def lstm_unit(hidden_t_prev, cell_t_prev, gates,
seq_lengths, timestep, forget_bias=0.0, drop_states=False):
D = cell_t_prev.shape[2]
G = gates.shape[2]
N = gates.shape[1]
t = (timestep * np.ones(shape=(N, D))).astype(np.int32)
assert t.shape == (N, D)
seq_lengths = (np.ones(shape=(N, D)) *
seq_lengths.reshape(N, 1)).astype(np.int32)
assert seq_lengths.shape == (N, D)
assert G == 4 * D
# Resize to avoid broadcasting inconsistencies with NumPy
gates = gates.reshape(N, 4, D)
cell_t_prev = cell_t_prev.reshape(N, D)
i_t = gates[:, 0, :].reshape(N, D)
f_t = gates[:, 1, :].reshape(N, D)
o_t = gates[:, 2, :].reshape(N, D)
g_t = gates[:, 3, :].reshape(N, D)
i_t = sigmoid(i_t)
f_t = sigmoid(f_t + forget_bias)
o_t = sigmoid(o_t)
g_t = tanh(g_t)
valid = (t < seq_lengths).astype(np.int32)
assert valid.shape == (N, D)
cell_t = ((f_t * cell_t_prev) + (i_t * g_t)) * (valid) + \
(1 - valid) * cell_t_prev * (1 - drop_states)
assert cell_t.shape == (N, D)
hidden_t = (o_t * tanh(cell_t)) * valid + hidden_t_prev * (
1 - valid) * (1 - drop_states)
hidden_t = hidden_t.reshape(1, N, D)
cell_t = cell_t.reshape(1, N, D)
return hidden_t, cell_t
def gru_reference(input, hidden_input,
reset_gate_w, reset_gate_b,
update_gate_w, update_gate_b,
output_gate_w, output_gate_b,
seq_lengths, drop_states=False,
linear_before_reset=False):
D = hidden_input.shape[hidden_input.ndim - 1]
T = input.shape[0]
N = input.shape[1]
G = input.shape[2]
print("Dimensions: T= ", T, " N= ", N, " G= ", G, " D= ", D)
hidden = np.zeros(shape=(T + 1, N, D))
hidden[0, :, :] = hidden_input
for t in range(T):
input_t = input[t].reshape(1, N, G)
hidden_t_prev = hidden[t].reshape(1, N, D)
# Split input contributions for three gates.
input_t = input_t.reshape(N, 3, D)
input_reset = input_t[:, 0, :].reshape(N, D)
input_update = input_t[:, 1, :].reshape(N, D)
input_output = input_t[:, 2, :].reshape(N, D)
reset_gate = np.dot(hidden_t_prev, reset_gate_w.T) + reset_gate_b
reset_gate = reset_gate + input_reset
update_gate = np.dot(hidden_t_prev, update_gate_w.T) + update_gate_b
update_gate = update_gate + input_update
if linear_before_reset:
with_linear = np.dot(
hidden_t_prev, output_gate_w.T) + output_gate_b
output_gate = sigmoid(reset_gate) * with_linear
else:
with_reset = hidden_t_prev * sigmoid(reset_gate)
output_gate = np.dot(with_reset, output_gate_w.T) + output_gate_b
output_gate = output_gate + input_output
gates_out_t = np.concatenate(
(reset_gate, update_gate, output_gate),
axis=2,
)
print(reset_gate, update_gate, output_gate, gates_out_t, sep="\n")
(hidden_t, ) = gru_unit(
hidden_t_prev,
gates_out_t,
seq_lengths,
t,
drop_states=drop_states
)
hidden[t + 1] = hidden_t
return (
hidden[1:],
hidden[-1].reshape(1, N, D),
)
def gru_reference(input, hidden_input,
reset_gate_w, reset_gate_b,
update_gate_w, update_gate_b,
output_gate_w, output_gate_b,
seq_lengths, drop_states=False,
linear_before_reset=False):
D = hidden_input.shape[hidden_input.ndim - 1]
T = input.shape[0]
N = input.shape[1]
G = input.shape[2]
print("Dimensions: T= ", T, " N= ", N, " G= ", G, " D= ", D)
hidden = np.zeros(shape=(T + 1, N, D))
hidden[0, :, :] = hidden_input
for t in range(T):
input_t = input[t].reshape(1, N, G)
hidden_t_prev = hidden[t].reshape(1, N, D)
# Split input contributions for three gates.
input_t = input_t.reshape(N, 3, D)
input_reset = input_t[:, 0, :].reshape(N, D)
input_update = input_t[:, 1, :].reshape(N, D)
input_output = input_t[:, 2, :].reshape(N, D)
reset_gate = np.dot(hidden_t_prev, reset_gate_w.T) + reset_gate_b
reset_gate = reset_gate + input_reset
update_gate = np.dot(hidden_t_prev, update_gate_w.T) + update_gate_b
update_gate = update_gate + input_update
if linear_before_reset:
with_linear = np.dot(hidden_t_prev, output_gate_w.T) + output_gate_b
output_gate = sigmoid(reset_gate) * with_linear
else:
with_reset = hidden_t_prev * sigmoid(reset_gate)
output_gate = np.dot(with_reset, output_gate_w.T) + output_gate_b
output_gate = output_gate + input_output
gates_out_t = np.concatenate(
(reset_gate, update_gate, output_gate),
axis=2,
)
print(reset_gate, update_gate, output_gate, gates_out_t, sep="\n")
(hidden_t, ) = gru_unit(
hidden_t_prev,
gates_out_t,
seq_lengths,
t,
drop_states=drop_states
)
hidden[t + 1] = hidden_t
return (
hidden[1:],
hidden[-1].reshape(1, N, D),
)
def gru_unit(*args, **kwargs):
'''
Implements one GRU unit, for one time step
Shapes:
hidden_t_prev.shape = (1, N, D)
gates_out_t.shape = (1, N, G)
seq_lenths.shape = (N,)
'''
drop_states = kwargs.get('drop_states', False)
sequence_lengths = kwargs.get('sequence_lengths', True)
if sequence_lengths:
hidden_t_prev, gates_out_t, seq_lengths, timestep = args
else:
hidden_t_prev, gates_out_t, timestep = args
N = hidden_t_prev.shape[1]
D = hidden_t_prev.shape[2]
G = gates_out_t.shape[2]
t = (timestep * np.ones(shape=(N, D))).astype(np.int32)
assert t.shape == (N, D)
assert G == 3 * D
# Calculate reset, update, and output gates separately
# because output gate depends on reset gate.
gates_out_t = gates_out_t.reshape(N, 3, D)
reset_gate_t = gates_out_t[:, 0, :].reshape(N, D)
update_gate_t = gates_out_t[:, 1, :].reshape(N, D)
output_gate_t = gates_out_t[:, 2, :].reshape(N, D)
# Calculate gate outputs.
reset_gate_t = sigmoid(reset_gate_t)
update_gate_t = sigmoid(update_gate_t)
output_gate_t = tanh(output_gate_t)
if sequence_lengths:
seq_lengths = (np.ones(shape=(N, D)) *
seq_lengths.reshape(N, 1)).astype(np.int32)
assert seq_lengths.shape == (N, D)
valid = (t < seq_lengths).astype(np.int32)
else:
valid = np.ones(shape=(N, D))
assert valid.shape == (N, D)
hidden_t = update_gate_t * hidden_t_prev + (1 -
update_gate_t) * output_gate_t
hidden_t = hidden_t * valid + hidden_t_prev * (1 - valid) * (1 -
drop_states)
hidden_t = hidden_t.reshape(1, N, D)
return (hidden_t, )
def gru_unit(*args, **kwargs):
'''
Implements one GRU unit, for one time step
Shapes:
hidden_t_prev.shape = (1, N, D)
gates_out_t.shape = (1, N, G)
seq_lenths.shape = (N,)
'''
drop_states = kwargs.get('drop_states', False)
sequence_lengths = kwargs.get('sequence_lengths', True)
if sequence_lengths:
hidden_t_prev, gates_out_t, seq_lengths, timestep = args
else:
hidden_t_prev, gates_out_t, timestep = args
N = hidden_t_prev.shape[1]
D = hidden_t_prev.shape[2]
G = gates_out_t.shape[2]
t = (timestep * np.ones(shape=(N, D))).astype(np.int32)
assert t.shape == (N, D)
assert G == 3 * D
# Calculate reset, update, and output gates separately
# because output gate depends on reset gate.
gates_out_t = gates_out_t.reshape(N, 3, D)
reset_gate_t = gates_out_t[:, 0, :].reshape(N, D)
update_gate_t = gates_out_t[:, 1, :].reshape(N, D)
output_gate_t = gates_out_t[:, 2, :].reshape(N, D)
# Calculate gate outputs.
reset_gate_t = sigmoid(reset_gate_t)
update_gate_t = sigmoid(update_gate_t)
output_gate_t = tanh(output_gate_t)
if sequence_lengths:
seq_lengths = (np.ones(shape=(N, D)) *
seq_lengths.reshape(N, 1)).astype(np.int32)
assert seq_lengths.shape == (N, D)
valid = (t < seq_lengths).astype(np.int32)
else:
valid = np.ones(shape=(N, D))
assert valid.shape == (N, D)
hidden_t = update_gate_t * hidden_t_prev + (1 - update_gate_t) * output_gate_t
hidden_t = hidden_t * valid + hidden_t_prev * (1 - valid) * (1 - drop_states)
hidden_t = hidden_t.reshape(1, N, D)
return (hidden_t, )
