def test_full_hawks_run_multiple(self):
gen = hawks.create_generator("validation.json")
gen.run()
# Run a second time to ensure there's no carryover
gen = hawks.create_generator("validation.json")
gen.run()
res = gen.get_stats()
known_result = pd.read_csv("validation.csv", index_col=False)
# Pandas can be iffy with data types
equals = np.allclose(res.values, known_result.values)
self.assertTrue(equals)
def test_full_hawks_run_multiple(self):
test_fpath = Path(hawks.__file__).parents[1] / "tests"
gen = hawks.create_generator(test_fpath / "validation.json")
gen.run()
# Run a second time to ensure there's no carryover
gen = hawks.create_generator(test_fpath / "validation.json")
gen.run()
res = gen.get_stats()
known_result = pd.read_csv(test_fpath / "validation.csv",
index_col=False)
print(res)
print(known_result)
# Pandas can be iffy with data types
equals = np.allclose(res.values, known_result.values)
self.assertTrue(equals)
def test_nested_config_arg(self):
gen = hawks.create_generator(
{"constraints": {
"overlap": {
"limit": "TEST"
}
}})
self.assertEqual(gen.full_config["constraints"]["overlap"]["limit"],
"TEST")
def test_full_hawks_run(self):
gen = hawks.create_generator("validation.json")
gen.run()
res = gen.get_stats()
known_result = pd.read_csv("validation.csv", index_col=False)
print(res)
print(known_result)
# Pandas can be iffy with data types
equals = np.allclose(res.values, known_result.values)
self.assertTrue(equals)
def test_full_hawks_run(self):
test_fpath = Path(hawks.__file__).parents[1] / "tests"
gen = hawks.create_generator(test_fpath / "validation.json")
gen.run()
res = gen.get_stats()
known_result = pd.read_csv(test_fpath / "validation.csv",
index_col=False)
print("Result:")
print(res)
print("Known result:")
print(known_result)
print("---")
# Pandas can be iffy with data types
equals = np.allclose(res.values, known_result.values)
self.assertTrue(equals)
def test_incorrect_config_arg(self):
with self.assertRaises(ValueError):
gen = hawks.create_generator({
"hawks": {
"seed_num": 4,
"num_runs": 1
},
"objectives": {
"silhouette": {
"target": 0.9
}
},
"constraints": {
"eigenval_ratio": {
"lim": "upper" # <--- error
}
}
})
def test_multiconfig_deep(self):
config = {
"dataset": {
"num_examples": [10, 100, 1000]
},
"constraints": {
"overlap": {
"limit": ["upper", "lower"]
}
},
"ga": {
"num_gens": [50, 100, 10, 200],
"mut_args_mean": {
"dims": ["each", "all"]
}
}
}
obj = hawks.create_generator(config)
total_configs, _, _ = obj._count_multiconfigs()
self.assertEqual(total_configs, 48)
def main(ndim, nclusters, run_id):
config = {
"hawks": {
"folder_name": "",
"save_best_data": True
},
"dataset": {
"num_clusters": nclusters,
"num_dims": ndim
},
"ga": {
"num_gens": 500
},
"constraints": {
"overlap": {
"threshold": 0.3,
"limit": "lower"
}
}
}
generator = hawks.create_generator(config)
print(generator.folder_name)
print(generator.save_best_data)
generator.run()
# Get the best dataset found and it's labels
datasets, label_sets = generator.get_best_dataset()
# Stored as a list for multiple runs
data, labels = datasets[0], label_sets[0]
# Run KMeans on the data
km = KMeans(n_clusters=len(np.unique(labels)), random_state=0).fit(data)
# Get the Adjusted Rand Index for KMeans on the data
ari = adjusted_rand_score(labels, km.labels_)
sil = silhouette_score(data, km.labels_)
print(f"ARI: {ari}, SIL: {sil}")
generator.plot_best_indivs(show=True)
"""
import seaborn as sns
import hawks
# Create the generator
generator = hawks.create_generator({
"hawks": {
"seed_num": 42,
"num_runs": 5
},
"dataset": {
"num_clusters": 5
},
"objectives": {
"silhouette": {
"target": 0.9
}
},
"constraints": {
"overlap": {
"threshold": 0.05,
"limit": "lower"
}
}
})
# Run HAWKS!
generator.run()
# Make a dictionary of options common to both plots
converg_kws = dict(show=True, xlabel="Generation", ci="sd", legend_type=None)
# Make the font etc. larger
sns.set_context("talk")
from pathlib import Path
import hawks
gen = hawks.create_generator("validation.json")
gen.run()
hawks.utils.df_to_csv(
gen.stats,
Path.cwd(),
"validation"
)
def setUp(self):
self.gen = hawks.create_generator("validation.json")
self.init_pop = []
for indiv in self.gen.create_individual():
self.init_pop.append(indiv)
from sklearn.datasets import make_blobs, make_moons
import seaborn as sns
import hawks
SEED_NUM = 10
NUM_RUNS = 10 # May take a few minutes
NUM_CLUSTERS = 5
generator = hawks.create_generator({
"hawks": {
"seed_num": SEED_NUM,
"num_runs": int(NUM_RUNS / 2) # for parity
},
"dataset": {
"num_clusters": NUM_CLUSTERS
},
"objectives": {
"silhouette": {
"target": [0.5, 0.9]
}
}
})
generator.run()
# Analyse the hawks datasets
df, _ = hawks.analysis.analyse_datasets(generator=generator,
source="HAWKS",
seed=SEED_NUM,
save=False)
# Make the blobs datasets
datasets = []
label_sets = []
import numpy as np
from sklearn.cluster import KMeans
from sklearn.metrics import adjusted_rand_score
import hawks
SEED_NUM = 42
# Fix the seed number
config = {
"hawks": {
"seed_num": SEED_NUM
}
}
# Any missing parameters will take the default seen in configs/defaults.json
generator = hawks.create_generator(config)
# Run the generator
generator.run()
# Get the best dataset found and it's labels
datasets, label_sets = generator.get_best_dataset()
# Stored as a list for multiple runs
data, labels = datasets[0], label_sets[0]
# Run KMeans on the data
km = KMeans(
n_clusters=len(np.unique(labels)), random_state=SEED_NUM
).fit(data)
# Get the Adjusted Rand Index for KMeans on the data
ari = adjusted_rand_score(labels, km.labels_)
print(f"ARI: {ari}")
def setUp(self):
self.gen = hawks.create_generator(
Path(hawks.__file__).parents[1] / "tests" / "validation.json")
self.init_pop = []
for indiv in self.gen.create_individual():
self.init_pop.append(indiv)
# Use a HAWKS and sklearn example, and show the performance (boxplot) for HAWKS, moons, and blobs
from pathlib import Path
from sklearn.datasets import make_blobs, make_moons
import hawks
SEED_NUM = 42
SAVE_FOLDER = Path.cwd()
NUM_RUNS = 5
NUM_CLUSTERS = 5
generator = hawks.create_generator({
"hawks": {
"seed_num": SEED_NUM,
"num_runs": NUM_RUNS
},
"dataset": {
"num_clusters": NUM_CLUSTERS
}
})
generator.run()
# Analyse the hawks datasets
df, _ = hawks.analysis.analyse_datasets(generator=generator,
source="HAWKS",
seed=SEED_NUM,
save=False)
# Make the blobs datasets
datasets = []
label_sets = []
for run in range(NUM_RUNS):
data, labels = make_blobs(n_samples=1000,
n_features=2,
