def test_recommend():
"""simple check for the moment: see if it runs fine"""
def objective(array, invert=False):
if invert is True:
return -np.sum(array**2)
else:
return np.sum(array**2)
# --------------------
# 2D input, continuous
# --------------------
g = Golem(forest_type='rf',
ntrees=10,
goal='min',
nproc=1,
random_state=42,
verbose=True)
param_space = [{
'type': 'continuous',
'low': -1,
'high': 1
}, {
'type': 'continuous',
'low': -1,
'high': 1
}]
g.set_param_space(param_space)
X_obs = []
y_obs = []
for i in range(5):
X_next = g.recommend(X=X_obs,
y=y_obs,
distributions=[Uniform(0.2),
Uniform(0.2)],
pop_size=10,
ngen=5)
y_next = objective(np.array(X_next))
X_obs.append(X_next)
y_obs.append(y_next)
# variation of the above
g = Golem(forest_type='rf',
ntrees=10,
goal='max',
nproc=1,
random_state=42,
verbose=True)
param_space = [{
'type': 'continuous',
'low': -1,
'high': 1
}, {
'type': 'continuous',
'low': -1,
'high': 1
}]
g.set_param_space(param_space)
X_obs = []
y_obs = []
for i in range(5):
X_next = g.recommend(X=X_obs,
y=y_obs,
distributions=[Normal(0.1),
Normal(0.1)],
pop_size=10,
ngen=5)
y_next = objective(np.array(X_next), invert=True)
X_obs.append(X_next)
y_obs.append(y_next)
# variation of the above
g = Golem(forest_type='rf',
ntrees=10,
goal='min',
nproc=None,
random_state=42,
verbose=True)
param_space = [{
'type': 'continuous',
'low': -1,
'high': 1
}, {
'type': 'continuous',
'low': -1,
'high': 1
}]
g.set_param_space(param_space)
X_obs = []
y_obs = []
for i in range(5):
X_next = g.recommend(X=X_obs,
y=y_obs,
distributions=[Uniform(0.2),
Uniform(0.2)],
pop_size=10,
ngen=5)
y_next = objective(np.array(X_next))
X_obs.append(X_next)
y_obs.append(y_next)
# -------------------------------------------
# 3D input: continuous, discrete, categorical
# -------------------------------------------
def objective(x_list):
a = x_list[0]
b = x_list[1]
c = x_list[2]
return a**2 + b / 2. + 3 * len(c)
cats = ['red', 'blue', 'green', 'pink', 'yellow']
param_space = [{
'type': 'continuous',
'low': 0.,
'high': 1.
}, {
'type': 'discrete',
'low': 1,
'high': 100
}, {
'type': 'categorical',
'categories': cats
}]
# test optimization passing DataFrames, 1 processor
# -------------------------------------------------
g = Golem(forest_type='rf',
ntrees=10,
random_state=42,
verbose=False,
goal='min',
nproc=1)
g.set_param_space(param_space)
dists = [Uniform(0.1), DiscreteLaplace(5), Categorical(cats, 0.4)]
df = pd.DataFrame({'x0': [], 'x1': [], 'x2': [], 'obj': []})
for i in range(5):
X = df.loc[:, ['x0', 'x1', 'x2']]
y = df.loc[:, 'obj']
X_next = g.recommend(X=X,
y=y,
distributions=dists,
pop_size=30,
ngen=10,
verbose=False)
y_next = objective(X_next)
# append to dataframe
df_next = pd.DataFrame({
'x0': [X_next[0]],
'x1': [X_next[1]],
'x2': [X_next[2]],
'obj': [y_next]
})
df = df.append(df_next)
# test optimization passing list of lists, multiple processors
# ------------------------------------------------------------
g = Golem(forest_type='rf',
ntrees=10,
random_state=42,
verbose=False,
goal='min',
nproc=None)
g.set_param_space(param_space)
dists = [Uniform(0.1), DiscreteLaplace(5), Categorical(cats, 0.4)]
X_obs = []
y_obs = []
for i in range(10):
X_next = g.recommend(X=X_obs,
y=y_obs,
distributions=dists,
pop_size=30,
ngen=10,
verbose=False)
y_next = objective(X_next)
X_obs.append(X_next)
y_obs.append(y_next)
