def test_data2(self):
gpfa = GPFA(bin_size=self.bin_size, x_dim=8, em_max_iters=self.n_iters)
gpfa.fit(self.data2)
n_trials = len(self.data2)
returned_data = gpfa.valid_data_names
seqs = gpfa.transform(self.data2, returned_data=returned_data)
for key, data in seqs.items():
self.assertEqual(len(data), n_trials,
msg="Failed ndarray field {0}".format(key))
t_start = self.data2[0][0].t_stop
t_stop = self.data2[0][0].t_start
n_bins = int(((t_start - t_stop) / self.bin_size).magnitude)
assert_array_equal(gpfa.transform_info['num_bins'],
[n_bins, ] * n_trials)
def test_data1(self):
gpfa = GPFA(x_dim=self.x_dim, em_max_iters=self.n_iters)
gpfa.fit(self.data1)
xorth = gpfa.transform(self.data1)
self.assertAlmostEqual(gpfa.transform_info['log_likelihood'],
-8172.004695554373, places=5)
# Since data1 is inherently 2 dimensional, only the first two
# dimensions of xorth should have finite power.
for i in [0, 1]:
self.assertNotEqual(xorth[0][i].mean(), 0)
self.assertNotEqual(xorth[0][i].var(), 0)
for i in [2, 3]:
self.assertAlmostEqual(xorth[0][i].mean(), 0, places=2)
self.assertAlmostEqual(xorth[0][i].var(), 0, places=2)
def test_cross_validation(self):
# If GPFA.__init__ is decorated, sklearn signature function parsing
# magic throws the error
# __init__() got an unexpected keyword argument 'args'
lls = []
for x_dim in range(1, self.x_dim + 1):
gpfa = GPFA(x_dim=x_dim, em_max_iters=self.n_iters)
lls.append(np.mean(cross_val_score(gpfa, self.data1, cv=5)))
self.assertTrue(np.argmax(lls) == 1)
def test_invalid_input_data(self):
invalid_data = [(0, [0, 1, 2])]
invalid_bin_size = 10
invalid_tau_init = 100
with self.assertRaises(ValueError):
_ = GPFA(bin_size=invalid_bin_size)
with self.assertRaises(ValueError):
_ = GPFA(tau_init=invalid_tau_init)
gpfa = GPFA()
with self.assertRaises(ValueError):
gpfa.fit(spiketrains=invalid_data)
with self.assertRaises(ValueError):
gpfa.fit(spiketrains=[])
def test_transform_testing_data(self):
# check if the num. of neurons in the test data matches the
# num. of neurons in the training data
gpfa1 = GPFA(x_dim=self.x_dim, em_max_iters=self.n_iters)
gpfa1.fit(self.data1)
with self.assertRaises(ValueError):
gpfa1.transform(self.data2)
def test_returned_data(self):
gpfa = GPFA(bin_size=self.bin_size, x_dim=8, em_max_iters=self.n_iters)
gpfa.fit(self.data2)
seqs = gpfa.transform(self.data2)
self.assertTrue(isinstance(seqs, np.ndarray))
returned_data = gpfa.valid_data_names
seqs = gpfa.transform(self.data2, returned_data=returned_data)
self.assertTrue(len(returned_data) == len(seqs))
self.assertTrue(isinstance(seqs, dict))
with self.assertRaises(ValueError):
seqs = gpfa.transform(self.data2, returned_data=['invalid_name'])
def test_fit_transform(self):
gpfa1 = GPFA(bin_size=self.bin_size, x_dim=self.x_dim,
em_max_iters=self.n_iters)
gpfa1.fit(self.data1)
xorth1 = gpfa1.transform(self.data1)
xorth2 = GPFA(bin_size=self.bin_size, x_dim=self.x_dim,
em_max_iters=self.n_iters).fit_transform(self.data1)
for i in range(len(self.data1)):
for j in range(self.x_dim):
assert_array_almost_equal(xorth1[i][j], xorth2[i][j])
def test_cross_validation(self):
lls = []
for x_dim in range(1, self.x_dim + 1):
gpfa = GPFA(x_dim=x_dim, em_max_iters=self.n_iters)
lls.append(np.mean(cross_val_score(gpfa, self.data1, cv=5)))
self.assertTrue(np.argmax(lls) == 1)
