Yusuke Oda

@tetsuok @neubig At least we need to fix a version of libraries we rely on. Setting deprecation warnings is a common process to remove a feature from the library, but...

`>=` indicates that we require the newest version (unless other restrictions, such as `

According to the SciPy's log (https://github.com/scipy/scipy/pull/16389) it attempts to introduce breaking changes in the minor version (`1.11.0`). So we need to fix minor versions too for this library (`~=1.N.0`)

@neubig The `~=N.x` specifier (fixing the major version, auto-update minor/patch versions) does never cause any breakages on libraries that follows semantic versioning. Others are case-by-case (no silver-bullet solution): we have...

The best solution here is to choose a version specifier that won't involve breaking changes. Fixing the major version is enough for semvar-based libraries. We realized that the minor version...

@shuaichenchang How about: 1. withdraw this pull request 2. `git checkout ` then `git checkout -b ` 3. propose a new pull request on the new tag

The cause of randomness may be that we didn't fix the random seed. https://github.com/neulab/ExplainaBoard/blob/9ba998c125587bb5a9b421706b1ca50028af220b/explainaboard/metrics/metric.py#L571 I think a possible fix is: 1. Provide a way to fix the seed from outside...

NumPy provides `SeedSequence` functionality which allows us to hierarchically initialize the random seeds over the whole library. I think it would be better that every class that calculates something has...

Plan to apply the change: 1. Implement `seed` argument in `Metric.__init__`, which is defaulting to `None`. If `seed` is None, it is initialized randomly. 2. Implement `Metric.get_seed` to return the...

I think it is enough to provide a functionality to retrieve a list of features corresponding to given analysis level name. ```python class MyProcessor(Processor): def get_feature_set(name: str) -> List[Feature]: if...