Adios dataset name 319 #320
Conversation
examples/csce/train_gap.py
Outdated
| @@ -288,6 +288,7 @@ def __getitem__(self, idx): | |||
| data = generate_graphdata_from_smilestr( | |||
| smilestr, ytarget, csce_node_types, var_config | |||
| ) | |||
| data.dataset_name = 'csce' | |||
There was a problem hiding this comment.
@jychoi-hpc @allaffa Should the dataset_name be added this way manually, or should we read it from the json file?
|
|
||
| # Look for the dataset_name in any one of the Data samples and add it as an ADIOS attribute | ||
| dataset_name = self._get_dataset_name() | ||
| if dataset_name is not None: |
There was a problem hiding this comment.
@kshitij-v-mehta
Why do we need this if the dataset_name attributed is already take care of in the parent class AbstractBaseDataset?
There was a problem hiding this comment.
This is the ADIOS writer. It checks for the presence of the dataset_name in the object. If it exists, it adds it to the ADIOS file.
Similarly for the reader, if it finds dataset_name in the file, it reads it and adds it to the object. The get method of the parent class then reads this attribute.
|
Can this PR be merged? The check is failing due to a formatting issue in |
|
@pzhanggit This PR adds the dataset name to the ADIOS file. Does it also need to be added to the Pickle output? |
Thanks, Kshitij! I think so, as long as we continue to support Pickle. @allaffa what do you think? |
Never mind, I added it to the pickle output as well. |
* add multiple branches to Base model * adjust based on Max's new datasets with dataset_name * add json for compute grad energy physics-informed force prediction * revert time change * reverse prior change * Add check to ensure correct output dim for energy in physics-informed force prediction * Implement the check after setting var_config * revert change * Shorten comment --------- Co-authored-by: Rylie Weaver <[email protected]> Co-authored-by: Rylie Weaver <[email protected]>
* data attributes updated for consistency across datasets * non-normalized chemical composition added as data attribute * download_dataset.sh added for transition1x example * download dataset flag updated * scripts updated * development of tranistion1x scripts continues * transiton1x scripts completed * black formatting fixed * printouts removed * parallelizatin of data reading introduced * blsck formatting fixed * detach().clone() used to defined normalized energy per atom and black formatting * add compute_grad_energy=False as explicit argument * add data name as attributed to each data object * compute_grad_energy is parsed as input argument with default value set to False * edge_index, edge_attr, and edge_shifts explicitly itnroduced in the definition of the Data object * changed data.force into data.forces for ani1x and qm7x examples * smiles_string added as data attribute * remove redundant logic on energy normalization from omat24 example * force threshold value increased to 1000 for ani-1x * Reverted smiles_utils.py to version from commit 3c3c434 * xyz2mol functionalities put in a separate file * download dataset script added for qm7x * renamed data.force as data.forces in ani1x * natoms converted into a tensor * verbosity level ntroduced for ani1x * Z corrected into atomic_numbers for qm7x example * bug fixed for data attributes in transition1x * try-except in transition1x rescoped * transform coordinates fixed in transition1x * iterate_tqdm used in utils.create_graph_data for transition1x example * total_enerfgy replaced with energy --------- Co-authored-by: Massimiliano Lupo Pasini <[email protected]> Co-authored-by: Massimiliano Lupo Pasini <[email protected]>
* Add multidataset example with deepspeed support * Change base.json to follow GPS' requirement
…tional transforms accounting for pbc
…e and shape for each attribute, and appropriate positional transform
Draft Fix for PBC Examples
Update Feb 03: PR ready for merging.
PR for review only. Not ready for merging yet.