Integrate get_molnet_dataset with Splitter #209
Conversation
mottodora
changed the title
Codecov Report
@@ Coverage Diff @@
## master #209 +/- ##
==========================================
- Coverage 82.32% 81.59% -0.74%
==========================================
Files 106 106
Lines 4974 5026 +52
==========================================
+ Hits 4095 4101 +6
- Misses 879 925 +46 |
| frac_test=.1, seed=777, return_smiles=False, | ||
| target_index=None): | ||
| target_index=None, task_index=0, **kwargs): |
There was a problem hiding this comment.
Can you add docstring? or want to do in other PR?
| if isinstance(split, str): | ||
| splitter = split_method_dict[split]() | ||
| elif isinstance(split, BaseSplitter): | ||
| splitter = split |
There was a problem hiding this comment.
I feel adding else: raise TypeError('message') is friendly.
| if isinstance(splitter, ScaffoldSplitter): | ||
| get_smiles = True | ||
| else: | ||
| get_smiles = return_smiles |
There was a problem hiding this comment.
I feel adding else: raise TypeError('message') is friendly.
There was a problem hiding this comment.
I think TypeError is not necessary for this line.
There was a problem hiding this comment.
I want to fix the message to show the "actual" type.
| @@ -7,8 +7,9 @@ | |||
| from chainer_chemistry.datasets import NumpyTupleDataset | |||
| from chainer_chemistry.datasets import molnet | |||
|
|
|||
| expect_bbbp_lengths = [1631, 203, 205] | |||
| expect_bbbp_lengths = [1633, 203, 203] | |||
There was a problem hiding this comment.
Do you know why this happen? Rdkit version??
If rdkit version issue, I want to fix recommended rdkit version for each chainer chemistry version.
There was a problem hiding this comment.
It seems that this change is caused by the difference between scaffold split and random split. But unexpected.
There was a problem hiding this comment.
I think this is not "unexpected" since you implement that StratifiedSplitter to have much "train" data for the remainder, right?
|
Fix. |
| frac_test=.1, seed=777, return_smiles=False, | ||
| target_index=None): | ||
| target_index=None, task_index=0, **kwargs): |
| if isinstance(splitter, ScaffoldSplitter): | ||
| get_smiles = True | ||
| else: | ||
| get_smiles = return_smiles |
There was a problem hiding this comment.
I want to fix the message to show the "actual" type.
| @@ -7,8 +7,9 @@ | |||
| from chainer_chemistry.datasets import NumpyTupleDataset | |||
| from chainer_chemistry.datasets import molnet | |||
|
|
|||
| expect_bbbp_lengths = [1631, 203, 205] | |||
| expect_bbbp_lengths = [1633, 203, 203] | |||
There was a problem hiding this comment.
I think this is not "unexpected" since you implement that StratifiedSplitter to have much "train" data for the remainder, right?
|
I want to fix some document and message, but I think it is faster to make PR instead of asking you to fix. |
Please merge after #201 #202