def run_iterations_chunked(f, bounds, num_iterations=3, chunk_size=5):
X = [np.zeros(len(bounds))]
y = np.array([f(x) for x in X]).flatten()
for jj in range(num_iterations):
sample_X = None
for cc in range(chunk_size):
sample, prob, pred, samples, vals, stds, sample_probs, prob_of_fail, pred_runtimes = bayes.next_sample(
X, y, bounds, current_X=sample_X, improvement=0.1)
if sample_X is None:
sample_X = np.array([sample])
else:
sample_X = np.append(sample_X, np.array([sample]), axis=0)
sample_X = np.append(X, np.array([sample]), axis=0)
X = np.append(X, sample_X, axis=0)
y = np.array([f(x) for x in X]).flatten()
def run_iterations(f, bounds, num_iterations=20):
X = [np.zeros(len(bounds))]
y = np.array([f(x) for x in X]).flatten()
for jj in range(num_iterations):
sample, prob, pred, samples, vals, stds, sample_probs, prob_of_fail, pred_runtimes = bayes.next_sample(
X, y, bounds, improvement=0.1)
print("X: {} prob(I): {} pred: {} value: {}".format(
sample, prob, pred, f(sample)))
X = np.append(X, np.array([sample]), axis=0)
y = np.array([f(x) for x in X]).flatten()
def test_nans():
f = squiggle
X = np.array([np.random.uniform([0.], [5.]) for x in range(200)])
Y = np.array([np.nan] * 200)
sample, prob, pred, samples, vals, stds, sample_probs, prob_of_fail, pred_runtimes = bayes.next_sample(
X, Y, [[-10, 10]])
assert sample[0] < 10. # trying all NaNs
X += np.array([np.random.uniform([0.], [5.]) for x in range(200)])
Y += np.array([np.nan] * 200)
sample, prob, pred, samples, vals, stds, sample_probs, prob_of_fail, pred_runtimes = bayes.next_sample(
X, Y, [[-10, 10]])
assert sample[0] < 10.
def test_squiggle_int():
f = squiggle
X = np.array([np.random.uniform([0.], [5.]) for x in range(200)])
Y = np.array([f(x) for x in X]).flatten()
sample, prob, pred, samples, vals, stds, sample_probs, prob_of_fail, pred_runtimes = bayes.next_sample(
X, Y, [[-10, 10]])
assert sample[0] < 0., "Greater than 0 {}".format(sample[0])
def test_squiggle():
f = squiggle
# we sample a ton of positive examples, ignoring the negative side
X = np.array([np.random.uniform([0.], [5.]) for x in range(200)])
Y = np.array([f(x) for x in X]).flatten()
sample, prob, pred, samples, vals, stds, sample_probs, prob_of_fail, pred_runtimes = bayes.next_sample(
X, Y, [[-5., 5.]], improvement=1.0)
assert sample[0] < 0., "Greater than 0 {}".format(sample[0])
# we sample missing a big chunk between 1 and 3
X = np.append(
np.array([np.random.uniform([0.], [1.]) for x in range(200)]),
np.array([np.random.uniform([0.], [1.]) + 4. for x in range(200)]),
axis=0,
)
Y = np.array([f(x) for x in X]).flatten()
sample, prob, pred, samples, vals, stds, sample_probs, prob_of_fail, pred_runtimes = bayes.next_sample(
X, Y, [[0., 4.]])
assert sample[0] > 1. and sample[
0] < 4., "Sample outside of 1-3 range: {}".format(sample[0])
