add an interface to eval descriptors#1483
Conversation
Codecov Report
@@ Coverage Diff @@
## devel #1483 +/- ##
==========================================
+ Coverage 75.99% 76.03% +0.04%
==========================================
Files 93 93
Lines 7674 7729 +55
==========================================
+ Hits 5832 5877 +45
- Misses 1842 1852 +10
Continue to review full report at Codecov.
|
wanghan-iapcm
left a comment
wanghan-iapcm
left a comment
There was a problem hiding this comment.
I would recommend provide an separate method for inferring the descriptors, because strictly speaking, they are intermediate variable, not the output of a model. The eval method is provided to infer the output of a model.
| if self.auto_batch_size is not None: | ||
| return self.auto_batch_size.execute_all(self._eval_inner, numb_test, natoms, | ||
| coords, cells, atom_types, fparam = fparam, aparam = aparam, atomic = atomic, efield = efield) | ||
| return self._eval_inner(coords, cells, atom_types, fparam = fparam, aparam = aparam, atomic = atomic, efield = efield) | ||
| else : | ||
| if self.auto_batch_size is not None: | ||
| e, f, v = self.auto_batch_size.execute_all(self._eval_inner, numb_test, natoms, | ||
| coords, cells, atom_types, fparam = fparam, aparam = aparam, atomic = atomic, efield = efield) | ||
| else: | ||
| e, f, v = self._eval_inner(coords, cells, atom_types, fparam = fparam, aparam = aparam, atomic = atomic, efield = efield) | ||
| if self.modifier_type is not None: | ||
| me, mf, mv = self.dm.eval(coords, cells, atom_types) | ||
| e += me.reshape(e.shape) | ||
| f += mf.reshape(f.shape) | ||
| v += mv.reshape(v.shape) | ||
| return e, f, v |
There was a problem hiding this comment.
Any reason why auto_batch_size is removed?
There was a problem hiding this comment.
Not removed. Instead, it's moved into a wrapper eval_func to make the code more clear.
deepmd/infer/deep_pot.py
Outdated
| e += me.reshape(e.shape) | ||
| f += mf.reshape(f.shape) | ||
| v += mv.reshape(v.shape) |
There was a problem hiding this comment.
output is not updated with me, mf, mv before return.
There was a problem hiding this comment.
I don't think so. Here is a minimal example:
>>> import numpy as np
>>> output = (np.ones(3), np.ones(3), np.ones(3), np.ones(3), np.ones(3))
>>> e,f,v = output[:3]
>>> e += 1
>>> f += 2
>>> v += 3
>>> output
(array([2., 2., 2.]), array([3., 3., 3.]), array([4., 4., 4.]), array([1., 1., 1.]), array([1., 1., 1.]))See https://stackoverflow.com/a/35910888/9567349. In Python, a+=b is different from a=a+b. a=a+b creates a new object, but a+=b modifies the original object.
There was a problem hiding this comment.
by e,f,v = output[:3] e, f, v are copies of the list, not refs of the list element. You may do a test as
In [1]: a = [1,2,3]
In [2]: a[2]+=1
In [3]: a
Out[3]: [1, 2, 4]
In [4]: e,f,v=a[:3]
In [5]: e+=1
In [6]: f+=1
In [7]: v+=1
In [8]: a
Out[8]: [1, 2, 4]
In [9]: e, f, v
Out[9]: (2, 3, 5)There was a problem hiding this comment.
I can reproduce your example.
It seems that python does not copy the numpy arrays , but do copy numbers.
Adding modification in this way confuses the readers of the code, I suggest you changing it to
output[0] += me.reshape(e.shape)
...
deepmd/infer/deep_pot.py
Outdated
| e += me.reshape(e.shape) | ||
| f += mf.reshape(f.shape) | ||
| v += mv.reshape(v.shape) |
There was a problem hiding this comment.
I can reproduce your example.
It seems that python does not copy the numpy arrays , but do copy numbers.
Adding modification in this way confuses the readers of the code, I suggest you changing it to
output[0] += me.reshape(e.shape)
...
3 participants
Fix #1393.