def test_infinite_bounds_are_valid_for_floats(self, minimum, maximum):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.DOUBLE
tensor_spec.min.double = minimum
tensor_spec.max.double = maximum
tensor_spec.name = 'foo'
tensor_spec_utils.bounds(tensor_spec)
def test_nonnumeric_type_raises_error(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.STRING
tensor_spec.max.int32s.array[:] = [1]
tensor_spec.name = 'foo'
with self.assertRaisesRegex(ValueError, 'foo.*non-numeric.*string'):
tensor_spec_utils.bounds(tensor_spec)
def test_invalid_min_shape(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.UINT32
tensor_spec.min.uint32s.array[:] = [1, 2]
with self.assertRaisesRegex(
ValueError, 'Scalar tensors must have exactly 1 element.*'):
tensor_spec_utils.bounds(tensor_spec)
def test_max_mismatches_type_raises_error(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.UINT32
tensor_spec.max.int32s.array[:] = [1]
tensor_spec.name = 'foo'
with self.assertRaisesRegex(ValueError, 'foo.*uint32.*max.*int32'):
tensor_spec_utils.bounds(tensor_spec)
def test_invalid_max_shape(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.UINT32
tensor_spec.max.uint32s.array[:] = [1, 2]
tensor_spec.shape[:] = (2, 2)
with self.assertRaisesRegex(
ValueError, 'cannot reshape array of size .* into shape.*'):
tensor_spec_utils.bounds(tensor_spec)
def test_max_less_than_min_raises_error(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.INT32
tensor_spec.max.int32s.array[:] = [-1]
tensor_spec.min.int32s.array[:] = [1]
tensor_spec.name = 'foo'
with self.assertRaisesRegex(ValueError, 'foo.*min 1.*max -1'):
tensor_spec_utils.bounds(tensor_spec)
def test_invalid_min_var_shape(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.INT32
tensor_spec.min.int32s.array[:] = [-1, -1]
tensor_spec.max.int32s.array[:] = [1]
tensor_spec.shape[:] = (-1, )
with self.assertRaisesRegex(
ValueError, "TensorSpec's with variable length shapes "
'can only have scalar ranges.'):
tensor_spec_utils.bounds(tensor_spec)
def test_min_and_max(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.INT32
tensor_spec.min.int32s.array[:] = [-1]
tensor_spec.max.int32s.array[:] = [1]
bounds = tensor_spec_utils.bounds(tensor_spec)
self.assertEqual((-1, 1), bounds)
def test_min_and_max_legacy(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.INT32
tensor_spec.min.int32 = -1
tensor_spec.max.int32 = 1
bounds = tensor_spec_utils.bounds(tensor_spec)
self.assertEqual((-1, 1), bounds)
def _create_test_value(spec):
"""Creates a scalar test value consistent with the TensorSpec `spec`."""
if _is_numeric_type(spec.dtype):
return tensor_spec_utils.bounds(spec).min,
else:
np_type = tensor_utils.data_type_to_np_type(spec.dtype)
return np_type()
def tensor_spec_to_dm_env_spec(
tensor_spec: dm_env_rpc_pb2.TensorSpec) -> specs.Array:
"""Returns a dm_env spec given a dm_env_rpc TensorSpec.
Args:
tensor_spec: A dm_env_rpc TensorSpec protobuf.
Returns:
Either a DiscreteArray, BoundedArray, StringArray or Array, depending on the
content of the TensorSpec.
"""
np_type = tensor_utils.data_type_to_np_type(tensor_spec.dtype)
if tensor_spec.HasField('min') or tensor_spec.HasField('max'):
bounds = tensor_spec_utils.bounds(tensor_spec)
if (not tensor_spec.shape and np.issubdtype(np_type, np.integer)
and bounds.min == 0 and tensor_spec.HasField('max')):
return specs.DiscreteArray(num_values=bounds.max + 1,
dtype=np_type,
name=tensor_spec.name)
else:
return specs.BoundedArray(shape=tensor_spec.shape,
dtype=np_type,
name=tensor_spec.name,
minimum=bounds.min,
maximum=bounds.max)
else:
if tensor_spec.dtype == dm_env_rpc_pb2.DataType.STRING:
return specs.StringArray(shape=tensor_spec.shape,
name=tensor_spec.name)
else:
return specs.Array(shape=tensor_spec.shape,
dtype=np_type,
name=tensor_spec.name)
def test_max_broadcast(self): tensor_spec = dm_env_rpc_pb2.TensorSpec() tensor_spec.dtype = dm_env_rpc_pb2.DataType.UINT32 tensor_spec.max.uint32s.array[:] = [1] tensor_spec.shape[:] = (2, 2) bounds = tensor_spec_utils.bounds(tensor_spec) np.testing.assert_array_equal(np.full(tensor_spec.shape, 0), bounds.min) np.testing.assert_array_equal(np.full(tensor_spec.shape, 1), bounds.max)
def test_broadcast_var_shape(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.INT32
tensor_spec.min.int32s.array[:] = [-1]
tensor_spec.max.int32s.array[:] = [1]
tensor_spec.shape[:] = (-1, )
bounds = tensor_spec_utils.bounds(tensor_spec)
self.assertEqual((-1, 1), bounds)
def test_max_scalar_doesnt_broadcast(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.UINT32
tensor_spec.max.uint32s.array[:] = [1]
tensor_spec.shape[:] = (2, 2)
bounds = tensor_spec_utils.bounds(tensor_spec)
self.assertEqual(0, bounds.min)
self.assertEqual(1, bounds.max)
def test_error_if_min_or_max_cannot_be_safely_cast_to_dtype(
self, minimum, maximum):
name = 'foo'
dtype = dm_env_rpc_pb2.DataType.INT8
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dtype
tensor_spec.min.int8 = minimum
tensor_spec.max.int8 = maximum
tensor_spec.name = name
with self.assertRaisesWithLiteralMatch(
ValueError,
tensor_spec_utils._BOUNDS_CANNOT_BE_SAFELY_CAST_TO_DTYPE.format(
name=name,
minimum=minimum,
maximum=maximum,
dtype=dm_env_rpc_pb2.DataType.Name(dtype).lower(),
)):
tensor_spec_utils.bounds(tensor_spec)
def test_max_n_shape(self):
maximum = np.array([[1, 2], [3, 4]])
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.UINT32
tensor_spec.max.uint32s.array[:] = maximum.flatten().data.tolist()
tensor_spec.shape[:] = maximum.shape
bounds = tensor_spec_utils.bounds(tensor_spec)
np.testing.assert_array_equal(np.full(maximum.shape, 0), bounds.min)
np.testing.assert_array_equal(maximum, bounds.max)
def test_all_numerical_observations_in_range(self):
numeric_uids = (uid for uid, spec in self.specs.observations.items()
if _is_numeric_type(spec.dtype))
response = self.step(requested_observations=numeric_uids)
for uid, observation in response.observations.items():
spec = self.specs.observations[uid]
with self.subTest(uid=uid, name=spec.name):
unpacked = tensor_utils.unpack_tensor(observation)
bounds = tensor_spec_utils.bounds(spec)
_assert_less_equal(unpacked, bounds.max)
_assert_greater_equal(unpacked, bounds.min)
def _above_max(spec):
"""Returns a value above spec's max or None if none."""
if not spec.HasField('max'):
return None
np_type = tensor_utils.data_type_to_np_type(spec.dtype)
max_type_value = _np_range_info(np_type).max
if max_type_value > tensor_spec_utils.bounds(spec).max:
return max_type_value
else:
return None
def _below_min(spec):
"""Returns a value below spec's min or None if none."""
if not spec.HasField('min'):
return None
np_type = tensor_utils.data_type_to_np_type(spec.dtype)
min_type_value = _np_range_info(np_type).min
if min_type_value < tensor_spec_utils.bounds(spec).min:
return min_type_value
else:
return None
def test_min(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.UINT32
tensor_spec.min.uint32s.array[:] = [1]
bounds = tensor_spec_utils.bounds(tensor_spec)
self.assertEqual((1, 2**32 - 1), bounds)
def test_unbounded_signed(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.INT32
bounds = tensor_spec_utils.bounds(tensor_spec)
self.assertEqual((-2**31, 2**31 - 1), bounds)
def test_min_0_stays_0(self):
tensor_spec = dm_env_rpc_pb2.TensorSpec()
tensor_spec.dtype = dm_env_rpc_pb2.DataType.INT8
tensor_spec.min.int8s.array = b'\x00'
tensor_spec.name = 'foo'
self.assertEqual((0, 127), tensor_spec_utils.bounds(tensor_spec))
