https://github.com/danielsabinasz/TensorSlow

Re-implementation of TensorFlow in pure python, with an emphasis on code understandability
https://github.com/danielsabinasz/TensorSlow

artificial-intelligence artificial-neural-networks deep-learning python tensorflow

Last synced: 10 months ago
JSON representation

Re-implementation of TensorFlow in pure python, with an emphasis on code understandability

• Host: GitHub
• URL: https://github.com/danielsabinasz/TensorSlow
• Owner: danielsabinasz
• Created: 2017-08-24T20:51:07.000Z (over 8 years ago)
• Default Branch: master
• Last Pushed: 2021-04-11T22:29:25.000Z (almost 5 years ago)
• Last Synced: 2024-11-12T11:40:22.230Z (over 1 year ago)
• Topics: artificial-intelligence, artificial-neural-networks, deep-learning, python, tensorflow
• Language: Jupyter Notebook
• Homepage: http://www.deepideas.net
• Size: 4.52 MB
• Stars: 679
• Watchers: 41
• Forks: 89
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# TensorSlow

## A re-implementation of TensorFlow functionality in pure python

TensorSlow is a minimalist machine learning API that mimicks the TensorFlow API, but is implemented in pure python (without a C backend). The source code has been built with maximal understandability in mind, rather than maximal efficiency. Therefore, TensorSlow should be used solely for educational purposes. If you want to understand how deep learning libraries like TensorFlow work under the hood, this may be your best shot. 

I have written an article in my blog at deepideas.net that develops this library step by step, explaining all the math and algorithms on the way: Deep Learning From Scratch.

## How to use

Import:

    import tensorslow as ts

Create a computational graph:

    ts.Graph().as_default()

Create input placeholders:

    training_features = ts.placeholder()

    training_classes = ts.placeholder()

Build a model:

	weights = ts.Variable(np.random.randn(2, 2))

	biases = ts.Variable(np.random.randn(2))

	model = ts.softmax(ts.add(ts.matmul(X, W), b))

Create training criterion:

    loss = ts.negative(ts.reduce_sum(ts.reduce_sum(ts.multiply(training_classes, ts.log(model)), axis=1)))

Create optimizer:

    optimizer = ts.train.GradientDescentOptimizer(learning_rate=0.01).minimize(J)

Create placeholder inputs:

	feed_dict = {

		training_features: my_training_features,

		training_classes: my_training_classes

	}

Create session:

	session = ts.Session()

Train:

	for step in range(100):

		loss_value = session.run(loss, feed_dict)

		if step % 10 == 0:

			print("Step:", step, " Loss:", loss_value)

		session.run(optimizer, feed_dict)

Retrieve model parameters:

	weights_value = session.run(weigths)

	biases_value = session.run(biases)

Check out the `examples` directory for more.