Published April 7, 2021
| Version 0.2.5
Software
Open
e3nn/e3nn:
Creators
- 1. @atomicarchitects
- 2. University of Amsterdam
- 3. MIR group @ Harvard University
- 4. University of Michigan
- 5. Harvard
- 6. Technical University of Denmark (DTU)
- 7. Redwood Center for Theoretical Neuroscience
Description
Added
- Added
e3nn.set_optimization_defaults()
ande3nn.get_optimization_defaults()
- Constructors for empty
Irreps
:Irreps()
andIrreps("")
- Additional tests, docs, and refactoring for
Irrep
andIrreps
. - Added
TensorProduct.weight_views()
andTensorProduct.weight_view_for_instruction()
- Fix Docs for ExtractIr
- Renamed
o3.TensorProduct
arguments inirreps_in1
,irreps_in2
andirreps_out
- Renamed
o3.spherical_harmonics
arguementxyz
intox
- Renamed
math.soft_one_hot_linspace
argumentendpoint
intocutoff
,cutoff = not endpoint
- Variances are now provided to
o3.TensorProduct
through explicitin1_var
,in2_var
,out_var
parameters - Submodules define
__all__
; documentation uses shorter module names for the classes/methods.
- Enabling/disabling einsum optimization no longer affects PyTorch RNG state.
- Variances can no longer be provided to
o3.TensorProduct
in the list-of-tuple format forirreps_in1
, etc.
Files
e3nn/e3nn-0.2.5.zip
Files
(707.6 kB)
Name | Size | Download all |
---|---|---|
md5:7cc871a3ac8adb9d0e672c8ac59ee20f
|
707.6 kB | Preview Download |
Additional details
Related works
- Is supplement to
- https://github.com/e3nn/e3nn/tree/0.2.5 (URL)