Skip to main content

No project description provided

Project description

# TFBuild

TensorFlow Build is an open source library for better science in deep
learning. The main features are (1) a growing repository of layers, models,
datasets, and full training and evaluation pipelines; and (2) abstractions for
the above that allow seamless prototyping, experimentation, and
reproducibility of research.

## Installation

We recommend installing with pip: `pip install tfbuild`.

## Basics

IMPORTANT: this section assumes basic familiarity with TensorFlow.

A TensorFlow graph includes all operations that are part of training and
inference routines. Almost always, this makes models messy, especially when
doing things such as: separating training and inference (and possibly other
modes), training models in a layer-wise fashion, composing multiple models,
training multiple models jointly, and more. TFBuild solves this by defining
abstractions that take care of all the issues above.

### Models

Models only define the mathematical operations done to the input(s) to create
one or more outputs (e.g. loss, discrete output from logits). These should be
input agnostic, meaning that we make no choice as to how we feed the inputs.

Models can define trainable variables, but they should not define training
operations! There are many reasons to opt for this choice. For example, we
don't care about defining optimizers in the graph during inference. This would
also complicate situations where we combine models and then train if both
define training operations.

TFBuild has a growing repository of these models. Since users are likely
interested only in a small subset of these, they are not downloaded by
default. Instead, they can be uploaded and downloaded using our CLI:

# Under construction.

### Wrappers

Wrappers, as the name suggests, wrap models with functionality. This could be,
for example, training or inference. They return the same model instance, but
change their behavior by adding more ops to the computational graph and adding
new bound methods. For example:

from tfbuild.wrappers import TrainingWrapper


model = TrainingWrapper(MyModel('my_model'))
model.train(examples) # not valid if not wrapped

### Input Pipelines


## Recipes


## Uploading to TF Build collections


## Contributing

See the contribution guidelines in ``.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for tfbuild, version 0.1.2
Filename, size File type Python version Upload date Hashes
Filename, size tfbuild-0.1.2-py3-none-any.whl (29.6 kB) File type Wheel Python version py3 Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page