def test_train(self):
"""
"""
print("\nTest 0: VAE initialization")
vae = VariationalAutoEncoder(input_dim=10, state_dim=3)
# save_on_valid_improvement=True) # OK!
vae.train(dataset, num_epochs=20)
def test_train_custom_node2(self):
"""
Test commit
"""
print("Test 1: with CustomNode\n")
builder = StaticBuilder("MyModel")
enc_dirs = {
'loc_numlayers': 2,
'loc_numnodes': 64,
'loc_activations': 'softplus',
'loc_netgrowrate': 1.0
}
input_dim = 10
in0 = builder.addInput(input_dim, name='Observation')
# Define Custom Recognition Model
cust_rec = builder.createCustomNode(inputs=[in0],
num_outputs=1,
name="Recognition")
cust_in2 = cust_rec.addTransformInner(3,
main_inputs=[0],
node_class=NormalTriLNode,
**enc_dirs)
cust_rec.declareOslot(oslot=0,
innernode_name=cust_in2,
inode_oslot_name='main')
# Define Custom Generative Model
cust_gen = builder.createCustomNode(inputs=cust_rec.name,
num_outputs=1,
name="Generative")
cust_ginn1 = cust_gen.addTransformInner(16, main_inputs=[0])
cust_ginn2 = cust_gen.addTransformInner(10,
main_inputs=cust_ginn1,
node_class=NormalTriLNode,
**enc_dirs)
cust_gen.declareOslot(oslot=0,
innernode_name=cust_ginn2,
inode_oslot_name='main')
# Define VAE and train
vae = VariationalAutoEncoder(builder=builder)
vae.train(dataset, num_epochs=10)
def test_train1(self):
"""
"""
print("Test 0:")
dirs = {}
builder = StaticBuilder(scope='vae')
odim = 10
idim = 3
hdim = 16
i1 = builder.addInput(odim, name='Observation', **dirs)
enc0 = builder.addTransformInner(hdim, main_inputs=i1, name='Inner')
enc1 = builder.addTransformInner(idim,
main_inputs=enc0,
name='Recognition',
node_class=NormalTriLNode)
builder.addTransformInner(odim,
main_inputs=enc1,
name='Generative',
node_class=NormalTriLNode)
vae = VariationalAutoEncoder(builder=builder)
# save_on_valid_improvement=True) # OK!
vae.train(dataset, num_epochs=20)
def test_train(self):
"""
"""
nsamps = 100
idim = 3
odim = 10
x = 1.0*np.random.randn(nsamps, idim)
W = np.random.randn(3, odim)
y = np.tanh(np.dot(x, W) + 0.1*np.random.randn(nsamps, odim)) # + 3*x[:,1:]**2 + 0.5*np.random.randn(100,1)
dataset = {'train_response' : y}
vae = VariationalAutoEncoder(latent_dim=3, output_dim=10)
vae.build()
vae.train(dataset, num_epochs=200)
def test_build(self):
"""
"""
print("\nTest 1: VAE build")
dc = VariationalAutoEncoder(latent_dim=3, output_dim=10, batch_size=1)
dc.build()
def test_init(self):
"""
"""
print("\nTest 0: VAE initialization")
VariationalAutoEncoder(latent_dim=3, output_dim=10, batch_size=1)
def test_build(self):
"""
"""
dc = VariationalAutoEncoder(latent_dim=3, output_dim=10, batch_size=1)
dc.build()
def test_init(self):
"""
"""
VariationalAutoEncoder(latent_dim=3, output_dim=10, batch_size=1)
def test_build(self):
"""
"""
print("\nTest 1: VAE build")
VariationalAutoEncoder(input_dim=3, state_dim=10, batch_size=1)