def test_sequential_side(M):
x1_np = 2
x2_np = 3
b_np = 1
x_np = np.array([1, 2, 3], dtype=np.float32)
x = ng.variable([M], initial_value=x_np)
x1 = ng.persistent_tensor(axes=(), initial_value=x1_np)
x2 = ng.persistent_tensor(axes=(), initial_value=x2_np)
x1_vo = ng.value_of(x1)
x2_vo = ng.value_of(x2)
b = ng.persistent_tensor(axes=(), initial_value=b_np)
y = ng.sequential([
x1_vo, x2_vo,
ng.assign(x1,
ng.sum(x, out_axes=()) + x1 * b + (1 - b)),
ng.assign(x2,
ng.mean(x, out_axes=()) + x2 * b + (1 - b)), x * 2
])
with ExecutorFactory() as ex:
main_effect = ex.executor((y, x1_vo, x2_vo, x1, x2))
current_values = ex.executor((x1, x2))
# Run main path #1
y_val, x1_init_val, x2_init_val, x1_final_val, x2_final_val = main_effect(
)
y_np = x_np * 2
assert np.allclose(y_val, y_np)
assert np.allclose(x1_init_val, x1_np)
assert np.allclose(x2_init_val, x2_np)
x1_np = np.sum(x_np) + x1_np * b_np + (1 - b_np)
x2_np = np.mean(x_np) + x2_np * b_np + (1 - b_np)
assert np.allclose(x1_final_val, x1_np)
assert np.allclose(x2_final_val, x2_np)
x1_val, x2_val = current_values()
assert np.allclose(x1_val, x1_np)
assert np.allclose(x2_val, x2_np)
# Run main path #2 (Should be the same as before)
y_val, x1_init_val, x2_init_val, x1_final_val, x2_final_val = main_effect(
)
y_np = x_np * 2
assert np.allclose(y_val, y_np)
assert np.allclose(x1_init_val, x1_np)
assert np.allclose(x2_init_val, x2_np)
x1_np = np.sum(x_np) + x1_np * b_np + (1 - b_np)
x2_np = np.mean(x_np) + x2_np * b_np + (1 - b_np)
assert np.allclose(x1_final_val, x1_np)
assert np.allclose(x2_final_val, x2_np)
'name': 'schedule',
'base_lr': 0.01,
'gamma': (1 / 250.)**(1 / 3.),
'schedule': [22, 44, 65]
}
optimizer = GradientDescentMomentum(lr_schedule,
0.0,
wdecay=0.0005,
iteration=inputs['iteration'])
train_prob = seq1(inputs['image'])
train_loss = ng.cross_entropy_multi(train_prob,
ng.one_hot(inputs['label'], axis=ax.Y))
batch_cost = ng.sequential(
[optimizer(train_loss),
ng.mean(train_loss, out_axes=())])
train_outputs = dict(batch_cost=batch_cost)
with closing(ngt.make_transformer()) as transformer:
train_computation = make_bound_computation(transformer, train_outputs,
inputs)
cbs = make_default_callbacks(transformer=transformer,
output_file=args.output_file,
frequency=args.iter_interval,
train_computation=train_computation,
total_iterations=args.num_iterations,
use_progress_bar=args.progress_bar)
loop_train(train_set, cbs)
schedule = list(np.arange(step, num_iterations, step))
learning_rate_policy = {'name': 'schedule',
'schedule': schedule,
'gamma': 0.95,
'base_lr': 0.01}
optimizer = GradientDescentMomentum(learning_rate=learning_rate_policy,
iteration=inputs['iteration'])
# Define the loss function (Cross entropy loss)
# Note that we convert the integer values of input['y'] to one hot here
fwd_prop = seq1(inputs['X'])
train_loss = ng.cross_entropy_multi(fwd_prop,
ng.one_hot(inputs['y'], axis=out_axis),
usebits=True)
# Train cost computation
batch_cost = ng.sequential([optimizer(train_loss), ng.mean(train_loss, out_axes=())])
train_computation = ng.computation([batch_cost, fwd_prop], "all")
train_outputs = dict(batch_cost=batch_cost)
# Forward prop of evaluation set
# Required for correct functioning of batch norm and dropout layers during inference mode
with Layer.inference_mode_on():
inference_prop = seq1(inputs['X'])
eval_loss = ng.cross_entropy_multi(inference_prop,
ng.one_hot(inputs['y'], axis=out_axis),
usebits=True)
eval_computation = ng.computation([ng.mean(eval_loss, out_axes=()), inference_prop], "all")
eval_outputs = dict(x_ent_loss=eval_loss)
# Computation for text generation - this is pure inference (fwd prop)
gen_computation = ng.computation(inference_prop, "all")
def __init__(self):
self.ng_computation = lambda Y, T: ng.mean(ng.square(Y - T),
out_axes=()) / 2.
def cifar_mean_subtract(x):
bgr_mean = ng.persistent_tensor(
axes=x.axes.find_by_name('C'),
initial_value=np.array([104., 119., 127.]))
return (x - bgr_mean) / 255.
seq1 = Sequential([Preprocess(functor=cifar_mean_subtract),
Affine(nout=200, weight_init=UniformInit(-0.1, 0.1), activation=Rectlin()),
Affine(axes=ax.Y, weight_init=UniformInit(-0.1, 0.1), activation=Softmax())])
optimizer = GradientDescentMomentum(0.1, 0.9)
train_prob = seq1(inputs['image'])
train_loss = ng.cross_entropy_multi(train_prob, ng.one_hot(inputs['label'], axis=ax.Y))
batch_cost = ng.sequential([optimizer(train_loss), ng.mean(train_loss, out_axes=())])
train_outputs = dict(batch_cost=batch_cost)
with Layer.inference_mode_on():
inference_prob = seq1(inputs['image'])
eval_loss = ng.cross_entropy_multi(inference_prob, ng.one_hot(inputs['label'], axis=ax.Y))
eval_outputs = dict(results=inference_prob, cross_ent_loss=eval_loss)
# Now bind the computations we are interested in
with closing(ngt.make_transformer()) as transformer:
train_computation = make_bound_computation(transformer, train_outputs, inputs)
loss_computation = make_bound_computation(transformer, eval_outputs, inputs)
cbs = make_default_callbacks(transformer=transformer,
output_file=args.output_file,
frequency=args.iter_interval,
optimizer = RMSProp(decay_rate=0.95,
learning_rate=2e-3,
epsilon=1e-6,
gradient_clip_value=gradient_clip_value)
# build network graph
one_hot_enc_out = one_hot_enc(inputs['inp_txt'])
one_hot_dec_out = one_hot_dec(inputs['prev_tgt'])
enc_out = enc(one_hot_enc_out)
dec_out = dec(one_hot_dec_out, init_state=enc_out)
output_prob = linear(dec_out)
loss = ng.cross_entropy_multi(output_prob,
ng.one_hot(inputs['tgt_txt'], axis=ax.Y),
usebits=True)
mean_cost = ng.mean(loss, out_axes=[])
updates = optimizer(loss)
train_outputs = dict(batch_cost=mean_cost, updates=updates)
loss_outputs = dict(cross_ent_loss=loss)
# inference graph
with Layer.inference_mode_on():
enc_out_inference = enc(one_hot_enc_out)
# Create decoder placeholders
axes = one_hot_dec_out.axes
axes = axes - axes.recurrent_axis() + ng.make_axis(length=1, name="REC")
decoder_input_inference = ng.placeholder(axes, name="input")
decoder_state_inference = ng.placeholder(enc_out_inference.axes,
name="state")
