fix(tf): fix bugs in tensor training and migrate to reformat data

deepmd/tf/fit/polar.py

@@ -202,7 +202,9 @@
 
                         polar_bias.append(
                             np.sum(
-                                all_stat["atomic_polarizability"][ss][:, index_lis, :]
+                                all_stat["atomic_polarizability"][ss].reshape(
+                                    nframes, len(atom_has_polar), -1
+                                )[:, index_lis, :]
                                 / nframes,
                                 axis=(0, 1),
                             ).reshape((1, 9))

deepmd/tf/model/tensor.py

@@ -86,7 +86,7 @@
         all_stat = make_stat_input(data, self.data_stat_nbatch, merge_sys=False)
         m_all_stat = merge_sys_stat(all_stat)
         self._compute_input_stat(m_all_stat, protection=self.data_stat_protect)
-        self._compute_output_stat(all_stat)
+        self._compute_output_stat(m_all_stat)
 
     def _compute_input_stat(self, all_stat, protection=1e-2):
         self.descrpt.compute_input_stats(

deepmd/utils/data.py

@@ -643,6 +643,7 @@
                                 pass
                             else:
                                 sel_mask = np.isin(self.atom_type, type_sel)
+                                data = data.reshape([nframes, natoms, ndof_])
                                 data = data[:, sel_mask]
                                 natoms = natoms_sel
                                 idx_map = idx_map_sel
@@ -669,7 +670,7 @@
         elif must:
             raise RuntimeError("%s not found!" % path)
         else:
-            if type_sel is not None and not output_natoms_for_type_sel:
+            if atomic and type_sel is not None and not output_natoms_for_type_sel:
                 ndof = ndof_ * natoms_sel
             data = np.full([nframes, ndof], default, dtype=dtype)
             if repeat != 1:

examples/water_tensor/dipole/dipole_input.json

@@ -58,16 +58,16 @@
   "training": {
     "training_data": {
       "systems": [
-        "./training_data/atomic_system",
-        "./training_data/global_system"
+        "./training_data_reformat/atomic_system",
+        "./training_data_reformat/global_system"
       ],
       "batch_size": "auto",
       "_comment8": "that's all"
     },
     "validation_data": {
       "systems": [
-        "./validation_data/atomic_system",
-        "./validation_data/global_system"
+        "./validation_data_reformat/atomic_system",
+        "./validation_data_reformat/global_system"
       ],
       "batch_size": 1,
       "numb_btch": 3,

examples/water_tensor/polar/polar_input.json

@@ -63,16 +63,16 @@
   "training": {
     "training_data": {
       "systems": [
-        "./training_data/atomic_system",
-        "./training_data/global_system"
+        "./training_data_reformat/atomic_system",
+        "./training_data_reformat/global_system"
       ],
       "batch_size": "auto",
       "_comment8": "that's all"
     },
     "validation_data": {
       "systems": [
-        "./validation_data/atomic_system",
-        "./validation_data/global_system"
+        "./validation_data_reformat/atomic_system",
+        "./validation_data_reformat/global_system"
       ],
       "batch_size": 1,
       "numb_btch": 3,

source/tests/tf/test_polar_se_a.py

@@ -1,4 +1,8 @@
 # SPDX-License-Identifier: LGPL-3.0-or-later
+from pathlib import (
+    Path,
+)
+
 import numpy as np
 
 from deepmd.tf.common import (
@@ -16,6 +20,9 @@
 from deepmd.tf.model import (
     PolarModel,
 )
+from deepmd.utils.data_system import (
+    DeepmdDataSystem,
+)
 
 from .common import (
     DataSystem,
@@ -198,3 +205,52 @@ def test_model(self):
         # make sure atomic virial sum to virial
         places = 10
         np.testing.assert_almost_equal(pv, spv, places)
+
+    def test_data_stat(self):
+        jfile = "polar_se_a.json"
+        jdata = j_loader(jfile)
+
+        systems = [
+            str(
+                Path(__file__).parent.parent
+                / "pt"
+                / "water_tensor"
+                / "polar"
+                / "global_system"
+            ),
+            str(
+                Path(__file__).parent.parent
+                / "pt"
+                / "water_tensor"
+                / "polar"
+                / "atomic_system"
+            ),
+        ]
+
+        batch_size = 1
+        test_size = 1
+        rcut = j_must_have(jdata["model"]["descriptor"], "rcut")
+
+        data = DeepmdDataSystem(systems, batch_size, test_size, rcut)
+        data.add(
+            "atomic_polarizability",
+            9,
+            atomic=True,
+            type_sel=jdata["model"]["fitting_net"]["sel_type"],
+        )
+        data.add(
+            "polarizability",
+            9,
+            atomic=False,
+        )
+
+        jdata["model"]["descriptor"].pop("type", None)
+        jdata["model"]["fitting_net"].pop("type", None)
+        descrpt = DescrptSeA(**jdata["model"]["descriptor"], uniform_seed=True)
+        jdata["model"]["fitting_net"]["ntypes"] = descrpt.get_ntypes()
+        jdata["model"]["fitting_net"]["dim_descrpt"] = descrpt.get_dim_out()
+        jdata["model"]["fitting_net"]["embedding_width"] = descrpt.get_dim_rot_mat_1()
+        fitting = PolarFittingSeA(**jdata["model"]["fitting_net"], uniform_seed=True)
+        model = PolarModel(descrpt, fitting)
+
+        model.data_stat(data)