def make_model(
config,
peptide_length=9):
"""
If we're using a learned vector embedding for amino acids
then generate a network that expects index inputs,
otherwise assume a 1-of-k binary encoding.
"""
print("===")
print(config)
if config.embedding_size:
return make_embedding_network(
peptide_length=peptide_length,
embedding_input_dim=20,
embedding_output_dim=config.embedding_size,
layer_sizes=[config.hidden_layer_size],
activation=config.activation,
init=config.init,
loss=config.loss,
dropout_probability=config.dropout_probability,
learning_rate=config.learning_rate,
optimizer=config.optimizer)
else:
return make_hotshot_network(
peptide_length=peptide_length,
layer_sizes=[config.hidden_layer_size],
activation=config.activation,
init=config.init,
loss=config.loss,
dropout_probability=config.dropout_probability,
learning_rate=config.learning_rate,
optimizer=config.optimizer)
def test_make_embedding_network_properties():
layer_sizes = [3, 4]
nn = make_embedding_network(
peptide_length=3,
n_amino_acids=3,
layer_sizes=layer_sizes,
loss=mse,
optimizer=RMSprop(lr=0.7, rho=0.9, epsilon=1e-6),
batch_normalization=False,)
eq_(nn.layers[0].input_dim, 3)
eq_(nn.loss, mse)
assert np.allclose(nn.optimizer.lr.eval(), 0.7)
print(nn.layers)
# embedding + flatten + (dense->activation) * hidden layers and last layer
eq_(len(nn.layers), 2 + 2 * (1 + len(layer_sizes)))
def test_make_embedding_network_properties():
layer_sizes = [3, 4]
nn = make_embedding_network(
peptide_length=3,
n_amino_acids=3,
layer_sizes=layer_sizes,
loss=mse,
optimizer=RMSprop(lr=0.7, rho=0.9, epsilon=1e-6),
batch_normalization=False,
)
eq_(nn.layers[0].input_dim, 3)
eq_(nn.loss, mse)
assert np.allclose(nn.optimizer.lr.eval(), 0.7)
print(nn.layers)
# embedding + flatten + (dense->activation) * hidden layers and last layer
eq_(len(nn.layers), 2 + 2 * (1 + len(layer_sizes)))
def test_make_embedding_network_small_dataset():
nn = make_embedding_network(peptide_length=3,
n_amino_acids=3,
layer_sizes=[3],
activation="tanh",
init="lecun_uniform",
loss="mse",
embedding_output_dim=20,
optimizer=RMSprop(lr=0.05,
rho=0.9,
epsilon=1e-6),
batch_normalization=False)
X_negative = np.array([
[0] * 3,
[1] * 3,
[1, 0, 0],
[1, 1, 0],
[0, 1, 0],
[0, 0, 1],
[1, 0, 1],
])
X_positive = np.array([
[0, 2, 0],
[1, 2, 0],
[1, 2, 1],
[0, 2, 1],
[2, 2, 0],
[2, 2, 1],
[2, 2, 2],
])
X_index = np.vstack([X_negative, X_positive])
Y = np.array([0.0] * len(X_negative) + [1.0] * len(X_positive))
nn.fit(X_index, Y, nb_epoch=20)
Y_pred = nn.predict(X_index)
print(Y)
print(Y_pred)
for (Y_i, Y_pred_i) in zip(Y, Y_pred):
assert abs(Y_i - Y_pred_i) <= 0.25, (Y_i, Y_pred_i)
def test_make_embedding_network_small_dataset():
nn = make_embedding_network(
peptide_length=3,
n_amino_acids=3,
layer_sizes=[3],
activation="tanh",
init="lecun_uniform",
loss="mse",
embedding_output_dim=20,
optimizer=RMSprop(lr=0.05, rho=0.9, epsilon=1e-6),
batch_normalization=False)
X_negative = np.array([
[0] * 3,
[1] * 3,
[1, 0, 0],
[1, 1, 0],
[0, 1, 0],
[0, 0, 1],
[1, 0, 1],
])
X_positive = np.array([
[0, 2, 0],
[1, 2, 0],
[1, 2, 1],
[0, 2, 1],
[2, 2, 0],
[2, 2, 1],
[2, 2, 2],
])
X_index = np.vstack([X_negative, X_positive])
Y = np.array([0.0] * len(X_negative) + [1.0] * len(X_positive))
nn.fit(X_index, Y, nb_epoch=20)
Y_pred = nn.predict(X_index)
print(Y)
print(Y_pred)
for (Y_i, Y_pred_i) in zip(Y, Y_pred):
assert abs(Y_i - Y_pred_i) <= 0.25, (Y_i, Y_pred_i)
