Kenshin Abe

## Motivation I simplified the implementation of `IntersectionSearchSpace.calculate` a little. ## Description of the changes Directly filter the search space dictionary instead of creating a delete list and deleting them...

Gather mask of None parameter in TPESampler

8

## Motivation Previously, in TPESampler, the mask of removing None parameters (= parameters that do not appear in the trial) was taken in two different places, which can easily cause...

Unify optimization history tests

4

## Motivation The main logic of `plot_optimization_history` has been unified between plotly and matplotlib by #3736. This PR unifies the parts of tests that exist in both plotly and matplotlib....

Improve constrained TPE

1

### Motivation The constraints option was supported in `TPESampler` by #3506. There might be room for improvement in how to define weights of infeasible trials for `_ParzenEstimator`. ### Description Copy-paste...

Add recommended setting for batch optimization

10

### What is an issue? There are several recommended settings for efficient batch optimization, such as the followings. - set `popsize` to a multiple of batch size in `CmaEsSampler` (supported...

`constant_liar=True` does not work in multi-objective `TPESampler`

1

### What is an issue? In multi-objective `TPESampler`, running trials are skipped even when `constant_liar=True`. I think we should at least mention this in the document and should probably give...

- GINSparse - RelGCNSparse - GIN + coo sparse_matmul

- cora - citeseer - reddit

Weight tying option has a bug

1

If `weight_tying` option is `True`, only one update layer is called regardless of `n_update_layers` https://github.com/pfnet-research/chainer-chemistry/blob/master/chainer_chemistry/models/gin.py#L37

Add issue template for code fix

4

## Motivation Currently, we don't have an appropriate issue template for a code fix. This PR adds a code-fix issue template. Preview:   ## Description of the changes I...