What is tensorpack?¶
Tensorpack is a training interface based on TensorFlow, which means: you’ll use mostly tensorpack high-level APIs to do training, rather than TensorFlow low-level APIs.
TensorFlow is powerful, but at the same time too complicated for a lot of people, especially when speed is a concern. Users can often write slow code with low-level APIs or other existing high-level wrappers. Even a lot of official TensorFlow examples are written for simplicity rather than efficiency, which as a result makes people think TensorFlow is slow.
The official TensorFlow benchmark said this in their README:
These models are designed for performance. For models that have clean and easy-to-read implementations, see the TensorFlow Official Models.
which seems to suggest that you cannot have performance and ease-of-use together. However you can have them both in tensorpack. Tensorpack uses TensorFlow efficiently, and hides performance details under its APIs. You no longer need to write data prefetch, multi-GPU replication, device placement, variables synchronization – anything that’s unrelated to the model itself. You still need to learn to write models with TF, but performance is all taken care of by tensorpack.
A High Level Glance¶
DataFlowis a library to load data efficiently in Python. Apart from DataFlow, native TF operators can be used for data loading as well. They will eventually be wrapped under the same
InputSourceinterface and go through prefetching.
You can use any TF-based symbolic function library to define a model, including a small set of functions within tensorpack.
ModelDescis an interface to connect the model with the
Tensorpack trainers manage the training loops for you. They also include data parallel logic for multi-GPU or distributed training. At the same time, you have the power of customization through callbacks.
Callbacks are like
tf.train.SessionRunHook, or plugins. During training, everything you want to do other than the main iterations can be defined through callbacks and easily reused.
All the components, though work perfectly together, are highly decorrelated: you can:
Use DataFlow alone as a data loading library, without tensorfow at all.
Use tensorpack to build the graph with multi-GPU or distributed support, then train it with your own loops.
Build the graph on your own, and train it with tensorpack callbacks.