Auto module update

gluoncv/auto/data/auto_data.py

@@ -6,6 +6,8 @@
 from ...utils.download import download
 from ...utils.filesystem import unzip, untar, PathTree
 
+__all__ = ['url_data']
+
 def url_data(url, path=None, overwrite=False, overwrite_folder=False, sha1_hash=None, root=None, disp_depth=1):
     """Download an given URL
 

gluoncv/auto/estimators/base_estimator.py

@@ -265,7 +265,7 @@ def __setstate__(self, state):
         self.__dict__.update(state)
         # logger
         self._logger = logging.getLogger(state.get('_name', self.__class__.__name__))
-        self._logger.setLevel(logging.INFO)
+        self._logger.setLevel(logging.ERROR)
         fh = logging.FileHandler(self._log_file)
         self._logger.addHandler(fh)
         try:
@@ -275,7 +275,7 @@ def __setstate__(self, state):
             with temporary_filename() as tfile:
                 with open(tfile, 'wb') as fo:
                     fo.write(net_params)
-                self.net.load_parameters(tfile)
+                self.net.load_parameters(tfile, ignore_extra=True)
             trainer_state = state['trainer']
             self._init_trainer()
             with temporary_filename() as tfile:
@@ -284,3 +284,4 @@ def __setstate__(self, state):
                 self.trainer.load_states(tfile)
         except ImportError:
             pass
+        self._logger.setLevel(logging.INFO)

gluoncv/auto/estimators/center_net/center_net.py

@@ -47,7 +47,11 @@ def _predict(self, x):
         short_size = min(self._cfg.center_net.data_shape)
         if isinstance(x, str):
             x = load_test(x, short=short_size, max_size=1024)[0]
+        elif isinstance(x, np.ndarray):
+            return self._predict(mx.nd.array(x))
         elif isinstance(x, mx.nd.NDArray):
+            if len(x.shape) != 3 or x.shape[-1] != 3:
+                raise ValueError('array input with shape (h, w, 3) is required for predict')
             x = transform_test(x, short=short_size, max_size=1024)[0]
         elif isinstance(x, pd.DataFrame):
             assert 'image' in x.columns, "Expect column `image` for input images"
@@ -56,6 +60,8 @@ def _predict_merge(x):
                 y['image'] = x
                 return y
             return pd.concat([_predict_merge(xx) for xx in x['image']]).reset_index(drop=True)
+        elif isinstance(x, (list, tuple)):
+            return pd.concat([self._predict(xx) for xx in x]).reset_index(drop=True)
         else:
             raise ValueError('Input is not supported: {}'.format(type(x)))
         height, width = x.shape[2:4]

gluoncv/auto/estimators/faster_rcnn/faster_rcnn.py

@@ -276,6 +276,10 @@ def _predict(self, x):
         short_size = self.net.short[-1] if isinstance(self.net.short, (tuple, list)) else self.net.short
         if isinstance(x, str):
             x = load_test(x, short=short_size, max_size=1024)[0]
+        elif isinstance(x, np.ndarray):
+            if len(x.shape) != 3 or x.shape[-1] != 3:
+                raise ValueError('array input with shape (h, w, 3) is required for predict')
+            return self._predict(mx.nd.array(x))
         elif isinstance(x, mx.nd.NDArray):
             x = transform_test(x, short=short_size, max_size=1024)[0]
         elif isinstance(x, pd.DataFrame):
@@ -285,6 +289,8 @@ def _predict_merge(x):
                 y['image'] = x
                 return y
             return pd.concat([_predict_merge(xx) for xx in x['image']]).reset_index(drop=True)
+        elif isinstance(x, (list, tuple)):
+            return pd.concat([self._predict(xx) for xx in x]).reset_index(drop=True)
         else:
             raise ValueError('Input is not supported: {}'.format(type(x)))
         height, width = x.shape[2:4]

gluoncv/auto/estimators/image_classification/image_classification.py

@@ -374,7 +374,11 @@ def _predict(self, x):
         resize = int(math.ceil(self.input_size / self._cfg.train.crop_ratio))
         if isinstance(x, str):
             x = transform_eval(mx.image.imread(x), resize_short=resize, crop_size=self.input_size)
+        elif isinstance(x, np.ndarray):
+            return self._predict(mx.nd.array(x))
         elif isinstance(x, mx.nd.NDArray):
+            if len(x.shape) != 3 or x.shape[-1] != 3:
+                raise ValueError('array input with shape (h, w, 3) is required for predict')
             x = transform_eval(x, resize_short=resize, crop_size=self.input_size)
         elif isinstance(x, pd.DataFrame):
             assert 'image' in x.columns, "Expect column `image` for input images"

gluoncv/auto/estimators/ssd/ssd.py

@@ -250,7 +250,11 @@ def _predict(self, x):
         short_size = int(self._cfg.ssd.data_shape)
         if isinstance(x, str):
             x = load_test(x, short=short_size, max_size=1024)[0]
+        elif isinstance(x, np.ndarray):
+            return self._predict(mx.nd.array(x))
         elif isinstance(x, mx.nd.NDArray):
+            if len(x.shape) != 3 or x.shape[-1] != 3:
+                raise ValueError('array input with shape (h, w, 3) is required for predict')
             x = transform_test(x, short=short_size, max_size=1024)[0]
         elif isinstance(x, pd.DataFrame):
             assert 'image' in x.columns, "Expect column `image` for input images"
@@ -259,6 +263,8 @@ def _predict_merge(x):
                 y['image'] = x
                 return y
             return pd.concat([_predict_merge(xx) for xx in x['image']]).reset_index(drop=True)
+        elif isinstance(x, (list, tuple)):
+            return pd.concat([self._predict(xx) for xx in x]).reset_index(drop=True)
         else:
             raise ValueError('Input is not supported: {}'.format(type(x)))
         height, width = x.shape[2:4]

gluoncv/auto/estimators/yolo/yolo.py

@@ -251,7 +251,11 @@ def _predict(self, x):
         short_size = int(self._cfg.yolo3.data_shape)
         if isinstance(x, str):
             x = load_test(x, short=short_size, max_size=1024)[0]
+        elif isinstance(x, np.ndarray):
+            return self._predict(mx.nd.array(x))
         elif isinstance(x, mx.nd.NDArray):
+            if len(x.shape) != 3 or x.shape[-1] != 3:
+                raise ValueError('array input with shape (h, w, 3) is required for predict')
             x = transform_test(x, short=short_size, max_size=1024)[0]
         elif isinstance(x, pd.DataFrame):
             assert 'image' in x.columns, "Expect column `image` for input images"
@@ -260,6 +264,8 @@ def _predict_merge(x):
                 y['image'] = x
                 return y
             return pd.concat([_predict_merge(xx) for xx in x['image']]).reset_index(drop=True)
+        elif isinstance(x, (list, tuple)):
+            return pd.concat([self._predict(xx) for xx in x]).reset_index(drop=True)
         else:
             raise ValueError('Input is not supported: {}'.format(type(x)))
         height, width = x.shape[2:4]

gluoncv/auto/tasks/image_classification.py

@@ -31,7 +31,7 @@ class LiteConfig:
     lr : Union[ag.Space, float] = 1e-2
     num_trials : int = 1
     epochs : Union[ag.Space, int] = 5
-    batch_size : Union[ag.Space, int] = 3  # 2 ** 3 == 8
+    batch_size : Union[ag.Space, int] = 8
     nthreads_per_trial : int = 32
     ngpus_per_trial : int = 0
     time_limits : int = 7 * 24 * 60 * 60  # 7 days
@@ -44,7 +44,7 @@ class DefaultConfig:
     lr : Union[ag.Space, float] = ag.Categorical(1e-2, 5e-2)
     num_trials : int = 3
     epochs : Union[ag.Space, int] = 15
-    batch_size : Union[ag.Space, int] = 4  # 2 ** 4 = 16
+    batch_size : Union[ag.Space, int] = 16
     nthreads_per_trial : int = 128
     ngpus_per_trial : int = 8
     time_limits : int = 7 * 24 * 60 * 60  # 7 days
@@ -61,6 +61,13 @@ def _train_image_classification(args, reporter):
     # train, val data
     train_data = args.pop('train_data')
     val_data = args.pop('val_data')
+    # exponential batch size for Int() space batch sizes
+    try:
+        exp_batch_size = args.pop('exp_batch_size')
+    except AttributeError:
+        exp_batch_size = False
+    if exp_batch_size and 'batch_size' in args:
+        args['batch_size'] = 2 ** args['batch_size']
     try:
         task = args.pop('task')
         dataset = args.pop('dataset')
@@ -141,6 +148,8 @@ def __init__(self, config=None, estimator=None, logger=None):
         else:
             if not config.get('dist_ip_addrs', None):
                 ngpus_per_trial = config.get('ngpus_per_trial', gpu_count)
+                if ngpus_per_trial > gpu_count:
+                    ngpus_per_trial = gpu_count
                 if ngpus_per_trial < 1:
                     self._logger.info('No GPU detected/allowed, using most conservative search space.')
                     default_config = LiteConfig()
@@ -167,10 +176,6 @@ def __init__(self, config=None, estimator=None, logger=None):
         # additional configs
         config['num_workers'] = nthreads_per_trial
         config['gpus'] = [int(i) for i in range(ngpus_per_trial)]
-        # if config['gpus']:
-        #     config['batch_size'] = config.get('batch_size', 8) * len(config['gpus'])
-        #     self._logger.info('Increase batch size to %d based on the number of gpus %d',
-        #                       config['batch_size'], len(config['gpus']))
         config['seed'] = config.get('seed', np.random.randint(32,767))
         self._config = config
 
@@ -242,17 +247,6 @@ def fit(self, train_data, val_data=None, train_size=0.9, random_state=None):
                               len(train), len(val))
             train_data, val_data = train, val
 
-        # automatically suggest some hyperparameters based on the dataset statistics(experimental)
-        # estimator = self._config.get('estimator', None)
-        # if estimator is None:
-        #     estimator = [ImageClassificationEstimator]
-        # elif isinstance(estimator, (tuple, list)):
-        #     pass
-        # else:
-        #     assert issubclass(estimator, BaseEstimator)
-        #     estimator = [estimator]
-        # self._config['estimator'] = ag.Categorical(*estimator)
-
         estimator = self._config.get('estimator', None)
         if estimator is None:
             estimator = [ImageClassificationEstimator]
@@ -266,7 +260,12 @@ def fit(self, train_data, val_data=None, train_size=0.9, random_state=None):
                     estimator[i] = ImageClassificationEstimator
                 else:
                     estimator.pop(e)
-        self._config['estimator'] = ag.Categorical(*estimator)
+        if not estimator:
+            raise ValueError('Unable to determine the estimator for fit function.')
+        if len(estimator) == 1:
+            self._config['estimator'] = estimator[0]
+        else:
+            self._config['estimator'] = ag.Categorical(*estimator)
 
         # register args
         config = self._config.copy()

gluoncv/auto/tasks/object_detection.py

@@ -59,39 +59,16 @@ def _train_object_detection(args, reporter):
     ----------
     args: <class 'autogluon.utils.edict.EasyDict'>
     """
-    # pruning for batch size
-    # if args.get('batch_size', None):
-    #     if args.estimator == FasterRCNNEstimator and args.batch_size not in [4, 8]:
-    #         logging.info('Estimator and batch size are not matched, this trial is skipped.')
-    #         return
-    #     elif args.estimator != FasterRCNNEstimator and args.batch_size in [4, 8]:
-    #         logging.info('Estimator and batch size are not matched, this trial is skipped.')
-    #         return
-
-    # pruning for base network
-    # if args.get('base_network', None):
-    #     if args.estimator == SSDEstimator and \
-    #             args.base_network not in ['vgg16_atrous', 'resnet18_v1', 'resnet50_v1',
-    #                                       'resnet101_v2', 'resnet152_v2', 'resnet34_v1b']:
-    #         logging.info('Estimator and base network are not matched, this trial is skipped.')
-    #         return
-    #     elif args.estimator == YOLOv3Estimator and \
-    #             args.base_network not in ['darknet53']:
-    #         logging.info('Estimator and base network are not matched, this trial is skipped.')
-    #         return
-    #     elif args.estimator == FasterRCNNEstimator and \
-    #             args.base_network not in ['resnet50_v1b', 'resnet101_v1d',
-    #                                       'resnest50', 'resnest101', 'resnest269']:
-    #         logging.info('Estimator and base network are not matched, this trial is skipped.')
-    #         return
-    #     elif args.estimator == CenterNetEstimator and \
-    #             args.base_network not in ['resnet18_v1b', 'resnet50_v1b', 'resnet101_v1b', 'dla34']:
-    #         logging.info('Estimator and base network are not matched, this trial is skipped.')
-    #         return
-
     # train, val data
     train_data = args.pop('train_data')
     val_data = args.pop('val_data')
+    # exponential batch size for Int() space batch sizes
+    try:
+        exp_batch_size = args.pop('exp_batch_size')
+    except AttributeError:
+        exp_batch_size = False
+    if exp_batch_size and 'batch_size' in args:
+        args['batch_size'] = 2 ** args['batch_size']
     try:
         task = args.pop('task')
         dataset = args.pop('dataset')
@@ -164,6 +141,8 @@ def __init__(self, config=None, logger=None):
         else:
             if not config.get('dist_ip_addrs', None):
                 ngpus_per_trial = config.get('ngpus_per_trial', gpu_count)
+                if ngpus_per_trial > gpu_count:
+                    ngpus_per_trial = gpu_count
                 if ngpus_per_trial < 1:
                     self._logger.info('No GPU detected/allowed, using most conservative search space.')
                     default_config = LiteConfig()
@@ -215,10 +194,6 @@ def __init__(self, config=None, logger=None):
         # additional configs
         config['num_workers'] = nthreads_per_trial
         config['gpus'] = [int(i) for i in range(ngpus_per_trial)]
-        # if config['gpus']:
-        #     config['batch_size'] = config.get('batch_size', 8) * len(config['gpus'])
-        #     self._logger.info('Increase batch size to %d based on the number of gpus %d',
-        #                       config['batch_size'], len(config['gpus']))
         config['seed'] = config.get('seed', np.random.randint(32,767))
         self._config = config
 

gluoncv/auto/tasks/utils.py

@@ -114,7 +114,12 @@ def auto_suggest(config, estimator, logger):
                 estimator[i] = FasterRCNNEstimator
             elif e == 'center_net':
                 estimator[i] = CenterNetEstimator
-        config['estimator'] = ag.Categorical(*estimator)
+        if not estimator:
+            raise ValueError('Unable to determine the estimator for fit function.')
+        if len(estimator) == 1:
+            config['estimator'] = estimator[0]
+        else:
+            config['estimator'] = ag.Categorical(*estimator)
 
     # get dataset statistics
     # user needs to define a Dataset object "train_dataset" when using custom dataset
@@ -217,36 +222,6 @@ def get_recursively(search_dict, field):
 
 def config_to_nested(config):
     """Convert config to nested version"""
-    # estimator = config.get('estimator', None)
-    # if estimator is None:
-    #     transfer = config.get('transfer', None)
-    #     assert transfer is not None, "estimator or transfer is required in search space"
-    #     if transfer.startswith('ssd'):
-    #         estimator = SSDEstimator
-    #     elif transfer.startswith('faster_rcnn'):
-    #         estimator = FasterRCNNEstimator
-    #     elif transfer.startswith('yolo3'):
-    #         estimator = YOLOv3Estimator
-    #     elif transfer.startswith('center_net'):
-    #         estimator = CenterNetEstimator
-    #     else:
-    #         estimator = ImageClassificationEstimator
-    # else:
-    #     # str to instance
-    #     if isinstance(estimator, str):
-    #         if estimator == 'ssd':
-    #             estimator = SSDEstimator
-    #         elif estimator == 'faster_rcnn':
-    #             estimator = FasterRCNNEstimator
-    #         elif estimator == 'yolo3':
-    #             estimator = YOLOv3Estimator
-    #         elif estimator == 'center_net':
-    #             estimator = CenterNetEstimator
-    #         elif estimator == 'img_cls':
-    #             estimator = ImageClassificationEstimator
-    #         else:
-    #             raise ValueError(f'Unknown estimator: {estimator}')
-
     estimator = config.get('estimator', None)
     transfer = config.get('transfer', None)
     # choose hyperparameters based on pretrained model in transfer learning
@@ -294,21 +269,28 @@ def config_to_nested(config):
     else:
         assert issubclass(estimator, BaseEstimator)
 
-    # batch size is the power of 2
-    if config.get('batch_size', None):
-        config['batch_size'] = 2 ** config['batch_size']
-
     cfg_map = estimator._default_cfg.asdict()
 
-    def _recursive_update(config, key, value):
+    def _recursive_update(config, key, value, auto_strs, auto_ints):
         for k, v in config.items():
+            if k in auto_strs:
+                config[k] = 'auto'
+            if k in auto_ints:
+                config[k] = -1
             if key == k:
                 config[key] = value
             elif isinstance(v, dict):
-                _recursive_update(v, key, value)
+                _recursive_update(v, key, value, auto_strs, auto_ints)
 
+    if 'use_rec' in config:
+        auto_strs = ['data_dir']
+        auto_ints = []
+    else:
+        auto_strs = ['data_dir', 'rec_train', 'rec_train_idx', 'rec_val', 'rec_val_idx',
+                     'dataset', 'dataset_root']
+        auto_ints = ['num_training_samples']
     for k, v in config.items():
-        _recursive_update(cfg_map, k, v)
+        _recursive_update(cfg_map, k, v, auto_strs, auto_ints)
     cfg_map['estimator'] = estimator
     return cfg_map