complete HyperNEAT!

config/__init__.py

@@ -0,0 +1 @@
+from .config import Configer

config/config.py

@@ -0,0 +1,70 @@
+import os
+import warnings
+import configparser
+
+import numpy as np
+
+
+class Configer:
+
+    @classmethod
+    def __load_default_config(cls):
+        par_dir = os.path.dirname(os.path.abspath(__file__))
+        default_config_path = os.path.join(par_dir, "default_config.ini")
+        return cls.__load_config(default_config_path)
+
+    @classmethod
+    def __load_config(cls, config_path):
+        c = configparser.ConfigParser()
+        c.read(config_path)
+        config = {}
+
+        for section in c.sections():
+            for key, value in c.items(section):
+                config[key] = eval(value)
+
+        return config
+
+    @classmethod
+    def __check_redundant_config(cls, default_config, config):
+        for key in config:
+            if key not in default_config:
+                warnings.warn(f"Redundant config: {key} in config!")
+
+    @classmethod
+    def __complete_config(cls, default_config, config):
+        for key in default_config:
+            if key not in config:
+                config[key] = default_config[key]
+
+    @classmethod
+    def load_config(cls, config_path=None):
+        default_config = cls.__load_default_config()
+        if config_path is None:
+            config = {}
+        elif not os.path.exists(config_path):
+            warnings.warn(f"config file {config_path} not exist!")
+            config = {}
+        else:
+            config = cls.__load_config(config_path)
+
+        cls.__check_redundant_config(default_config, config)
+        cls.__complete_config(default_config, config)
+
+        cls.refactor_activation(config)
+        cls.refactor_aggregation(config)
+
+        config['input_idx'] = np.arange(config['num_inputs'])
+        config['output_idx'] = np.arange(config['num_inputs'], config['num_inputs'] + config['num_outputs'])
+
+        return config
+
+    @classmethod
+    def refactor_activation(cls, config):
+        config['activation_default'] = 0
+        config['activation_options'] = np.arange(len(config['activation_option_names']))
+
+    @classmethod
+    def refactor_aggregation(cls, config):
+        config['aggregation_default'] = 0
+        config['aggregation_options'] = np.arange(len(config['aggregation_option_names']))

config/default_config.ini

@@ -0,0 +1,82 @@
+[basic]
+random_seed = 0
+generation_limit = 1000
+
+[problem]
+fitness_threshold = 3.9999
+num_inputs = 2
+num_outputs = 1
+
+[neat]
+network_type = "feedforward"
+activate_times = 5
+maximum_nodes = 50
+maximum_conns = 50
+maximum_species = 10
+
+[hyperneat]
+below_threshold = 0.2
+max_weight = 3
+h_activation = "sigmoid"
+h_aggregation = "sum"
+h_activate_times = 5
+
+[substrate]
+input_coors = [[-1, 1], [0, 1], [1, 1]]
+hidden_coors = [[-1, 0], [0, 0], [1, 0]]
+output_coors = [[0, -1]]
+
+[population]
+pop_size = 10
+
+[genome]
+compatibility_disjoint = 1.0
+compatibility_weight = 0.5
+conn_add_prob = 0.4
+conn_delete_prob = 0
+node_add_prob = 0.2
+node_delete_prob = 0
+
+[species]
+compatibility_threshold = 3.0
+species_elitism = 2
+max_stagnation = 15
+genome_elitism = 2
+survival_threshold = 0.2
+min_species_size = 1
+spawn_number_change_rate = 0.5
+
+[gene]
+# bias
+bias_init_mean = 0.0
+bias_init_std = 1.0
+bias_mutate_power = 0.5
+bias_mutate_rate = 0.7
+bias_replace_rate = 0.1
+
+# response
+response_init_mean = 1.0
+response_init_std = 0.0
+response_mutate_power = 0.0
+response_mutate_rate = 0.0
+response_replace_rate = 0.0
+
+# activation
+activation_default = "sigmoid"
+activation_option_names = ["tanh"]
+activation_replace_rate = 0.0
+
+# aggregation
+aggregation_default = "sum"
+aggregation_option_names = ["sum"]
+aggregation_replace_rate = 0.0
+
+# weight
+weight_init_mean = 0.0
+weight_init_std = 1.0
+weight_mutate_power = 0.5
+weight_mutate_rate = 0.8
+weight_replace_rate = 0.1
+
+[visualize]
+renumber_nodes = True