update keras example

docs/tutorial/input-source.md

@@ -72,8 +72,9 @@ Let's take a look at what users are asking for:
 * [Different ways to pad your data](https://github.com/tensorflow/tensorflow/issues/13969)
 * [Handle none values in data](https://github.com/tensorflow/tensorflow/issues/13865)
 * [Handle dataset that's not a multiple of batch size](https://github.com/tensorflow/tensorflow/issues/13745)
-* [Take variable-length np array](https://github.com/tensorflow/tensorflow/issues/13018)
 * [Different levels of determinism](https://github.com/tensorflow/tensorflow/issues/13932)
+* [Sort/skip some data](https://github.com/tensorflow/tensorflow/issues/14250)
+* [Take variable-length np array](https://github.com/tensorflow/tensorflow/issues/13018)
 
 To support these features which could've been done with 3 lines of code in Python, you need either a new TF
 API, or ask [Dataset.from_generator](https://www.tensorflow.org/versions/r1.4/api_docs/python/tf/contrib/data/Dataset#from_generator)
@@ -82,8 +83,8 @@ API, or ask [Dataset.from_generator](https://www.tensorflow.org/versions/r1.4/ap
 It only makes sense to use TF to read data, if your data is originally very clean and well-formated.
 If not, you may feel like writing a script to clean your data, but then you're almost writing a Python loader already!
 
-Think about it: it's a waste of time to write a Python script to transform from raw data to TFRecords,
-then a TF script to transform from TFRecords to tensors.
+Think about it: it's a waste of time to write a Python script to transform from raw data to clean format (e.g. TFRecords),
+then a TF script to transform from this format to tensors.
 The intermediate step (TFRecords) doesn't have to exist.
 You just need the right interface to connect Python to the graph directly, efficiently.
 `tensorpack.InputSource` is such an interface.

examples/mnist-keras-v2.py

@@ -57,11 +57,6 @@ if __name__ == '__main__':
         metrics=['accuracy']
     )
     M.fit(
-        callbacks=[
-            ModelSaver(),
-            InferenceRunner(
-                dataset_test,
-                [ScalarStats(['total_loss', 'accuracy'])]),
-        ],
+        validation_data=dataset_test,
         steps_per_epoch=dataset_train.size(),
     )

tensorpack/callbacks/saver.py

@@ -8,7 +8,6 @@ import os
 
 from .base import Callback
 from ..utils import logger
-from ..utils.develop import log_deprecated
 from ..tfutils.common import get_tf_version_number
 
 __all__ = ['ModelSaver', 'MinSaver', 'MaxSaver']
@@ -22,8 +21,7 @@ class ModelSaver(Callback):
     def __init__(self, max_to_keep=10,
                  keep_checkpoint_every_n_hours=0.5,
                  checkpoint_dir=None,
-                 var_collections=tf.GraphKeys.GLOBAL_VARIABLES,
-                 keep_recent=None, keep_freq=None):
+                 var_collections=tf.GraphKeys.GLOBAL_VARIABLES):
         """
         Args:
             max_to_keep (int): the same as in ``tf.train.Saver``.
@@ -33,12 +31,6 @@ class ModelSaver(Callback):
         """
         self._max_to_keep = max_to_keep
         self._keep_every_n_hours = keep_checkpoint_every_n_hours
-        if keep_recent is not None or keep_freq is not None:
-            log_deprecated("ModelSaver(keep_recent=, keep_freq=)", "Use max_to_keep and keep_checkpoint_every_n_hours!")
-            if keep_recent is not None:
-                self._max_to_keep = keep_recent
-            if keep_freq is not None:
-                self._keep_every_n_hours = keep_freq
 
         if not isinstance(var_collections, list):
             var_collections = [var_collections]

tensorpack/contrib/keras.py

@@ -9,7 +9,9 @@ import keras
 from ..graph_builder import InputDesc
 from ..tfutils.tower import get_current_tower_context
 from ..tfutils.collection import freeze_collection
-from ..callbacks import Callback, InferenceRunner, CallbackToHook
+from ..callbacks import (
+    Callback, InferenceRunner, CallbackToHook,
+    ScalarStats, ModelSaver)
 from ..tfutils.summary import add_moving_summary
 from ..utils.gpu import get_nr_gpu
 from ..train import Trainer, SimpleTrainer, SyncMultiGPUTrainerParameterServer
@@ -107,6 +109,9 @@ class KerasModel(object):
         """
         Args:
             model (keras.model.Model):
+            input (InputSource):
+            trainer (Trainer): the default will check the number of available
+                GPUs and use them all.
         """
         self.model = model
         if trainer is None:
@@ -117,10 +122,16 @@ class KerasModel(object):
                 trainer = SyncMultiGPUTrainerParameterServer(nr_gpu)
         assert isinstance(trainer, Trainer), trainer
 
-        self.trainer = trainer
         self.input = input
+        self.trainer = trainer
 
     def compile(self, optimizer, loss, metrics):
+        """
+        Args:
+            optimizer (tf.train.Optimizer):
+            loss, metrics: same as in `keras.model.Model.compile()`.
+        """
+        self._metrics = metrics
         setup_keras_trainer(
             self.trainer, model=self.model,
             input=self.input,
@@ -128,10 +139,21 @@ class KerasModel(object):
             loss=loss,
             metrics=metrics)
 
-    def fit(self, **kwargs):
+    def fit(self, validation_data=None, **kwargs):
+        """
+        Args:
+            validation_data (DataFlow or InputSource): to be used for inference.
+            kwargs: same as `self.trainer.train_with_defaults`.
+        """
         callbacks = kwargs.pop('callbacks', [])
         callbacks.extend(self.get_default_callbacks())
-        self.trainer.train_with_defaults(**kwargs)
+        if validation_data is not None:
+            callbacks.append(
+                InferenceRunner(
+                    validation_data, ScalarStats(self._metrics + ['total_loss'])))
+        self.trainer.train_with_defaults(callbacks=callbacks, **kwargs)
 
     def get_default_callbacks(self):
-        return []
+        return [
+            ModelSaver(keep_checkpoint_every_n_hours=0.2)
+        ]

tensorpack/dataflow/common.py

@@ -219,6 +219,10 @@ class FixedSizeData(ProxyDataFlow):
     def size(self):
         return self._size
 
+    def reset_state(self):
+        super(FixedSizeData, self).reset_state()
+        self.itr = self.ds.get_data()
+
     def get_data(self):
         with self._guard:
             if self.itr is None: