def reinforce(self, train_dataset, encoder, output_dir):
interpreter = self.interpreter()
collate = Collate(
test_case_tensor=CollateOptions(False, 0, 0),
variables_tensor=CollateOptions(True, 0, 0),
previous_actions=CollateOptions(True, 0, -1),
hidden_state=CollateOptions(False, 0, 0),
state=CollateOptions(False, 0, 0),
ground_truth_actions=CollateOptions(True, 0, -1),
reward=CollateOptions(False, 0, 0)
)
collate_fn = Sequence(OrderedDict([
("to_episode", Map(self.to_episode(encoder,
interpreter))),
("flatten", Flatten()),
("transform", Map(self.transform(
encoder, interpreter, Parser()))),
("collate", collate.collate)
]))
model = self.prepare_model(encoder)
optimizer = self.prepare_optimizer(model)
train_REINFORCE(
os.path.join(output_dir, "pretrain"), os.path.join(output_dir, "RL"),
train_dataset,
self.prepare_synthesizer(model, encoder, interpreter),
model, optimizer,
torch.nn.Sequential(OrderedDict([
("policy",
torch.nn.Sequential(OrderedDict([
("loss",
Apply(
module=mlprogram.nn.action_sequence.Loss(
reduction="none"
),
in_keys=[
"rule_probs",
"token_probs",
"reference_probs",
"ground_truth_actions",
],
out_key="action_sequence_loss",
)),
("weight_by_reward",
Apply(
[("reward", "lhs"),
("action_sequence_loss", "rhs")],
"action_sequence_loss",
mlprogram.nn.Function(Mul())))
]))),
("value",
torch.nn.Sequential(OrderedDict([
("reshape_reward",
Apply(
[("reward", "x")],
"value_loss_target",
Reshape([-1, 1]))),
("BCE",
Apply(
[("value", "input"),
("value_loss_target", "target")],
"value_loss",
torch.nn.BCELoss(reduction='sum'))),
("reweight",
Apply(
[("value_loss", "lhs")],
"value_loss",
mlprogram.nn.Function(Mul()),
constants={"rhs": 1e-2})),
]))),
("aggregate",
Apply(
["action_sequence_loss", "value_loss"],
"loss",
AggregatedLoss())),
("normalize",
Apply(
[("loss", "lhs")],
"loss",
mlprogram.nn.Function(Div()),
constants={"rhs": 1})),
("pick",
mlprogram.nn.Function(
Pick("loss")))
])),
EvaluateSynthesizer(
train_dataset,
self.prepare_synthesizer(model, encoder, interpreter,
rollout=False),
{}, top_n=[]),
"generation_rate",
metrics.use_environment(
metric=metrics.TestCaseResult(
interpreter=interpreter,
metric=metrics.use_environment(
metric=metrics.Iou(),
in_keys=["actual", "expected"],
value_key="actual",
)
),
in_keys=["test_cases", "actual"],
value_key="actual",
transform=Threshold(threshold=0.9, dtype="float"),
),
collate_fn,
1, 1,
Epoch(10), evaluation_interval=Epoch(10),
snapshot_interval=Epoch(10),
use_pretrained_model=True,
use_pretrained_optimizer=True,
threshold=1.0)
def prepare_synthesizer(self, model, encoder, interpreter, rollout=True):
collate = Collate(
test_case_tensor=CollateOptions(False, 0, 0),
input_feature=CollateOptions(False, 0, 0),
test_case_feature=CollateOptions(False, 0, 0),
reference_features=CollateOptions(True, 0, 0),
variables_tensor=CollateOptions(True, 0, 0),
previous_actions=CollateOptions(True, 0, -1),
hidden_state=CollateOptions(False, 0, 0),
state=CollateOptions(False, 0, 0),
ground_truth_actions=CollateOptions(True, 0, -1)
)
subsampler = ActionSequenceSampler(
encoder, IsSubtype(),
Sequence(OrderedDict([
("tinput",
Apply(
module=TransformInputs(),
in_keys=["test_cases"],
out_key="test_case_tensor",
)),
("tvariable",
Apply(
module=TransformVariables(),
in_keys=["variables", "test_case_tensor"],
out_key="variables_tensor"
)),
])),
Compose(OrderedDict([
("add_previous_actions",
Apply(
module=AddPreviousActions(encoder, n_dependent=1),
in_keys=["action_sequence", "reference"],
out_key="previous_actions",
constants={"train": False},
)),
("add_state", AddState("state")),
("add_hidden_state", AddState("hidden_state"))
])),
collate, model,
rng=np.random.RandomState(0))
subsampler = mlprogram.samplers.transform(
subsampler,
Parser().unparse
)
subsynthesizer = SMC(
5, 1,
subsampler,
max_try_num=1,
to_key=Pick("action_sequence"),
rng=np.random.RandomState(0)
)
sampler = SequentialProgramSampler(
subsynthesizer,
Apply(
module=TransformInputs(),
in_keys=["test_cases"],
out_key="test_case_tensor",
),
collate,
model.encode_input,
interpreter=interpreter,
expander=Expander(),
rng=np.random.RandomState(0))
if rollout:
sampler = FilteredSampler(
sampler,
metrics.use_environment(
metric=metrics.TestCaseResult(
interpreter,
metric=metrics.use_environment(
metric=metrics.Iou(),
in_keys=["actual", "expected"],
value_key="actual",
)
),
in_keys=["test_cases", "actual"],
value_key="actual"
),
1.0
)
return SMC(3, 1, sampler, rng=np.random.RandomState(0),
to_key=Pick("interpreter_state"), max_try_num=1)
else:
sampler = SamplerWithValueNetwork(
sampler,
Sequence(OrderedDict([
("tinput",
Apply(
module=TransformInputs(),
in_keys=["test_cases"],
out_key="test_case_tensor",
)),
("tvariable",
Apply(
module=TransformVariables(),
in_keys=["variables", "test_case_tensor"],
out_key="variables_tensor"
)),
])),
collate,
torch.nn.Sequential(OrderedDict([
("encoder", model.encoder),
("value", model.value),
("pick",
mlprogram.nn.Function(
Pick("value")))
])))
synthesizer = SynthesizerWithTimeout(
SMC(3, 1, sampler, rng=np.random.RandomState(0),
to_key=Pick("interpreter_state"),
max_try_num=1),
1
)
return FilteredSynthesizer(
synthesizer,
metrics.use_environment(
metric=metrics.TestCaseResult(
interpreter,
metric=metrics.use_environment(
metric=metrics.Iou(),
in_keys=["actual", "expected"],
value_key="actual",
)
),
in_keys=["test_cases", "actual"],
value_key="actual"
),
1.0
)
def pretrain(self, output_dir):
dataset = Dataset(4, 1, 2, 1, 45, seed=0)
"""
"""
train_dataset = ListDataset([
Environment(
{"ground_truth": Circle(1)},
set(["ground_truth"]),
),
Environment(
{"ground_truth": Rectangle(1, 2)},
set(["ground_truth"]),
),
Environment(
{"ground_truth": Rectangle(1, 1)},
set(["ground_truth"]),
),
Environment(
{"ground_truth": Rotation(45, Rectangle(1, 1))},
set(["ground_truth"]),
),
Environment(
{"ground_truth": Translation(1, 1, Rectangle(1, 1))},
set(["ground_truth"]),
),
Environment(
{"ground_truth": Difference(Circle(1), Circle(1))},
set(["ground_truth"]),
),
Environment(
{"ground_truth": Union(Rectangle(1, 2), Circle(1))},
set(["ground_truth"]),
),
Environment(
{"ground_truth": Difference(Rectangle(1, 1), Circle(1))},
set(["ground_truth"]),
),
])
interpreter = self.interpreter()
train_dataset = data_transform(
train_dataset,
Apply(
module=AddTestCases(interpreter),
in_keys=["ground_truth"],
out_key="test_cases",
is_out_supervision=False,
)
)
encoder = self.prepare_encoder(dataset, Parser())
collate = Collate(
test_case_tensor=CollateOptions(False, 0, 0),
variables_tensor=CollateOptions(True, 0, 0),
previous_actions=CollateOptions(True, 0, -1),
hidden_state=CollateOptions(False, 0, 0),
state=CollateOptions(False, 0, 0),
ground_truth_actions=CollateOptions(True, 0, -1)
)
collate_fn = Sequence(OrderedDict([
("to_episode", Map(self.to_episode(encoder,
interpreter))),
("flatten", Flatten()),
("transform", Map(self.transform(
encoder, interpreter, Parser()))),
("collate", collate.collate)
]))
model = self.prepare_model(encoder)
optimizer = self.prepare_optimizer(model)
train_supervised(
os.path.join(output_dir, "pretrain"),
train_dataset, model, optimizer,
torch.nn.Sequential(OrderedDict([
("loss",
Apply(
module=Loss(
reduction="sum",
),
in_keys=[
"rule_probs",
"token_probs",
"reference_probs",
"ground_truth_actions",
],
out_key="action_sequence_loss",
)),
("normalize", # divided by batch_size
Apply(
[("action_sequence_loss", "lhs")],
"loss",
mlprogram.nn.Function(Div()),
constants={"rhs": 1})),
("pick",
mlprogram.nn.Function(
Pick("loss")))
])),
None, "score",
collate_fn,
1, Epoch(100), evaluation_interval=Epoch(10),
snapshot_interval=Epoch(100)
)
return encoder, train_dataset
def parser():
return Parser()