Skip to content

un_hadamard

UnHadamard

Bases: TensorOp

Convert hadamard encoded class representations into onehot probabilities.

Parameters:

Name Type Description Default
inputs Union[str, List[str]]

Key of the input tensor(s) to be converted.

 required
outputs Union[str, List[str]]

Key of the output tensor(s) as class probabilities.

 required
n_classes int

How many classes are there in the inputs.

 required
code_length Optional[int]

What code length to use. Will default to the smallest power of 2 which is >= the number of classes.

 None
mode Union[None, str, Iterable[str]]

What mode(s) to execute this Op in. For example, "train", "eval", "test", or "infer". To execute regardless of mode, pass None. To execute in all modes except for a particular one, you can pass an argument like "!infer" or "!train".

 None
ds_id Union[None, str, Iterable[str]]

What dataset id(s) to execute this Op in. To execute regardless of ds_id, pass None. To execute in all ds_ids except for a particular one, you can pass an argument like "!ds1".

 None
Source code in fastestimator/fastestimator/op/tensorop/un_hadamard.py 
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
@traceable()
class UnHadamard(TensorOp):
    """Convert hadamard encoded class representations into onehot probabilities.

    Args:
        inputs: Key of the input tensor(s) to be converted.
        outputs: Key of the output tensor(s) as class probabilities.
        n_classes: How many classes are there in the inputs.
        code_length: What code length to use. Will default to the smallest power of 2 which is >= the number of classes.
        mode: What mode(s) to execute this Op in. For example, "train", "eval", "test", or "infer". To execute
            regardless of mode, pass None. To execute in all modes except for a particular one, you can pass an argument
            like "!infer" or "!train".
        ds_id: What dataset id(s) to execute this Op in. To execute regardless of ds_id, pass None. To execute in all
            ds_ids except for a particular one, you can pass an argument like "!ds1".
    """
    def __init__(self,
                 inputs: Union[str, List[str]],
                 outputs: Union[str, List[str]],
                 n_classes: int,
                 code_length: Optional[int] = None,
                 mode: Union[None, str, Iterable[str]] = None,
                 ds_id: Union[None, str, Iterable[str]] = None) -> None:
        super().__init__(inputs=inputs, outputs=outputs, mode=mode, ds_id=ds_id)
        self.in_list, self.out_list = True, True
        self.n_classes = n_classes
        if code_length is None:
            code_length = 1 << (n_classes - 1).bit_length()
        if code_length <= 0 or (code_length & (code_length - 1) != 0):
            raise ValueError(f"code_length must be a positive power of 2, but got {code_length}.")
        if code_length < n_classes:
            raise ValueError(f"code_length must be >= n_classes, but got {code_length} and {n_classes}")
        self.code_length = code_length
        self.labels = None
        self.eps = None

    def build(self, framework: str, device: Optional[torch.device] = None) -> None:
        labels = hadamard(self.code_length).astype(np.float32)
        labels[np.arange(0, self.code_length, 2), 0] = -1  # Make first column alternate
        labels = labels[:self.n_classes]
        self.labels = to_tensor(labels, target_type=framework)
        max_prob = 0.99999  # This will only be approximate since the first column is alternating
        power = 1.0
        self.eps = to_tensor(
            np.array((self.code_length + 1) * math.pow((1.0 - max_prob) / (max_prob * (self.n_classes - 1)), 1 / power),
                     dtype=np.float32),
            target_type=framework)
        if framework == "torch":
            self.labels = self.labels.to(device)
            self.eps = self.eps.to(device)

    def forward(self, data: List[Tensor], state: Dict[str, Any]) -> List[Tensor]:
        results = []
        for elem in data:
            # L1 Distance
            x = reduce_sum(abs(expand_dims(elem, axis=1) - self.labels), axis=-1)
            x = iwd(x, power=1.0, eps=self.eps)
            results.append(x)
        return results