Merge branch 'master' into lb_fix_yaml

domainlab/algos/trainers/a_trainer.py

@@ -37,7 +37,6 @@ def __init__(self, successor_node=None):
         #
         self.loader_tr = None
         self.loader_te = None
-        self.dict_loader_tr = None
         self.num_batches = None
         self.flag_update_hyper_per_epoch = None
         self.flag_update_hyper_per_batch = None
@@ -67,7 +66,6 @@ def init_business(self, model, task, observer, device, aconf, flag_accept=True):
         #
         self.loader_tr = task.loader_tr
         self.loader_te = task.loader_te
-        self.dict_loader_tr = task.dict_loader_tr
 
         if flag_accept:
             self.observer.accept(self)

domainlab/algos/trainers/zoo_trainer.py

@@ -5,11 +5,9 @@
 from domainlab.algos.trainers.train_dial import TrainerDIAL
 from domainlab.algos.trainers.train_matchdg import TrainerMatchDG
 from domainlab.algos.trainers.train_mldg import TrainerMLDG
-from domainlab.algos.trainers.train_fishr import TrainerFishr
 from domainlab.algos.trainers.train_hyper_scheduler import TrainerHyperScheduler
 
 
-
 class TrainerChainNodeGetter(object):
     """
     Chain of Responsibility: node is named in pattern Trainer[XXX] where the string
@@ -40,7 +38,6 @@ def __call__(self, lst_candidates=None, default=None, lst_excludes=None):
         chain = TrainerDIAL(chain)
         chain = TrainerMatchDG(chain)
         chain = TrainerMLDG(chain)
-        chain = TrainerFishr(chain)
         chain = TrainerHyperScheduler(chain)  # FIXME: change to warmup
         node = chain.handle(self.request)
         return node

domainlab/compos/exp/exp_utils.py

@@ -63,6 +63,9 @@ def mk_model_na(self, tag=None, dd_cut=19):
         model_name = "_".join(list4mname)
         if self.host.args.debug:
             model_name = "debug_" + model_name
+        slurm = os.environ.get('SLURM_JOB_ID')
+        if slurm:
+            model_name = model_name + '_' + slurm
         logger = Logger.get_logger()
         logger.info(f"model name: {model_name}")
         return model_name

domainlab/models/a_model.py

@@ -6,13 +6,6 @@
 
 from torch import nn
 
-try:
-    from backpack import extend
-except:
-    backpack = None
-
-
-
 
 class AModel(nn.Module, metaclass=abc.ABCMeta):
     """
@@ -65,4 +58,3 @@ def forward(self, tensor_x, tensor_y, tensor_d, others=None):
         :param d:
         """
         return self.cal_loss(tensor_x, tensor_y, tensor_d, others)
-

domainlab/models/a_model_classif.py

@@ -10,24 +10,13 @@
 from torch import nn as nn
 from torch.nn import functional as F
 
-
-try:
-    from backpack import backpack, extend
-    from backpack.extensions import BatchGrad, Variance
-except:
-    backpack = None
-
-
-
 from domainlab.models.a_model import AModel
 from domainlab.utils.utils_class import store_args
 from domainlab.utils.utils_classif import get_label_na, logit2preds_vpic
 from domainlab.utils.perf import PerfClassif
 from domainlab.utils.perf_metrics import PerfMetricClassif
 from domainlab.utils.logger import Logger
 
-loss_cross_entropy_extended = extend(nn.CrossEntropyLoss())
-
 
 class AModelClassif(AModel, metaclass=abc.ABCMeta):
     """
@@ -124,7 +113,7 @@ def cal_task_loss(self, tensor_x, tensor_y):
             y_target = tensor_y
         else:
             _, y_target = tensor_y.max(dim=1)
-        lc_y = loss_cross_entropy_extended(logit_y, y_target)
+        lc_y = F.cross_entropy(logit_y, y_target, reduction="none")
         # cross entropy always return a scalar, no need for inside instance reduction
         return lc_y
 

domainlab/models/model_deep_all.py

@@ -1,10 +1,3 @@
-
-try:
-    from backpack import extend
-except:
-    backpack = None
-
-
 from domainlab.models.a_model_classif import AModelClassif
 from domainlab.utils.override_interface import override_interface
 
@@ -59,9 +52,4 @@ def cal_loss(self, tensor_x, tensor_y, tensor_d, others=None):
             lc_y = self.cal_task_loss(tensor_x, tensor_y)
             return lc_y
 
-        def convert4backpack(self):
-            """
-            convert the module to backpack for 2nd order gradients
-            """
-            self.net = extend(self.net, use_converter=True)
     return ModelDeepAll

domainlab/tasks/a_task.py

@@ -24,7 +24,6 @@ def __init__(self, succ=None):
         self._args = None
         self.dict_dset_all = {}  # persist
         self.dict_dset_tr = {}  # versatile variable: which domains to use as training
-        self.dict_loader_tr = {}
         self.dict_dset_te = {}  # versatile
         self.dict_dset_val = {}  # versatile
         self.dict_domain_class_count = {}

domainlab/tasks/b_task.py

@@ -40,7 +40,6 @@ def init_business(self, args, node_algo_builder=None):
                 ddset_tr = node_algo_builder.dset_decoration_args_algo(args, ddset_tr)
                 ddset_val = node_algo_builder.dset_decoration_args_algo(args, ddset_val)
             self.dict_dset_tr.update({na_domain: ddset_tr})
-            self.dict_loader_tr.update({na_domain: mk_loader(ddset_tr, args.bs)})
             self.dict_dset_val.update({na_domain: ddset_val})
         ddset_mix = ConcatDataset(tuple(self.dict_dset_tr.values()))
         self._loader_tr = mk_loader(ddset_mix, args.bs)