def test_flatten_layer_with_model():
model = Sequential()
model.add(Embedding(2, 2))
model.add(Flatten())
model.add(Linear(4, 1))
inputs = np.array([0, 1])
out = model.feed_forward(inputs)
assert out.shape == (1, )
def test_model_backprop_with_rnn():
inputs = np.array([[[1, 1], [2, 2], [3, 3], [1, 1]]])
targets = np.array([[[2, 2], [3, 3], [4, 4], [2, 2]]])
model = Sequential()
model.add(RNN(2, input_shape=(4, 2)))
model.skeleton()
model.compile(batch_size=1, optimizer=SGD(momentum=0))
his = model.train(inputs, targets, 30, verbose=True)
preds = model.predict(inputs)
preds = preds.astype('int32')
assert preds.shape == (1, 4, 2)
import matplotlib.pyplot as plt
INPUT_SIZE = 3
HIDDEN_SIZE = 50
OUTPUT_SIZE = 1
inputs = np.array([[0,0,1],
[1,1,1],
[1,0,1],
[0,1,1]])
outputs = np.array([0,1,1,0])
# init sequential model
model = Sequential()
# add layers
model.add(Linear(INPUT_SIZE, OUTPUT_SIZE, activation='relu'))
# we can check structure of the model
model.skeleton()
# we need to compile the model
model.compile(
loss='mse',
batch_size=2,
optimizer=SGD(),
lr=0.001,
)
# to train we just use train method
return [x >> i & 1 for i in range(10)]
if __name__ == "__main__":
inputs = np.array([
binary_encode(x)
for x in range(101, 1024)
])
targets = np.array([
fizz_buzz_encode(x)
for x in range(101, 1024)
])
print(targets)
model = Sequential()
model.add(Linear(10 ,50))
model.add(Tanh())
model.add(Linear(50,4))
model.skeleton()
model.compile(lr=0.001)
model.train(inputs, targets, 2000)
#model.save_weights("test")
#model.feed_forward([0,1,0,0,0,1,1,0,0,0])
# [0100001001] -> [1000]
# model.compile(batch_size=32, lr=0.001)
# model.train(inputs, targets, 2000)
def test_predict_tensor_expand_dim():
model = Sequential()
model.add(Linear(2, 1))
preds = model.predict(np.array([1, 2]))
assert preds.shape == (1, 1)
def test_predict_function_output_shape():
model = Sequential()
model.add(Linear(5, 3))
preds = model.predict(np.array([[1, 2, 4, 6, 7]]))
assert preds.shape == (1, 3)
def test_skeleton_function():
model = Sequential()
model.add(Linear(2, 1))
model.add(RNN(1, 2))
model.skeleton()
def test_embedding_layer_with_ndim_2_input():
model = Sequential()
model.add(Embedding(3, 6))
output = model.feed_forward(np.array([[0], [1], [2]]))
assert output.shape == (3, 1, 6)
def test_embedding_layer_with_sequential_model():
model = Sequential()
model.add(Embedding(3, 6))
word_dim = np.array([0, 1, 2])
output = model.feed_forward(word_dim)
assert output.shape == (3, 6)