sizes}, Cost time: {cost_time}")

168
algorithms/neat/population.py
View File

@@ -1,168 +0,0 @@
from functools import partial
import jax
import jax.numpy as jnp
from jax import jit, vmap
from jax import Array
from .genome import distance, mutate, crossover
from .genome.utils import I_INT, fetch_first, argmin_with_mask
@jit
def create_next_generation_then_speciate(rand_key, pop_nodes, pop_cons, winner_part, loser_part, elite_mask,
new_node_keys,
pre_spe_center_nodes, pre_spe_center_cons, species_keys, new_species_key_start,
species_kwargs, mutate_kwargs):
# create next generation
pop_nodes, pop_cons = create_next_generation(rand_key, pop_nodes, pop_cons, winner_part, loser_part, elite_mask,
new_node_keys, **mutate_kwargs)
# speciate
idx2specie, spe_center_nodes, spe_center_cons, species_keys = speciate(pop_nodes, pop_cons, pre_spe_center_nodes,
pre_spe_center_cons, species_keys,
new_species_key_start, **species_kwargs)
return pop_nodes, pop_cons, idx2specie, spe_center_nodes, spe_center_cons, species_keys
@jit
def speciate(pop_nodes: Array, pop_cons: Array, spe_center_nodes: Array, spe_center_cons: Array,
species_keys, new_species_key_start,
disjoint_coe: float = 1., compatibility_coe: float = 0.5, compatibility_threshold=3.0
):
"""
args:
pop_nodes: (pop_size, N, 5)
pop_cons: (pop_size, C, 4)
spe_center_nodes: (species_size, N, 5)
spe_center_cons: (species_size, C, 4)
"""
pop_size, species_size = pop_nodes.shape[0], spe_center_nodes.shape[0]
# prepare distance functions
distance_with_args = partial(distance, disjoint_coe=disjoint_coe, compatibility_coe=compatibility_coe)
o2p_distance_func = vmap(distance_with_args, in_axes=(None, None, 0, 0))
s2p_distance_func = vmap(
o2p_distance_func, in_axes=(0, 0, None, None)
)
# idx to specie key
idx2specie = jnp.full((pop_size,), I_INT, dtype=jnp.int32) # I_INT means not assigned to any species
# part 1: find new centers
# the distance between each species' center and each genome in population
s2p_distance = s2p_distance_func(spe_center_nodes, spe_center_cons, pop_nodes, pop_cons)
def find_new_centers(i, carry):
i2s, scn, scc = carry
# find new center
idx = argmin_with_mask(s2p_distance[i], mask=i2s == I_INT)
# check species[i] exist or not
# if not exist, set idx and i to I_INT, jax will not do array value assignment
idx = jnp.where(species_keys[i] != I_INT, idx, I_INT)
i = jnp.where(species_keys[i] != I_INT, i, I_INT)
i2s = i2s.at[idx].set(species_keys[i])
scn = scn.at[i].set(pop_nodes[idx])
scc = scc.at[i].set(pop_cons[idx])
return i2s, scn, scc
idx2specie, spe_center_nodes, spe_center_cons = jax.lax.fori_loop(0, species_size, find_new_centers, (idx2specie, spe_center_nodes, spe_center_cons))
def continue_execute_while(carry):
i, i2s, scn, scc, sk, ck = carry # sk is short for species_keys, ck is short for current key
not_all_assigned = ~jnp.all(i2s != I_INT)
not_reach_species_upper_bounds = i < species_size
return not_all_assigned & not_reach_species_upper_bounds
def deal_with_each_center_genome(carry):
i, i2s, scn, scc, sk, ck = carry # scn is short for spe_center_nodes, scc is short for spe_center_cons
center_nodes, center_cons = spe_center_nodes[i], spe_center_cons[i]
i2s, scn, scc, sk, ck = jax.lax.cond(
jnp.all(jnp.isnan(center_nodes)), # whether the center genome is valid
create_new_specie, # if not valid, create a new specie
update_exist_specie, # if valid, update the specie
(i, i2s, scn, scc, sk, ck)
)
return i + 1, i2s, scn, scc, sk, ck
def create_new_specie(carry):
i, i2s, scn, scc, sk, ck = carry
# pick the first one who has not been assigned to any species
idx = fetch_first(i2s == I_INT)
# assign it to new specie
sk = sk.at[i].set(ck)
i2s = i2s.at[idx].set(ck)
# update center genomes
scn = scn.at[i].set(pop_nodes[idx])
scc = scc.at[i].set(pop_cons[idx])
i2s, scn, scc = speciate_by_threshold((i, i2s, scn, scc, sk))
return i2s, scn, scc, sk, ck + 1 # change to next new speciate key
def update_exist_specie(carry):
i, i2s, scn, scc, sk, ck = carry
i2s, scn, scc = speciate_by_threshold((i, i2s, scn, scc, sk))
return i2s, scn, scc, sk, ck
def speciate_by_threshold(carry):
i, i2s, scn, scc, sk = carry
# distance between such center genome and ppo genomes
o2p_distance = o2p_distance_func(scn[i], scc[i], pop_nodes, pop_cons)
close_enough_mask = o2p_distance < compatibility_threshold
# when it is close enough, assign it to the species, remember not to update genome has already been assigned
i2s = jnp.where(close_enough_mask & (i2s == I_INT), sk[i], i2s)
return i2s, scn, scc
current_new_key = new_species_key_start
# update idx2specie
_, idx2specie, spe_center_nodes, spe_center_cons, species_keys, new_species_key_start = jax.lax.while_loop(
continue_execute_while,
deal_with_each_center_genome,
(0, idx2specie, spe_center_nodes, spe_center_cons, species_keys, current_new_key)
)
# if there are still some pop genomes not assigned to any species, add them to the last genome
# this condition seems to be only happened when the number of species is reached species upper bounds
idx2specie = jnp.where(idx2specie == I_INT, species_keys[-1], idx2specie)
return idx2specie, spe_center_nodes, spe_center_cons, species_keys
@jit
def create_next_generation(rand_key, pop_nodes, pop_cons, winner_part, loser_part, elite_mask, new_node_keys,
**mutate_kwargs):
# prepare functions
batch_crossover = vmap(crossover)
mutate_with_args = vmap(partial(mutate, **mutate_kwargs))
pop_size = pop_nodes.shape[0]
k1, k2 = jax.random.split(rand_key, 2)
crossover_rand_keys = jax.random.split(k1, pop_size)
mutate_rand_keys = jax.random.split(k2, pop_size)
# batch crossover
wpn = pop_nodes[winner_part] # winner pop nodes
wpc = pop_cons[winner_part] # winner pop connections
lpn = pop_nodes[loser_part] # loser pop nodes
lpc = pop_cons[loser_part] # loser pop connections
npn, npc = batch_crossover(crossover_rand_keys, wpn, wpc, lpn, lpc) # new pop nodes, new pop connections
m_npn, m_npc = mutate_with_args(mutate_rand_keys, npn, npc, new_node_keys) # mutate_new_pop_nodes
# elitism don't mutate
pop_nodes = jnp.where(elite_mask[:, None, None], npn, m_npn)
pop_cons = jnp.where(elite_mask[:, None, None], npc, m_npc)
return pop_nodes, pop_cons

271
algorithms/neat/species.py
View File

@@ -1,271 +0,0 @@
from typing import List, Tuple, Dict, Union, Callable
from itertools import count
import jax
import numpy as np
from numpy.typing import NDArray
from .genome.utils import I_INT
class Species(object):
def __init__(self, key, generation):
self.key = key
self.created = generation
self.last_improved = generation
self.representative: Tuple[NDArray, NDArray] = (None, None) # (center_nodes, center_connections)
self.members: NDArray = None # idx in pop_nodes, pop_connections,
self.fitness = None
self.member_fitnesses = None
self.adjusted_fitness = None
self.fitness_history: List[float] = []
def update(self, representative, members):
self.representative = representative
self.members = members
def get_fitnesses(self, fitnesses):
return fitnesses[self.members]
class SpeciesController:
"""
A class to control the species
"""
def __init__(self, config):
self.config = config
self.species_elitism = self.config.neat.species.species_elitism
self.pop_size = self.config.neat.population.pop_size
self.max_stagnation = self.config.neat.species.max_stagnation
self.min_species_size = self.config.neat.species.min_species_size
self.genome_elitism = self.config.neat.species.genome_elitism
self.survival_threshold = self.config.neat.species.survival_threshold
self.species_idxer = count(0)
self.species: Dict[int, Species] = {} # species_id -> species
def init_speciate(self, pop_nodes: NDArray, pop_connections: NDArray):
"""
speciate for the first generation
:param pop_connections:
:param pop_nodes:
:return:
"""
pop_size = pop_nodes.shape[0]
species_id = next(self.species_idxer)
s = Species(species_id, 0)
members = np.array(list(range(pop_size)))
s.update((pop_nodes[0], pop_connections[0]), members)
self.species[species_id] = s
def __update_species_fitnesses(self, fitnesses):
"""
update the fitness of each species
:param fitnesses:
:return:
"""
for sid, s in self.species.items():
# TODO: here use mean to measure the fitness of a species, but it may be other functions
s.member_fitnesses = s.get_fitnesses(fitnesses)
# s.fitness = np.mean(s.member_fitnesses)
s.fitness = np.max(s.member_fitnesses)
s.fitness_history.append(s.fitness)
s.adjusted_fitness = None
def __stagnation(self, generation):
"""
code modified from neat-python!
:param generation:
:return: whether the species is stagnated
"""
species_data = []
for sid, s in self.species.items():
if s.fitness_history:
prev_fitness = max(s.fitness_history)
else:
prev_fitness = float('-inf')
if prev_fitness is None or s.fitness > prev_fitness:
s.last_improved = generation
species_data.append((sid, s))
# Sort in descending fitness order.
species_data.sort(key=lambda x: x[1].fitness, reverse=True)
result = []
for idx, (sid, s) in enumerate(species_data):
if idx < self.species_elitism: # elitism species never stagnate!
is_stagnant = False
else:
stagnant_time = generation - s.last_improved
is_stagnant = stagnant_time > self.max_stagnation
result.append((sid, s, is_stagnant))
return result
def __reproduce(self, fitnesses: NDArray, generation: int) -> Tuple[NDArray, NDArray, NDArray]:
"""
code modified from neat-python!
:param fitnesses:
:param generation:
:return: crossover_pair for next generation.
# int -> idx in the pop_nodes, pop_connections of elitism
# (int, int) -> the father and mother idx to be crossover
"""
# Filter out stagnated species, collect the set of non-stagnated
# species members, and compute their average adjusted fitness.
# The average adjusted fitness scheme (normalized to the interval
# [0, 1]) allows the use of negative fitness values without
# interfering with the shared fitness scheme.
min_fitness = np.inf
max_fitness = -np.inf
remaining_species = []
for stag_sid, stag_s, stagnant in self.__stagnation(generation):
if not stagnant:
min_fitness = min(min_fitness, np.min(stag_s.member_fitnesses))
max_fitness = max(max_fitness, np.max(stag_s.member_fitnesses))
remaining_species.append(stag_s)
# No species left.
assert remaining_species
# Compute each species' member size in the next generation.
# Do not allow the fitness range to be zero, as we divide by it below.
# TODO: The ``1.0`` below is rather arbitrary, and should be configurable.
fitness_range = max(1.0, max_fitness - min_fitness)
for afs in remaining_species:
# Compute adjusted fitness.
msf = afs.fitness
af = (msf - min_fitness) / fitness_range # make adjusted fitness in [0, 1]
afs.adjusted_fitness = af
adjusted_fitnesses = [s.adjusted_fitness for s in remaining_species]
previous_sizes = [len(s.members) for s in remaining_species]
min_species_size = max(self.min_species_size, self.genome_elitism)
spawn_amounts = compute_spawn(adjusted_fitnesses, previous_sizes, self.pop_size, min_species_size)
assert sum(spawn_amounts) == self.pop_size
# generate new population and speciate
self.species = {}
# int -> idx in the pop_nodes, pop_connections of elitism
# (int, int) -> the father and mother idx to be crossover
part1, part2, elite_mask = [], [], []
for spawn, s in zip(spawn_amounts, remaining_species):
assert spawn >= self.genome_elitism
# retain remain species to next generation
old_members, member_fitnesses = s.members, s.member_fitnesses
s.members = []
self.species[s.key] = s
# add elitism genomes to next generation
sorted_members, sorted_fitnesses = sort_element_with_fitnesses(old_members, member_fitnesses)
if self.genome_elitism > 0:
for m in sorted_members[:self.genome_elitism]:
part1.append(m)
part2.append(m)
elite_mask.append(True)
spawn -= 1
if spawn <= 0:
continue
# add genome to be crossover to next generation
repro_cutoff = int(np.ceil(self.survival_threshold * len(sorted_members)))
repro_cutoff = max(repro_cutoff, 2)
# only use good genomes to crossover
sorted_members = sorted_members[:repro_cutoff]
list_idx1, list_idx2 = np.random.choice(len(sorted_members), size=(2, spawn), replace=True)
part1.extend(sorted_members[list_idx1])
part2.extend(sorted_members[list_idx2])
elite_mask.extend([False] * spawn)
part1_fitness, part2_fitness = fitnesses[part1], fitnesses[part2]
is_part1_win = part1_fitness >= part2_fitness
winner_part = np.where(is_part1_win, part1, part2)
loser_part = np.where(is_part1_win, part2, part1)
return winner_part, loser_part, np.array(elite_mask)
def tell(self, idx2specie, spe_center_nodes, spe_center_cons, species_keys, generation):
for idx, key in enumerate(species_keys):
if key == I_INT:
continue
members = np.where(idx2specie == key)[0]
assert len(members) > 0
if key not in self.species:
s = Species(key, generation)
self.species[key] = s
self.species[key].update((spe_center_nodes[idx], spe_center_cons[idx]), members)
def ask(self, fitnesses, generation, S, N, C):
self.__update_species_fitnesses(fitnesses)
winner_part, loser_part, elite_mask = self.__reproduce(fitnesses, generation)
pre_spe_center_nodes = np.full((S, N, 5), np.nan)
pre_spe_center_cons = np.full((S, C, 4), np.nan)
species_keys = np.full((S,), I_INT)
for idx, (key, specie) in enumerate(self.species.items()):
pre_spe_center_nodes[idx] = specie.representative[0]
pre_spe_center_cons[idx] = specie.representative[1]
species_keys[idx] = key
next_new_specie_key = max(self.species.keys()) + 1
return winner_part, loser_part, elite_mask, pre_spe_center_nodes, \
pre_spe_center_cons, species_keys, next_new_specie_key
def compute_spawn(adjusted_fitness, previous_sizes, pop_size, min_species_size):
"""
Code from neat-python, the only modification is to fix the population size for each generation.
Compute the proper number of offspring per species (proportional to fitness).
"""
af_sum = sum(adjusted_fitness)
spawn_amounts = []
for af, ps in zip(adjusted_fitness, previous_sizes):
if af_sum > 0:
s = max(min_species_size, af / af_sum * pop_size)
else:
s = min_species_size
d = (s - ps) * 0.5
c = int(round(d))
spawn = ps
if abs(c) > 0:
spawn += c
elif d > 0:
spawn += 1
elif d < 0:
spawn -= 1
spawn_amounts.append(spawn)
# Normalize the spawn amounts so that the next generation is roughly
# the population size requested by the user.
total_spawn = sum(spawn_amounts)
norm = pop_size / total_spawn
spawn_amounts = [max(min_species_size, int(round(n * norm))) for n in spawn_amounts]
# for batch parallelization, pop size must be a fixed value.
total_amounts = sum(spawn_amounts)
spawn_amounts[0] += pop_size - total_amounts
assert sum(spawn_amounts) == pop_size, "Population size is not stable."
return spawn_amounts
def sort_element_with_fitnesses(members: NDArray, fitnesses: NDArray) \
-> Tuple[NDArray, NDArray]:
sorted_idx = np.argsort(fitnesses)[::-1]
return members[sorted_idx], fitnesses[sorted_idx]