✨ Add Adaptive Calibration plots

[Copilot]

This PR implements the missing plot() method for AdaptiveCalibrationError that was previously raising NotImplementedError.

Problem

The AdaptiveCalibrationError.plot() method was not implemented and would raise NotImplementedError when called:

from torch_uncertainty.metrics import AdaptiveCalibrationError
import torch

ace = AdaptiveCalibrationError(task='binary', num_bins=5, norm='l1')
ace.update(torch.tensor([0.1, 0.5, 0.9]), torch.tensor([0, 1, 1]))
ace.plot()  # NotImplementedError

Solution

Added plot() methods to both BinaryAdaptiveCalibrationError and MulticlassAdaptiveCalibrationError classes that:

• Use adaptive binning (_equal_binning_bucketize) for the actual calibration computation
• Generate reliability diagrams showing the relationship between predicted confidence and actual accuracy
• Include confidence distribution histograms
• Return the same _PLOT_OUT_TYPE (Figure, Axes) as other calibration metrics
• Are consistent with the existing CalibrationError.plot() interface

Usage

from torch_uncertainty.metrics import AdaptiveCalibrationError
import torch

# Binary classification
ace = AdaptiveCalibrationError(task='binary', num_bins=5, norm='l1')
ace.update(torch.tensor([0.1, 0.5, 0.9]), torch.tensor([0, 1, 1]))
fig, ax = ace.plot()  # Now works!

# Multiclass classification  
ace_mc = AdaptiveCalibrationError(task='multiclass', num_classes=3, num_bins=4, norm='l1')
ace_mc.update(probs, targets)
fig, ax = ace_mc.plot()  # Also works!

Testing

Added comprehensive tests for both binary and multiclass adaptive calibration plotting that verify:

• Plot generation works without errors
• Correct figure and axes types are returned
• Proper axis labels are set
• Integration with existing test suite

Fixes #214.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.