def test_fit():
npr.seed(1)
N = 10
D = 5
burnin = 100
mcmc_iters = 100
num_pending = 3
num_fantasies = 2
gp = GPClassifier(D,
burnin=burnin,
mcmc_iters=mcmc_iters,
num_fantasies=num_fantasies)
inputs = np.vstack((0.1 * npr.rand(N, D), npr.rand(N, D)))
inputs[12] = np.ones(D)
pending = npr.rand(3, D)
W = npr.randn(D, 1)
vals = (inputs - inputs.mean(0)).dot(W).flatten() > 0
gp.fit(inputs, vals, pending)
probs = np.zeros(inputs.shape[0])
for i in xrange(gp.num_states):
gp.set_state(i)
probs += (gp.latent_values.value > 0) / float(mcmc_iters)
assert np.all(probs[:N] < 0.5) and np.all(probs[N:] > 0.5)
assert gp.values.shape[0] == 2 * N + num_pending
assert gp.values.shape[1] == 2
assert gp.chain_length == burnin + mcmc_iters
assert all(
[np.all(p.value != p.initial_value) for p in gp.params.values()])
assert len(gp._cache_list) == mcmc_iters
assert len(gp._hypers_list) == mcmc_iters
assert len(gp._latent_values_list) == mcmc_iters
assert len(gp._fantasy_values_list) == mcmc_iters
def test_fit():
npr.seed(1)
N = 10
D = 5
burnin = 100
mcmc_iters = 100
num_pending = 3
num_fantasies = 2
gp = GPClassifier(D, burnin=burnin, mcmc_iters=mcmc_iters, num_fantasies=num_fantasies)
inputs = np.vstack((0.1*npr.rand(N,D),npr.rand(N,D)))
inputs[12] = np.ones(D)
pending = npr.rand(3,D)
W = npr.randn(D,1)
vals = (inputs - inputs.mean(0)).dot(W).flatten() > 0
gp.fit(inputs, vals, pending)
probs = np.zeros(inputs.shape[0])
for i in xrange(gp.num_states):
gp.set_state(i)
probs += (gp.latent_values.value > 0) / float(mcmc_iters)
assert np.all(probs[:N] < 0.5) and np.all(probs[N:] > 0.5)
assert gp.values.shape[0] == 2*N + num_pending
assert gp.values.shape[1] == 2
assert gp.chain_length == burnin + mcmc_iters
assert all([np.all(p.value != p.initial_value) for p in gp.params.values()])
assert len(gp._cache_list) == mcmc_iters
assert len(gp._hypers_list) == mcmc_iters
assert len(gp._latent_values_list) == mcmc_iters
assert len(gp._fantasy_values_list) == mcmc_iters
