tensorplay.testing API

Legacy alias of assert_close() with positional tolerances.

Asserts that actual and expected are close.

If actual and expected are strided and finite, they are considered close if

Non-finite values (-inf and inf) are only considered close if and only if they are equal. NaN’s are only considered equal to each other if equal_nan is True.

In addition, they are only considered close if they have the same

• device (if check_device is True),

• dtype (if check_dtype is True),

• layout (if check_layout is True), and

• stride (if check_stride is True).

If either actual or expected is a scalar or a nested python container, the other side is converted to a tensor-like value before the comparison.

Parameters:

• actual (Any) – Actual input.

• expected (Any) – Expected input.

• allow_subclasses (bool) – If True (default) and other than exact type match, inputs that are subclasses of each other are considered close.

• rtol (Optional[float]) – Relative tolerance. If specified atol must also be specified. If omitted, default values based on the dtype are selected. See below for details.

• atol (Optional[float]) – Absolute tolerance. If specified rtol must also be specified. If omitted, default values based on the dtype are selected. See below for details.

• equal_nan (bool) – If True, two NaN values are considered equal. Defaults to False.

• check_device (bool) – If True (default), asserts that corresponding tensors are on the same device.

• check_dtype (bool) – If True (default), asserts that corresponding tensors have the same dtype.

• check_layout (bool) – If True (default), asserts that corresponding tensors have the same layout.

• check_stride (bool) – If True, asserts that corresponding strided tensors have the same stride.

• msg (Optional[Union[str, Callable[[str], str]]]) – Optional error message to use in case of failure.

Raises:

• ValueError – If only rtol or only atol is specified.

• AssertionError – If corresponding values are not close.

Default tolerances by dtype:

Returns the default absolute and relative testing tolerances for a set of inputs based on the dtype.

Returns:

Loosest tolerances of all input dtypes.

Return type:

(Tuple[float, float])

Gets absolute and relative tolerances to be used for numeric comparisons.

If both rtol and atol are specified, this is a no-op. If neither is specified, default_tolerances() is used. Specifying only one raises a ValueError, since a single tolerance might lead to surprising results.

Creates a tensor with the given shape, device, and dtype, filled with values drawn uniformly from [low, high).

If low or high are outside the range of the dtype’s representable finite values, they are clamped to the lowest or highest representable finite value, respectively. If None, they default to -9 and 9 respectively (0 and 2 for bool).

Parameters:

• shape (Tuple[int, ...]) – Single integer or a collection of integers defining the shape of the output tensor.

• dtype (tensorplay.dtype) – The data type of the returned tensor.

• device (Union[str, tensorplay.device]) – The device of the returned tensor.

• low (Optional[Number]) – Sets the lower limit of the range of the values in the returned tensor.

• high (Optional[Number]) – Sets the upper limit of the range of the values in the returned tensor.

• requires_grad (bool) – If True, the returned tensor is set to require gradient.

• noncontiguous (bool) – If True, the returned tensor is non-contiguous.

• exclude_zero (bool) – If True, zeros in the returned tensor are replaced by the smallest normal value of the dtype (1 for boolean and integral dtypes).

Raises:

• ValueError – If low >= high, or if the interval [low, high) does not intersect the dtype’s representable range.

• TypeError – For unsupported dtypes.