Skip to main content

A light wrapper over TensorFlow that enables you to easily create complex deep neural networks using the Builder Pattern through a functional fluent immutable API

Project description

# Tensor Builder
TensorBuilder is light-weight extensible library that enables you to easily create complex deep neural networks through a functional [fluent](https://en.wikipedia.org/wiki/Fluent_interface) [immutable](https://en.wikipedia.org/wiki/Immutable_object) API based on the Builder Pattern. Tensor Builder also comes with a DSL based on [applicatives](http://learnyouahaskell.com/functors-applicative-functors-and-monoids) and function composition that enables you to express more clearly the structure of your network, make changes faster, and reuse code.

### Goals

* Be a light-wrapper around Tensor-based libraries
* Enable users to easily create complex branched topologies while maintaining a fluent API (see [Builder.branch](http://cgarciae.github.io/tensorbuilder/api/builder.m.html#tensorbuilder.api.builder.Builder.branch))
* Let users be expressive and productive through a DSL

## Installation
Tensor Builder assumes you have a working `tensorflow` installation. We don't include it in the `requirements.txt` since the installation of tensorflow varies depending on your setup.

#### From github
1. `pip install git+https://github.com/cgarciae/tensorbuilder.git@0.0.10`

#### From pip
Coming soon!

## Getting Started

Create neural network with a [5, 10, 3] architecture with a `softmax` output layer and a `tanh` hidden layer through a Builder and then get back its tensor:

import tensorflow as tf
from tensorbuilder import tb

x = tf.placeholder(tf.float32, shape=[None, 5])
keep_prob = tf.placeholder(tf.float32)

h = (
tb
.build(x)
.tanh_layer(10) # tanh(x * w + b)
.dropout(keep_prob) # dropout(x, keep_prob)
.softmax_layer(3) # softmax(x * w + b)
.tensor()
)

## Features
* **Branching**: Enable to easily express complex complex topologies with a fluent API. See [Branching](https://cgarciae.gitbooks.io/tensorbuilder/content/branching/).
* **Scoping**: Enable you to express scopes for your tensor graph using methods such as `tf.device` and `tf.variable_scope` with the same fluent API. [Scoping](https://cgarciae.gitbooks.io/tensorbuilder/content/scoping/).
* **DSL**: Use an abbreviated notation with a functional style to make the creation of networks faster, structural changes easier, and reuse code. See [DSL](https://cgarciae.gitbooks.io/tensorbuilder/content/dsl/).
* **Patches**: Add functions from other Tensor-based libraries as methods of the Builder class. TensorBuilder gives you a curated patch plus some specific patches from `TensorFlow` and `TFLearn`, but you can build you own to make TensorBuilder what you want it to be. See [Patches](https://cgarciae.gitbooks.io/tensorbuilder/content/patches/).

## Documentation
* [Complete API](http://cgarciae.github.io/tensorbuilder/api/index.html).
* [Core API](http://cgarciae.github.io/tensorbuilder/core/index.html).
* [Complete Documentation](http://cgarciae.github.io/tensorbuilder/index.html)

## The Guide
Check out [The Guide](https://cgarciae.gitbooks.io/tensorbuilder/content/) to learn to code in TensorBuilder.

## Full Example
Next is an example with all the features of TensorBuilder including the DSL, branching and scoping. It creates a branched computation where each branch is executed on a different device. All branches are then reduced to a single layer, but the computation is the branched again to obtain both the activation function and the trainer.

import tensorflow as tf
from tensorbuilder import tb

x = placeholder(tf.float32, shape=[None, 10])
y = placeholder(tf.float32, shape=[None, 5])

[activation, trainer] = tb.pipe(
x,
[
{ tf.device("/gpu:0"):
tb.relu_layer(20)
}
,
{ tf.device("/gpu:1"):
tb.sigmoid_layer(20)
}
,
{ tf.device("/cpu:0"):
tb.tanh_layer(20)
}
],
tb.linear_layer(5),
[
tb.softmax() # activation
,
tb
.softmax_cross_entropy_with_logits(y) # loss
.map(tf.train.AdamOptimizer(0.01).minimize) # trainer
],
tb.tensors()
)

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 tensorbuilder, version 0.0.14
Filename, size File type Python version Upload date Hashes
Filename, size tensorbuilder-0.0.14.tar.gz (20.6 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page