Float64 conversion #83
Conversation
openqdc/datasets/base.py
Outdated
| from torch_geometric.data import Data | ||
| from torch_geometric.loader import DataLoader | ||
|
|
||
| return DataLoader( |
There was a problem hiding this comment.
Add support for num_workers, pin_memory and additional data loader arguments that can be passed (similar to how pyg dataloader does it.
openqdc/datasets/base.py
Outdated
| @@ -546,6 +546,48 @@ def wrapper(idx): | |||
| datum["idxs"] = idxs | |||
| return datum | |||
|
|
|||
| @classmethod | |||
There was a problem hiding this comment.
Why class method? Why not a normal method and remove the additional arguments. Also, say I split my dataset in train and val dataset, after which I want to create dataloaders? Will this work?
Can we address this in a different PR?
There was a problem hiding this comment.
Also, we should test this if we are adding this feature.
There was a problem hiding this comment.
Ops, it was not supposed to be inside the PR. It was one of my custom piece of code that i use for utilities
| return { | ||
| "atomic_inputs": np.float32, | ||
| "position_idx_range": np.int32, | ||
| "energies": np.float32, |
There was a problem hiding this comment.
Shouldn't it be np.float64?
There was a problem hiding this comment.
Interaction energies don't use total energies so the energy values is fine at float32
openqdc/datasets/potential/dummy.py
Outdated
| return { | ||
| "atomic_inputs": np.float32, | ||
| "position_idx_range": np.int32, | ||
| "energies": np.float32, |
There was a problem hiding this comment.
As this is a potential dataset, shouldn't we do float64.
There was a problem hiding this comment.
This dataset is used for static testing and we should regenerate the data to be float64 in case. As the accuracy is not actually a real issue here I find that just overriding the energy type is good enough
There was a problem hiding this comment.
Yup, can we regenerate to follow the same format.
There was a problem hiding this comment.
Regerated to float64
Conversion to float64 for Potential Dataset