Torchmetrics

Fixed

- Fixed evaluation of `R2Score` with near constant target ([1576](https://github.com/Lightning-AI/torchmetrics/pull/1576))
- Fixed dtype conversion when metric is submodule ([1583](https://github.com/Lightning-AI/torchmetrics/pull/1583))
- Fixed bug related to `top_k>1` and `ignore_index!=None` in `StatScores` based metrics ([1589](https://github.com/Lightning-AI/torchmetrics/pull/1589))
- Fixed corner case for `PearsonCorrCoef` when running in ddp mode but only on single device ([1587](https://github.com/Lightning-AI/torchmetrics/pull/1587))
- Fixed overflow error for specific cases in `MAP` when big areas are calculated ([1607](https://github.com/Lightning-AI/torchmetrics/pull/1607))

Fixed

- Fixed classification metrics for `byte` input ([1521](https://github.com/Lightning-AI/torchmetrics/pull/1474))
- Fixed the use of `ignore_index` in `MulticlassJaccardIndex` ([1386](https://github.com/Lightning-AI/torchmetrics/pull/1386))

Fixed

- Fixed compatibility between XLA in `_bincount` function ([1471](https://github.com/Lightning-AI/torchmetrics/pull/1471))
- Fixed type hints in methods belonging to `MetricTracker` wrapper ([1472](https://github.com/Lightning-AI/torchmetrics/pull/1472))
- Fixed `multilabel` in `ExactMatch` ([1474](https://github.com/Lightning-AI/torchmetrics/pull/1474))

Fixed

- Fixed type checking on the `maximize` parameter at the initialization of `MetricTracker` ([1428](https://github.com/Lightning-AI/torchmetrics/pull/1430))
- Fixed mixed precision autocast for `SSIM` metric ([1454](https://github.com/Lightning-AI/torchmetrics/pull/1454))
- Fixed checking for `nltk.punkt` in `RougeScore` if a machine is not online ([1456](https://github.com/Lightning-AI/torchmetrics/pull/1456))
- Fixed wrongly reset method in `MultioutputWrapper` ([1460](https://github.com/Lightning-AI/torchmetrics/pull/1460))
- Fixed dtype checking in `PrecisionRecallCurve` for `target` tensor ([1457](https://github.com/Lightning-AI/torchmetrics/pull/1457))

Added

- Added `MulticlassExactMatch` to classification metrics ([1343](https://github.com/Lightning-AI/torchmetrics/pull/1343))
- Added `TotalVariation` to image package ([978](https://github.com/Lightning-AI/torchmetrics/pull/978))
- Added `CLIPScore` to new multimodal package ([1314](https://github.com/Lightning-AI/torchmetrics/pull/1314))
- Added regression metrics:
* `KendallRankCorrCoef` ([1271](https://github.com/Lightning-AI/torchmetrics/pull/1271))
* `LogCoshError` ([1316](https://github.com/Lightning-AI/torchmetrics/pull/1316))
- Added new nominal metrics:
* `CramersV` ([1298](https://github.com/Lightning-AI/torchmetrics/pull/1298))
* `PearsonsContingencyCoefficient` ([1334](https://github.com/Lightning-AI/torchmetrics/pull/1334))
* `TschuprowsT` ([1334](https://github.com/Lightning-AI/torchmetrics/pull/1334))
* `TheilsU` ([1337](https://github.com/Lightning-AI/torchmetrics/pull/1334))
- Added option to pass `distributed_available_fn` to metrics to allow checks for custom communication backend for making `dist_sync_fn` actually useful ([1301](https://github.com/Lightning-AI/torchmetrics/pull/1301))
- Added `normalize` argument to `Inception`, `FID`, `KID` metrics ([1246](https://github.com/Lightning-AI/torchmetrics/pull/1246))

Changed

- Changed minimum Pytorch version to be 1.8 ([1263](https://github.com/Lightning-AI/torchmetrics/pull/1263))
- Changed interface for all functional and modular classification metrics after refactor ([1252](https://github.com/Lightning-AI/torchmetrics/pull/1252))

Removed

- Removed deprecated `BinnedAveragePrecision`, `BinnedPrecisionRecallCurve`, `RecallAtFixedPrecision` ([1251](https://github.com/Lightning-AI/torchmetrics/pull/1251))
- Removed deprecated `LabelRankingAveragePrecision`, `LabelRankingLoss` and `CoverageError` ([1251](https://github.com/Lightning-AI/torchmetrics/pull/1251))
- Removed deprecated `KLDivergence` and `AUC` ([1251](https://github.com/Lightning-AI/torchmetrics/pull/1251))

Fixed

- Fixed precision bug in `pairwise_euclidean_distance` ([1352](https://github.com/Lightning-AI/torchmetrics/pull/1352))

---

Fixed

- Fixed bug in `Metrictracker.best_metric` when `return_step=False` ([1306](https://github.com/Lightning-AI/torchmetrics/pull/1306))
- Fixed bug to prevent users from going into an infinite loop if trying to iterate of a single metric ([1320](https://github.com/Lightning-AI/torchmetrics/pull/1320))