release v0.1.dev21

pytorch_lightning/models/trainer.py

@@ -5,6 +5,7 @@ from pytorch_lightning.root_module.memory import get_gpu_memory_map
 import traceback
 from pytorch_lightning.root_module.model_saving import TrainerIO
 from torch.optim.lr_scheduler import MultiStepLR
+from torch.nn import DataParallel
 import pdb
 
 try:
@@ -33,6 +34,8 @@ class Trainer(TrainerIO):
                  log_save_interval=1, add_log_row_interval=1,
                  lr_scheduler_milestones=None,
                  use_amp=False,
+                 check_grad_nans=False,
+                 amp_level='O2',
                  nb_sanity_val_steps=5):
 
         # Transfer params
@@ -58,6 +61,10 @@ class Trainer(TrainerIO):
         self.nb_sanity_val_steps = nb_sanity_val_steps
         self.lr_scheduler_milestones = [] if lr_scheduler_milestones is None else [int(x.strip()) for x in lr_scheduler_milestones.split(',')]
         self.lr_schedulers = []
+        self.amp_level = amp_level
+        self.check_grad_nans = check_grad_nans
+        self.data_parallel_device_ids = [0]
+        self.data_parallel = False
 
         # training state
         self.optimizers = None
@@ -122,21 +129,21 @@ class Trainer(TrainerIO):
         self.tqdm_metrics = {}
 
         # determine number of training batches
-        nb_tng_batches = self.model.nb_batches(self.tng_dataloader)
-        self.nb_tng_batches = int(nb_tng_batches * self.train_percent_check)
+        self.nb_tng_batches = self.model.nb_batches(self.tng_dataloader)
+        self.nb_tng_batches = int(self.nb_tng_batches * self.train_percent_check)
 
         # determine number of validation batches
-        nb_val_batches = self.model.nb_batches(self.val_dataloader)
-        nb_val_batches = int(nb_val_batches * self.val_percent_check)
-        nb_val_batches = max(1, nb_val_batches)
-        self.nb_val_batches = nb_val_batches
+        self.nb_val_batches = self.model.nb_batches(self.val_dataloader)
+        self.nb_val_batches = int(self.nb_val_batches * self.val_percent_check)
+        self.nb_val_batches = max(1, self.nb_val_batches)
+        self.nb_val_batches = self.nb_val_batches
 
         # determine number of test batches
-        nb_test_batches = self.model.nb_batches(self.test_dataloader)
-        self.nb_test_batches = int(nb_test_batches * self.test_percent_check)
+        self.nb_test_batches = self.model.nb_batches(self.test_dataloader)
+        self.nb_test_batches = int(self.nb_test_batches * self.test_percent_check)
 
         # determine when to check validation
-        self.val_check_batch = int(nb_tng_batches * self.val_check_interval)
+        self.val_check_batch = int(self.nb_tng_batches * self.val_check_interval)
 
     def __add_tqdm_metrics(self, metrics):
         for k, v in metrics.items():
@@ -163,19 +170,19 @@ class Trainer(TrainerIO):
         outputs = []
 
         # run training
-        for i, data_batch in enumerate(dataloader):
+        for batch_i, data_batch in enumerate(dataloader):
 
             if data_batch is None:
                 continue
 
             # stop short when on fast dev run
-            if max_batches is not None and i >= max_batches:
+            if max_batches is not None and batch_i >= max_batches:
                 break
 
             # -----------------
             # RUN VALIDATION STEP
             # -----------------
-            output = model.validation_step(data_batch)
+            output = model.validation_step(data_batch, batch_i)
             outputs.append(output)
 
             # batch done
@@ -222,7 +229,7 @@ class Trainer(TrainerIO):
         if self.use_amp:
             # An example
             self.model, optimizer = amp.initialize(
-                self.model, self.optimizers[0], opt_level="O2",
+                self.model, self.optimizers[0], opt_level=self.amp_level,
             )
             self.optimizers[0] = optimizer
             model.trainer = self
@@ -238,6 +245,9 @@ class Trainer(TrainerIO):
 
         # put on gpu if needed
         if self.on_gpu:
+            if self.data_parallel:
+                model = DataParallel(model, device_ids=self.data_parallel_device_ids)
+            else:
                 model = model.cuda()
 
         # run tiny validation to make sure program won't crash during val
@@ -290,7 +300,7 @@ class Trainer(TrainerIO):
                 # ---------------
                 # RUN TRAIN STEP
                 # ---------------
-                batch_result = self.__run_tng_batch(data_batch)
+                batch_result = self.__run_tng_batch(data_batch, batch_nb)
                 early_stop_epoch = batch_result == -1
 
                 # ---------------
@@ -348,7 +358,7 @@ class Trainer(TrainerIO):
                     return
 
 
-    def __run_tng_batch(self, data_batch):
+    def __run_tng_batch(self, data_batch, batch_nb):
         if data_batch is None:
             return 0
 
@@ -363,7 +373,7 @@ class Trainer(TrainerIO):
 
         # forward pass
         # return a scalar value and a dic with tqdm metrics
-        loss, model_specific_tqdm_metrics_dic = self.model.training_step(data_batch)
+        loss, model_specific_tqdm_metrics_dic = self.model.training_step(data_batch, batch_nb)
         self.__add_tqdm_metrics(model_specific_tqdm_metrics_dic)
 
         # backward pass
@@ -374,6 +384,10 @@ class Trainer(TrainerIO):
         else:
             loss.backward()
 
+        if self.check_grad_nans:
+            for param in self.model.parameters():
+                print(param.grad.float().sum())
+
         self.batch_loss_value += loss.item()
 
         # gradient update with accumulated gradients

pytorch_lightning/root_module/model_saving.py

@@ -1,6 +1,7 @@
 import torch
 import os
 import re
+import pdb
 
 
 class ModelIO(object):
@@ -99,6 +100,9 @@ class TrainerIO(object):
     # PRIVATE OPS
     # ----------------------------------
     def hpc_save(self, folderpath, experiment):
+        # make sure the checkpoint folder exists
+        os.makedirs(folderpath, exist_ok=True)
+
         # save exp to make sure we get all the metrics
         experiment.save()
 
@@ -130,6 +134,10 @@ class TrainerIO(object):
 
     def max_ckpt_in_folder(self, path):
         files = os.listdir(path)
+        files = [x for x in files if 'ckpt_' in x]
+        if len(files) == 0:
+            return 0
+
         ckpt_vs = []
         for name in files:
             name = name.split('ckpt_')[-1]

pytorch_lightning/root_module/root_module.py

@@ -51,7 +51,7 @@ class RootModule(GradInformation, ModelIO, OptimizerConfig, ModelHooks):
         """
         raise NotImplementedError
 
-    def validation_step(self, data_batch):
+    def validation_step(self, data_batch, batch_nb):
         """
         return whatever outputs will need to be aggregated in validation_end
         :param data_batch:
@@ -67,7 +67,7 @@ class RootModule(GradInformation, ModelIO, OptimizerConfig, ModelHooks):
         """
         raise NotImplementedError
 
-    def training_step(self, data_batch):
+    def training_step(self, data_batch, batch_nb):
         """
         return loss, dict with metrics for tqdm
         :param data_batch:

pytorch_lightning/utils/arg_parse.py

@@ -51,6 +51,9 @@ def add_default_args(parser, root_dir, rand_seed=None, possible_model_names=None
     parser.add_argument('--disable_cuda', dest='disable_cuda', action='store_true')
     parser.add_argument('--default_tensor_type', default='torch.cuda.FloatTensor', type=str)
     parser.add_argument('--use_amp', dest='use_amp', action='store_true')
+    parser.add_argument('--check_grad_nans', dest='check_grad_nans', action='store_true')
+    parser.add_argument('--amp_level', default='O2',type=str)
+
 
     # run on hpc
     parser.add_argument('--on_cluster', dest='on_cluster', action='store_true')

setup.py

@@ -7,7 +7,7 @@ from setuptools import setup, find_packages
 # http://blog.ionelmc.ro/2014/05/25/python-packaging/
 setup(
     name="pytorch-lightning",
-    version='0.1.dev18',
+    version='0.1.dev21',
     description="The Keras for ML researchers using PyTorch",
     author="William Falcon",
     author_email="waf2107@columbia.edu",
@@ -17,12 +17,13 @@ setup(
     keywords=["deep learning", "pytorch", "AI"],
     python_requires=">=3.5",
     install_requires=[
-        "torch",
+        "torch>=1.0.0",
         "tqdm",
         "test-tube",
     ],
     packages=find_packages(),
     long_description=open("README.md", encoding="utf-8").read(),
+    long_description_content_type='text/markdown',
     include_package_data=True,
     zip_safe=False,
 )