Upgrade to newest version of pytorch (#1070)

* Update to newest version of pytorch:
* A couple of interface changes when creating the resnet model, in the Trainer and the model summary
* Updated the yaml files to remove pytorch versions
* Bumped version

donkeycar/parts/pytorch/torch_train.py

@@ -2,6 +2,7 @@
 from pathlib import Path
 import torch
 import pytorch_lightning as pl
+from pytorch_lightning.utilities.model_summary import summarize
 from donkeycar.parts.pytorch.torch_data import TorchTubDataModule
 from donkeycar.parts.pytorch.torch_utils import get_model_by_type
 
@@ -16,21 +17,17 @@ def train(cfg, tub_paths, model_output_path, model_type, checkpoint_path=None):
     if is_torch_model:
         model = f'{model_name}.ckpt'
     else:
-        print("Unrecognized model file extension for model_output_path: '{}'. Please use the '.ckpt' extension.".format(
-            model_output_path))
-
+        print(f"Unrecognized model file extension for model_output_path: '"
+              f"{model_output_path}'. Please use the '.ckpt' extension.")
 
     if not model_type:
         model_type = cfg.DEFAULT_MODEL_TYPE
 
     tubs = tub_paths.split(',')
     tub_paths = [os.path.expanduser(tub) for tub in tubs]
     output_path = os.path.expanduser(model_output_path)
-
-    output_dir = Path(model_output_path).parent
-
+    output_dir = str(Path(model_output_path).parent)
     model = get_model_by_type(model_type, cfg, checkpoint_path=checkpoint_path)
-
     if torch.cuda.is_available():
         print('Using CUDA')
         gpus = -1
@@ -40,15 +37,17 @@ def train(cfg, tub_paths, model_output_path, model_type, checkpoint_path=None):
 
     logger = None
     if cfg.VERBOSE_TRAIN:
-        print("Tensorboard logging started. Run `tensorboard --logdir ./tb_logs` in a new terminal")
+        print("Tensorboard logging started. Run `tensorboard --logdir "
+              "./tb_logs` in a new terminal")
         from pytorch_lightning.loggers import TensorBoardLogger
 
         # Create Tensorboard logger
         logger = TensorBoardLogger('tb_logs', name=model_name)
 
-    weights_summary = 'full' if cfg.PRINT_MODEL_SUMMARY else 'top'
-    trainer = pl.Trainer(gpus=gpus, logger=logger, progress_bar_refresh_rate=30,
-                         max_epochs=cfg.MAX_EPOCHS, default_root_dir=output_dir, weights_summary=weights_summary)
+    if cfg.PRINT_MODEL_SUMMARY:
+        summarize(model)
+    trainer = pl.Trainer(gpus=gpus, logger=logger, max_epochs=cfg.MAX_EPOCHS,
+                         default_root_dir=output_dir)
 
     data_module = TorchTubDataModule(cfg, tub_paths)
     trainer.fit(model, data_module)

donkeycar/tests/test_torch.py

@@ -102,8 +102,7 @@ def test_training_pipeline(config: Config, model_type: str, car_dir: str) \
         gpus = 0
 
     # Overfit the data
-    trainer = pl.Trainer(gpus=gpus, overfit_batches=2,
-                         progress_bar_refresh_rate=30, max_epochs=30)
+    trainer = pl.Trainer(gpus=gpus, overfit_batches=2, max_epochs=30)
     trainer.fit(model, data_module)
     final_loss = model.loss_history[-1]
     assert final_loss < 0.35, \

install/envs/mac.yml

@@ -8,6 +8,7 @@ channels:
 
 dependencies:
   - python=3.7
+  - numpy=1.19
   - h5py
   - pillow
   - opencv
@@ -26,11 +27,10 @@ dependencies:
   - PrettyTable
   - pyfiglet
   - mypy
-  - pytorch=1.7.1
-  - torchvision
+  - pytorch
+  - torchvision=0.12
   - torchaudio
   - pytorch-lightning
-  - numpy
   - psutil
   - kivy=2.0.0
   - plotly

install/envs/ubuntu.yml

@@ -8,6 +8,7 @@ channels:
 
 dependencies:
   - python=3.7
+  - numpy=1.19
   - h5py
   - pillow
   - opencv
@@ -27,11 +28,10 @@ dependencies:
   - PrettyTable
   - pyfiglet
   - mypy
-  - pytorch=1.7.1
-  - torchvision
+  - pytorch
+  - torchvision=0.12
   - torchaudio
   - pytorch-lightning
-  - numpy
   - psutil
   - kivy=2.0.0
   - plotly

install/envs/windows.yml

@@ -8,6 +8,7 @@ channels:
 
 dependencies:
   - python=3.7
+  - numpy=1.19
   - h5py
   - pillow
   - opencv
@@ -27,11 +28,10 @@ dependencies:
   - PrettyTable
   - pyfiglet
   - mypy
-  - pytorch=1.7.1
-  - torchvision
+  - pytorch
+  - torchvision=0.12
   - torchaudio
   - pytorch-lightning
-  - numpy
   - kivy=2.0.0
   - plotly
   - pyyaml

setup.py

@@ -24,7 +24,7 @@ def package_files(directory, strip_leading):
     long_description = fh.read()
 
 setup(name='donkeycar',
-      version="4.4.1-main",
+      version="4.4.2-main",
       long_description=long_description,
       description='Self driving library for python.',
       url='https://github.com/autorope/donkeycar',
@@ -89,8 +89,8 @@ def package_files(directory, strip_leading):
           'ci': ['codecov'],
           'tf': ['tensorflow==2.2.0'],
           'torch': [
-              'pytorch>=1.7.1',
-              'torchvision',
+              'pytorch',
+              'torchvision==0.12',
               'torchaudio',
               'fastai'
           ],