Metric#

class ignite.metrics.metric.Metric(output_transform=<function Metric.<lambda>>, device=device(type='cpu'), skip_unrolling=False)[source]#

Base class for all Metrics.

Parameters

output_transform (Callable) – a callable that is used to transform the Engine’s process_function’s output into the form expected by the metric. This can be useful if, for example, you have a multi-output model and you want to compute the metric with respect to one of the outputs. By default, metrics require the output as (y_pred, y) or {'y_pred': y_pred, 'y': y}.
device (Union[str, device]) – specifies which device updates are accumulated on. Setting the metric’s device to be the same as your update arguments ensures the update method is non-blocking. By default, CPU.

skip_unrolling (bool) –

specifies whether output should be unrolled before being fed to update method. Should be true for multi-output model, for example, if y_pred contains multi-ouput as (y_pred_a, y_pred_b) Alternatively, output_transform can be used to handle this.

Examples

The following example shows a custom loss metric that expects input from a multi-output model.

import torch
import torch.nn as nn
import torch.nn.functional as F

from ignite.engine import create_supervised_evaluator
from ignite.metrics import Loss

class MyLoss(nn.Module):
    def __init__(self, ca: float = 1.0, cb: float = 1.0) -> None:
        super().__init__()
        self.ca = ca
        self.cb = cb

    def forward(self,
                y_pred: Tuple[torch.Tensor, torch.Tensor],
                y_true: Tuple[torch.Tensor, torch.Tensor]) -> torch.Tensor:
        a_true, b_true = y_true
        a_pred, b_pred = y_pred
        return self.ca * F.mse_loss(a_pred, a_true) + self.cb * F.cross_entropy(b_pred, b_true)


def prepare_batch(batch, device, non_blocking):
    return torch.rand(4, 1), (torch.rand(4, 1), torch.rand(4, 2))


class MyModel(nn.Module):

    def forward(self, x):
        return torch.rand(4, 1), torch.rand(4, 2)


model = MyModel()

device = "cpu"
loss = MyLoss(0.5, 1.0)
metrics = {
    "Loss": Loss(loss, skip_unrolling=True)
}
train_evaluator = create_supervised_evaluator(model, metrics, device, prepare_batch=prepare_batch)


data = range(10)
train_evaluator.run(data)
train_evaluator.state.metrics["Loss"]

required_output_keys#

dictionary defines required keys to be found in engine.state.output if the latter is a dictionary. By default, ("y_pred", "y"). This is useful with custom metrics that can require other arguments than predictions y_pred and targets y. See an example below.

Type: Optional[Tuple]

Examples

Let’s implement a custom metric that requires y_pred, y and x as input for update function. In the example below we show how to setup standard metric like Accuracy and the custom metric using by an evaluator created with create_supervised_evaluator() method.

For more information on how metric works with Engine, visit Attach Engine API.

# https://discuss.pytorch.org/t/how-access-inputs-in-custom-ignite-metric/91221/5

import torch
import torch.nn as nn

from ignite.metrics import Metric, Accuracy
from ignite.engine import create_supervised_evaluator

class CustomMetric(Metric):

    required_output_keys = ("y_pred", "y", "x")

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

    def update(self, output):
        y_pred, y, x = output
        # ...

    def reset(self):
        # ...
        pass

    def compute(self):
        # ...
        pass

model = ...

metrics = {
    "Accuracy": Accuracy(),
    "CustomMetric": CustomMetric()
}

evaluator = create_supervised_evaluator(
    model,
    metrics=metrics,
    output_transform=lambda x, y, y_pred: {"x": x, "y": y, "y_pred": y_pred}
)

res = evaluator.run(data)

Changed in version 0.4.2: required_output_keys became public attribute.

Changed in version 0.5.1: skip_unrolling argument is added.

Methods

`attach`	Attaches current metric to provided engine.
`completed`	Helper method to compute metric's value and put into the engine.
`compute`	Computes the metric based on its accumulated state.
`detach`	Detaches current metric from the engine and no metric's computation is done during the run.
`is_attached`	Checks if current metric is attached to provided engine.
`iteration_completed`	Helper method to update metric's computation.
`load_state_dict`	Method replaces internal state of the class with provided state dict data.
`reset`	Resets the metric to its initial state.
`started`	Helper method to start data gathering for metric's computation.
`state_dict`	Method returns state dict with attributes of the metric specified in its _state_dict_all_req_keys attribute.
`update`	Updates the metric's state using the passed batch output.

attach(engine, name, usage=<ignite.metrics.metric.EpochWise object>)[source]#

Attaches current metric to provided engine. On the end of engine’s run, engine.state.metrics dictionary will contain computed metric’s value under provided name.

Parameters

engine (Engine) – the engine to which the metric must be attached
name (str) – the name of the metric to attach
usage (Union[str, MetricUsage]) – the usage of the metric. Valid string values should be ignite.metrics.metric.EpochWise.usage_name (default) or ignite.metrics.metric.BatchWise.usage_name.

Return type

None

Examples

metric = ...
metric.attach(engine, "mymetric")

assert "mymetric" in engine.run(data).metrics

assert metric.is_attached(engine)

Example with usage:

metric = ...
metric.attach(engine, "mymetric", usage=BatchWise.usage_name)

assert "mymetric" in engine.run(data).metrics

assert metric.is_attached(engine, usage=BatchWise.usage_name)

completed(engine, name)[source]#

Helper method to compute metric’s value and put into the engine. It is automatically attached to the engine with attach(). If metrics’ value is torch tensor, it is explicitly sent to CPU device.

Parameters

engine (Engine) – the engine to which the metric must be attached
name (str) – the name of the metric used as key in dict engine.state.metrics

Return type

None

Changed in version 0.4.3: Added dict in metrics results.

Changed in version 0.4.5: metric’s value is put on CPU if torch tensor.

abstract compute()[source]#

Computes the metric based on its accumulated state.

By default, this is called at the end of each epoch.

Returns

the actual quantity of interest. However, if a Mapping is returned, it will be (shallow) flattened into engine.state.metrics when completed() is called.

Return type

Any

Raises

NotComputableError – raised when the metric cannot be computed.

detach(engine, usage=<ignite.metrics.metric.EpochWise object>)[source]#

Detaches current metric from the engine and no metric’s computation is done during the run. This method in conjunction with attach() can be useful if several metrics need to be computed with different periods. For example, one metric is computed every training epoch and another metric (e.g. more expensive one) is done every n-th training epoch.

Parameters

engine (Engine) – the engine from which the metric must be detached
usage (Union[str, MetricUsage]) – the usage of the metric. Valid string values should be ‘epoch_wise’ (default) or ‘batch_wise’.

Return type

None

Examples

metric = ...
engine = ...
metric.detach(engine)

assert "mymetric" not in engine.run(data).metrics

assert not metric.is_attached(engine)

Example with usage:

metric = ...
engine = ...
metric.detach(engine, usage="batch_wise")

assert "mymetric" not in engine.run(data).metrics

assert not metric.is_attached(engine, usage="batch_wise")

is_attached(engine, usage=<ignite.metrics.metric.EpochWise object>)[source]#

Checks if current metric is attached to provided engine. If attached, metric’s computed value is written to engine.state.metrics dictionary.

Parameters

engine (Engine) – the engine checked from which the metric should be attached
usage (Union[str, MetricUsage]) – the usage of the metric. Valid string values should be ‘epoch_wise’ (default) or ‘batch_wise’.

Return type

bool

iteration_completed(engine)[source]#

Helper method to update metric’s computation. It is automatically attached to the engine with attach().

Parameters: engine (Engine) – the engine to which the metric must be attached
Return type: None

Note

engine.state.output is used to compute metric values. The majority of implemented metrics accept the following formats for engine.state.output: (y_pred, y) or {'y_pred': y_pred, 'y': y}. y_pred and y can be torch tensors or list of tensors/numbers if applicable.

Changed in version 0.4.5: y_pred and y can be torch tensors or list of tensors/numbers

load_state_dict(state_dict)[source]#

Method replaces internal state of the class with provided state dict data.

If there’s an active distributed configuration, the process uses its rank to pick the proper value from the list of values saved under each attribute’s name in the dict.

Parameters: state_dict (Mapping) – a dict containing attributes of the metric specified in its _state_dict_all_req_keys attribute.
Return type: None

abstract reset()[source]#

Resets the metric to its initial state.

By default, this is called at the start of each epoch.

Return type: None

started(engine)[source]#

Helper method to start data gathering for metric’s computation. It is automatically attached to the engine with attach().

Parameters: engine (Engine) – the engine to which the metric must be attached
Return type: None

state_dict()[source]#

Method returns state dict with attributes of the metric specified in its _state_dict_all_req_keys attribute. Can be used to save internal state of the class.

Return type: OrderedDict

abstract update(output)[source]#

Updates the metric’s state using the passed batch output.

By default, this is called once for each batch.

Parameters: output (Any) – the is the output from the engine’s process function.
Return type: None