def test_saved_iter_jVAE(model):
X, _ = gen_image_data()
X = X[:, 0, ...]
vae_model = model((8, 8))
vae_model.fit(X, training_cycles=4, batch_size=2, filename="jvae_metadict")
num_iter = vae_model.kdict_["num_iter"]
loaded_model = load_model("jvae_metadict.tar")
assert_equal(num_iter, loaded_model.kdict_["num_iter"])
def test_saved_optimizer_imspec():
X, X_test = gen_image_data()
y, y_test = gen_spectra()
i2s_model = ImSpec((8, 8), (16, ))
i2s_model.fit(X, y, X_test, y_test, training_cycles=4, batch_size=2)
opt1 = i2s_model.optimizer
loaded_model = load_model("model_metadict_final.tar")
opt2 = loaded_model.optimizer
compare_optimizers(opt1, opt2)
def test_saved_optimizer_VAE(model):
X, _ = gen_image_data()
X = X[:, 0, ...]
vae_model = model((8, 8))
vae_model.fit(X, training_cycles=4, batch_size=2, filename="vae_metadict")
opt1 = vae_model.optim
loaded_model = load_model("vae_metadict.tar")
opt2 = loaded_model.optim
compare_optimizers(opt1, opt2)
def test_io_imspec():
X, X_test = gen_image_data()
y, y_test = gen_spectra()
i2s_model = ImSpec((8, 8), (16, ))
i2s_model.fit(X, y, X_test, y_test, training_cycles=4, batch_size=2)
loaded_model = load_model("model_metadict_final.tar")
for p1, p2 in zip(loaded_model.net.parameters(),
i2s_model.net.parameters()):
assert_array_equal(p1.detach().cpu().numpy(),
p2.detach().cpu().numpy())
def test_io_segmentor(model):
X, X_test = gen_image_data()
y, y_test = gen_image_labels()
segmodel = Segmentor(model, nb_classes=3)
segmodel.fit(X, y, X_test, y_test, training_cycles=4, batch_size=2)
loaded_model = load_model("model_metadict_final.tar")
for p1, p2 in zip(loaded_model.net.parameters(),
segmodel.net.parameters()):
assert_array_equal(p1.detach().cpu().numpy(),
p2.detach().cpu().numpy())
def test_resume_training(model):
X, _ = gen_image_data()
X = X[:, 0, ...]
vae_model = model((8, 8))
vae_model.fit(X, training_cycles=4, batch_size=2, filename="vae_metadict")
loss0 = abs(vae_model.loss_history["train_loss"][0])
loaded_model = load_model("vae_metadict.tar")
loaded_model.fit(X,
training_cycles=4,
batch_size=2,
filename="vae_metadict")
loss1 = abs(loaded_model.loss_history["train_loss"][0])
assert_(not np.isnan(loss1))
assert_(loss1 < loss0)
def test_io_VAE(model):
X, _ = gen_image_data()
X = X[:, 0, ...]
vae_model = model((8, 8))
vae_model.fit(X, training_cycles=4, batch_size=2, filename="vae_metadict")
loaded_model = load_model("vae_metadict.tar")
for p1, p2 in zip(loaded_model.encoder_net.parameters(),
vae_model.encoder_net.parameters()):
assert_array_equal(p1.detach().cpu().numpy(),
p2.detach().cpu().numpy())
for p1, p2 in zip(loaded_model.decoder_net.parameters(),
vae_model.decoder_net.parameters()):
assert_array_equal(p1.detach().cpu().numpy(),
p2.detach().cpu().numpy())
def test_saved_optimizer_segmentor(model):
X, X_test = gen_image_data()
y, y_test = gen_image_labels()
segmodel = Segmentor(model, nb_classes=3)
segmodel.fit(X,
y,
X_test,
y_test,
training_cycles=4,
batch_size=2,
filename="segmodel")
opt1 = segmodel.optimizer
loaded_model = load_model("segmodel_metadict_final.tar")
opt2 = loaded_model.optimizer
compare_optimizers(opt1, opt2)
