auto convert input to v1 compatibility

Author: Han Wang

source/train/CMakeLists.txt

@@ -2,7 +2,7 @@
 
 configure_file("RunOptions.py.in" "${CMAKE_CURRENT_BINARY_DIR}/RunOptions.py" @ONLY)
 
-file(GLOB LIB_PY main.py common.py env.py Network.py Deep*.py Data.py DataSystem.py Model*.py Descrpt*.py Fitting.py Loss.py LearningRate.py Trainer.py TabInter.py ${CMAKE_CURRENT_BINARY_DIR}/RunOptions.py)
+file(GLOB LIB_PY main.py common.py env.py compat.py Network.py Deep*.py Data.py DataSystem.py Model*.py Descrpt*.py Fitting.py Loss.py LearningRate.py Trainer.py TabInter.py ${CMAKE_CURRENT_BINARY_DIR}/RunOptions.py)
 
 file(GLOB CLS_PY  Local.py Slurm.py)
 

source/train/compat.py

@@ -0,0 +1,138 @@
+import os,json,warnings
+from deepmd.common import j_have,j_must_have,j_must_have_d
+
+def convert_input_v0_v1(jdata, warning = True, dump = None) :
+    output = {}
+    if 'with_distrib' in jdata:
+        output['with_distrib'] = jdata['with_distrib']
+    if jdata['use_smooth'] :
+        output['model'] = _smth_model(jdata)
+    else:
+        output['model'] = _nonsmth_model(jdata)
+    output['learning_rate'] = _learning_rate(jdata)
+    output['loss'] = _loss(jdata)
+    output['training'] = _training(jdata)
+    _warnning_input_v0_v1(dump)
+    if dump is not None:
+       with open(dump, 'w') as fp:
+          json.dump(output, fp, indent=4)        
+    return output
+
+def _warnning_input_v0_v1(fname) :
+    msg = 'It seems that you are using a deepmd-kit input of version 0.x.x, which is deprecated. we have converted the input to >1.0.0 compatible'
+    if fname is not None:
+        msg += ', and output it to file ' + fname
+    warnings.warn(msg)
+
+def _nonsmth_model(jdata):
+    model = {}
+    model['descriptor'] = _nonsmth_descriptor(jdata)
+    model['fitting_net'] = _fitting_net(jdata)
+    return model
+
+def _smth_model(jdata):
+    model = {}
+    model['descriptor'] = _smth_descriptor(jdata)
+    model['fitting_net'] = _fitting_net(jdata)
+    return model
+
+def _nonsmth_descriptor(jdata) :
+    output = {}
+    seed = None
+    if j_have (jdata, 'seed') :
+        seed = jdata['seed']
+    # model
+    descriptor = {}
+    descriptor['type'] = 'loc_frame'
+    descriptor['sel_a'] = jdata['sel_a']
+    descriptor['sel_r'] = jdata['sel_r']
+    descriptor['rcut'] = jdata['rcut']
+    descriptor['axis_rule'] = jdata['axis_rule']
+    return descriptor
+
+def _smth_descriptor(jdata):
+    descriptor = {}
+    seed = None
+    if j_have (jdata, 'seed') :
+        seed = jdata['seed']
+    descriptor['type'] = 'se_a'
+    descriptor['sel'] = jdata['sel_a']
+    descriptor['rcut'] = jdata['rcut']
+    if j_have(jdata, 'rcut_smth') :
+        descriptor['rcut_r_smth'] = jdata['rcut_smth']
+    else :
+        descriptor['rcut_r_smth'] = descriptor['rcut']
+    descriptor['neuron'] = j_must_have (jdata, 'filter_neuron')
+    descriptor['axis_neuron'] = j_must_have_d (jdata, 'axis_neuron', ['n_axis_neuron'])
+    descriptor['resnet_dt'] = False
+    if j_have(jdata, 'resnet_dt') :
+        descriptor['resnet_dt'] = jdata['filter_resnet_dt']
+    if seed is not None:
+        descriptor['seed'] = seed
+    return descriptor
+
+def _fitting_net(jdata):
+    fitting_net = {}
+    seed = None
+    if j_have (jdata, 'seed') :
+        seed = jdata['seed']
+    fitting_net['neuron']= j_must_have_d (jdata, 'fitting_neuron', ['n_neuron'])
+    fitting_net['resnet_dt'] = True
+    if j_have(jdata, 'resnet_dt') :
+        fitting_net['resnet_dt'] = jdata['resnet_dt']
+    if j_have(jdata, 'fitting_resnet_dt') :
+        fitting_net['resnet_dt'] = jdata['fitting_resnet_dt']    
+    if seed is not None:
+        fitting_net['seed'] = seed
+    return fitting_net
+
+def _learning_rate(jdata):
+    # learning rate
+    learning_rate = {}
+    learning_rate['type'] = 'exp'
+    learning_rate['decay_steps'] = j_must_have(jdata, 'decay_steps')
+    learning_rate['decay_rate'] = j_must_have(jdata, 'decay_rate')
+    learning_rate['start_lr'] = j_must_have(jdata, 'start_lr')
+    return learning_rate
+
+def _loss(jdata):
+    # loss
+    loss = {}
+    loss['start_pref_e'] = j_must_have (jdata, 'start_pref_e')
+    loss['limit_pref_e'] = j_must_have (jdata, 'limit_pref_e')
+    loss['start_pref_f'] = j_must_have (jdata, 'start_pref_f')
+    loss['limit_pref_f'] = j_must_have (jdata, 'limit_pref_f')
+    loss['start_pref_v'] = j_must_have (jdata, 'start_pref_v')
+    loss['limit_pref_v'] = j_must_have (jdata, 'limit_pref_v')
+    if j_have(jdata, 'start_pref_ae') :
+        loss['start_pref_ae'] = jdata['start_pref_ae']
+    if j_have(jdata, 'limit_pref_ae') :
+        loss['limit_pref_ae'] = jdata['limit_pref_ae']
+    return loss
+
+def _training(jdata):
+    # training
+    training = {}
+    seed = None
+    if j_have (jdata, 'seed') :
+        seed = jdata['seed']
+    training['systems'] = jdata['systems']
+    training['set_prefix'] = jdata['set_prefix']
+    training['stop_batch'] = jdata['stop_batch']
+    training['batch_size'] = jdata['batch_size']
+    if seed is not None:
+        training['seed'] = seed
+    training['disp_file'] = "lcurve.out"
+    if j_have (jdata, "disp_file") : training['disp_file'] = jdata["disp_file"]
+    training['disp_freq'] = j_must_have (jdata, 'disp_freq')
+    training['numb_test'] = j_must_have (jdata, 'numb_test')
+    training['save_freq'] = j_must_have (jdata, 'save_freq')
+    training['save_ckpt'] = j_must_have (jdata, 'save_ckpt')
+    training['display_in_training'] = j_must_have (jdata, 'disp_training')
+    training['timing_in_training'] = j_must_have (jdata, 'time_training')
+    training['profiling'] = False
+    if j_have (jdata, 'profiling') :
+        training['profiling'] = jdata['profiling']
+        if training['profiling'] :
+            training['profiling_file'] = j_must_have (jdata, 'profiling_file')
+    return training

source/train/train.py

@@ -7,6 +7,7 @@
 import argparse
 import json
 from deepmd.env import tf
+from deepmd.compat import convert_input_v0_v1
 
 lib_path = os.path.dirname(os.path.realpath(__file__)) + "/../lib/"
 sys.path.append (lib_path)
@@ -54,6 +55,10 @@ def train (args) :
     # load json database
     fp = open (args.INPUT, 'r')
     jdata = json.load (fp)
+    if not 'model' in jdata.keys():
+       jdata = convert_input_v0_v1(jdata, 
+                                   warning = True, 
+                                   dump = 'input_v1_compat.json')
     # run options
     with_distrib = False 
     if 'with_distrib' in jdata: