|
1 | 1 | # Changelog
|
2 | 2 |
|
| 3 | +## v0.0.1-alpha.1 |
| 4 | + |
| 5 | +Initial PyPi Release |
| 6 | + |
| 7 | +Current list of modules: |
| 8 | + |
| 9 | +- `clu.checkpoint` |
| 10 | +- `clu.deterministic_training` |
| 11 | +- `clu.metric_writers` |
| 12 | +- `clu.periodic_actions` |
| 13 | +- `clu.platform` |
| 14 | +- `clu.profiler` |
| 15 | + |
| 16 | +## v0.0.1-alpha.2 |
| 17 | + |
| 18 | +- Adds `metrics` module and some minor changes. |
| 19 | + |
3 | 20 | ## v0.0.1a3
|
4 | 21 |
|
5 | 22 | - Added `metric_writers.TorchTensorboardWriter`
|
| 23 | + |
| 24 | +## v0.0.2 |
| 25 | + |
| 26 | +- Added preprocess_spec. |
| 27 | +- Improvements to periodic_actions. |
| 28 | + |
| 29 | +## v0.0.3 |
| 30 | + |
| 31 | +- `metric_writers`: Lets `SummaryWriter` write nested dictionaries. |
| 32 | +- `internal`: Adds `async.Pool`. |
| 33 | +- `preprocess_spec`: Support nested dictionaries. |
| 34 | +- `profile`: Use JAX profiler APIs instead of TF profiler APIs. |
| 35 | + |
| 36 | +## v0.0.4 |
| 37 | + |
| 38 | +`deterministic_data` |
| 39 | + |
| 40 | +- Support non-positive input value for pad_up_to_batches. |
| 41 | +- Support padding dataset when data dimension is unknown. |
| 42 | +- Support TFDS specs in get_read_instruction_for_host. |
| 43 | +- Allow changing drop_remainder for batching. |
| 44 | +- Add RemainderOptions in deterministic_data. |
| 45 | + |
| 46 | +`metric_writers` |
| 47 | + |
| 48 | +- Support multiple writers in metric_writers.ensure_flushes. |
| 49 | + |
| 50 | +`metrics` |
| 51 | + |
| 52 | +- Makes internal.flatten_dict() work with ConfigDicts. |
| 53 | +- Forwards mask model output to metrics created via `Metric.from_output()`. |
| 54 | +- Forwards mask model output to metrics created via `Metric.from_fun()`. |
| 55 | +- Added `Collections.unreplicate()`, `Collections.create()`. |
| 56 | + |
| 57 | +`periodic_actions` |
| 58 | + |
| 59 | +- Formats long time strings in '{days}d{hours}h{mins}m' format. |
| 60 | + |
| 61 | +`preprocess_spec` |
| 62 | + |
| 63 | +- Make feature description of features in PreprocessFn more compact. |
| 64 | +- Better type check in `preprocess_spec.get_all_ops()`. |
| 65 | + |
| 66 | +Documentation: |
| 67 | + |
| 68 | +- Added `clu_synopsis.ipynb` Colab |
0 commit comments