add Trinity model and debug the lem #256
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
QG-phy
reviewed
Jun 1, 2025
|
Kindly suggest the flag 'init2b' change to 'only2b'. |
QG-phy
reviewed
Jun 1, 2025
QG-phy
reviewed
Jun 1, 2025
QG-phy
reviewed
Jun 2, 2025
QG-phy
reviewed
Jun 5, 2025
QG-phy
reviewed
Jun 5, 2025
|
|
不需要了，我comment下 |
QG-phy
reviewed
Jun 5, 2025
dptb/nnops/loss.py
Outdated
| ref_data[AtomicDataDict.NODE_FEATURES_KEY] = ref_data[AtomicDataDict.NODE_FEATURES_KEY] + mu * ref_data[AtomicDataDict.NODE_OVERLAP_KEY] | ||
| ref_data[AtomicDataDict.EDGE_FEATURES_KEY] = ref_data[AtomicDataDict.EDGE_FEATURES_KEY] + mu * ref_data[AtomicDataDict.EDGE_OVERLAP_KEY] | ||
| elif batch.max() >= 1: | ||
| slices = [data["__slices__"]["pos"][i]-data["__slices__"]["pos"][i-1] for i in range(1,len(data["__slices__"]["pos"]))] | ||
| slices = [0] + slices | ||
| ndiag_batch = torch.stack([i.sum() for i in self.idp.mask_to_ndiag[data[AtomicDataDict.ATOM_TYPE_KEY].flatten()].split(slices)]) | ||
| ndiag_batch = torch.stack([i.shape[0] for i in self.idp.mask_to_ndiag[data[AtomicDataDict.ATOM_TYPE_KEY].flatten()].split(slices)]) |
There was a problem hiding this comment.
这里要改为torch.tensor(i.shape[0]) 不然会报错。因为torch.stack要对tensor操作不能是int 之前i.sum() 没问题。改成i.shape[0]就会报错了。
…lements Initialize full_mask_to_diag tensor to track diagonal orbital pairs in the reduced matrix. This helps in identifying diagonal elements during further processing.
…alculation The shift_mu function was extracted to avoid code duplication across multiple loss classes. The onsite shift calculation was simplified by using a more accurate formula that accounts for both node and edge features.
QG-phy
reviewed
Jun 10, 2025
| for orbs, islice in self.orbpair_maps.items(): | ||
| fio, fjo = orbs.split('-') | ||
| if fio == fjo: | ||
| self.full_mask_to_diag[islice] = True | ||
|
|
There was a problem hiding this comment.
这一块我新加了一个属性。获取在full basis 的feature 中轨道相同的那些指标。其实就是feature to block 里需要*0.5的那部分！
QG-phy
reviewed
Jun 10, 2025
| norm_ss_e = (ref_data[AtomicDataDict.EDGE_OVERLAP_KEY] * ref_data[AtomicDataDict.EDGE_OVERLAP_KEY]).sum(dim=-1) | ||
| norm_ss_e_diag = (ref_data[AtomicDataDict.EDGE_OVERLAP_KEY][:,idp.full_mask_to_diag] * ref_data[AtomicDataDict.EDGE_OVERLAP_KEY][:,idp.full_mask_to_diag]).sum(dim=-1) | ||
|
|
||
| return mu_n, mu_e, mu_e_diag, norm_ss_n, norm_ss_e, norm_ss_e_diag |
QG-phy
reviewed
Jun 10, 2025
|
|
||
| mu = mu_n + 2 * mu_e - mu_e_diag | ||
| ss = norm_ss_n + 2 * norm_ss_e - norm_ss_e_diag | ||
| mu = mu / ss |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This pull request introduces a new
ScalarMLPFunctionmodule for modular MLP functionality, integrates theTrinityembedding method, and refactors existing code to enhance modularity and maintainability. Key changes include the addition of new functionality for two-body predictions in thee3tbmethod, improved handling of onsite shifts in loss computation, and cleanup of duplicate code.New Features and Modules:
Added
ScalarMLPFunctiontodptb/nn/base.py, which implements a modular MLP with configurable dimensions, nonlinearity, initialization, and optional dropout and batch normalization. This replaces duplicate implementations in other files. [1] [2] [3]Introduced the
Trinityembedding method indptb/nn/embedding/__init__.pyand integrated it into thee3tbprediction workflow indptb/nn/deeptb.py. This enables additional embedding options for specific use cases. [1] [2] [3]Enhancements to
e3tbPredictions:e3tbby introducingedge_prediction_h2andh2miltonianmodules indptb/nn/deeptb.py. These handle additional edge attributes and Hamiltonian calculations. [1] [2] [3]Refactoring and Code Cleanup:
Removed duplicate
ScalarMLPFunctionimplementations fromdptb/nn/embedding/lem.pyanddptb/nn/embedding/slem.py, replacing them with the centralized implementation indptb/nn/base.py. This reduces redundancy and simplifies maintenance. [1] [2]Updated
lem.pyto use the newScalarMLPFunctionand fixed dimension mismatches in latent and output scalars. [1] [2]Loss Computation Improvements:
dptb/nnops/loss.pyto account for overlap weights and avoid overflow issues. This ensures more accurate loss adjustment for batch data.