Welcome to Gradientzoo’s documentation!¶

Contents:

Using Keras with Gradientzoo¶

This doc will demonstrate how to save and load your trained neural network in Python using Keras.

Initialization¶

from gradientzoo import KerasGradientzoo

This is the client that you’ll create to save and load your model to the service.

param model_slug:
param username:	The gradientzoo username of the project you want to connect to
	The slug (short url-safe name) of the model
param api_base:	Optional URL prefix to specify where the gradientzoo API is located, which is especially useful if you are running your own instance of gradientzoo
param auth_token_id:
	Your authentication token as provided by the service
param default_dir:
	Optional When a download directory is asked for later, this is what it will default to if you pass None

Note: For convenience and aesthetics, you can pass both the username and model_slug as username/model_slug to the username parameter.

# Examples of how to instantiate the client
zoo = KerasGradientzoo('exampleuser/modelslug')
zoo = KerasGradientzoo('exampleuser', 'modelslug', auth_token_id='12345af')
zoo = KerasGradientzoo('exampleuser/modelslug', api_base='https://api2.gradientzoo.com')

Examples

The next few examples will assume zoo, an instantiated KerasGradientzoo client:

from gradientzoo import KerasGradientzoo
zoo = KerasGradientzoo('exampleuser/modelslug')

Loading a Model¶

If you want to load a model from the remote cloud, that can be done by calling the load method on the client.

returns file_model:
param model:	The Keras model you want to load the weights into
param filename:	Optional The filename of the thing you would like to download
param id:	Optional If there is a specific version of the file you want to download, provide the file id (as seen on the website) here
param dir:	Optional The directory to download the file to
	A dictionary with metadata about the file you downloaded

# Example of how to load a model
zoo.load(your_keras_model)

Saving a Model¶

If you want to save your moel to the remote cloud, simply call the upload_file method on the client.

param model:	The Keras model you want to save the weights for
param filename:	Optional The name of the file you’re uploading
param metadata:	Optional A bag of key/value pairs to be associated with this file (must be JSON encodable)
param dir:	Optional The temporary directory to save the model weights to before uploading them

# Example of how to save a model
zoo.save(your_keras_model)

Creating a Keras callback that will save automatically¶

Often times you don’t want to manage saving the model yourself. To make this easier, we provide a Keras callback that will save every N batches or every N epochs. To get a callback, call the make_callback method on the client.

param after_batches:
param model:	The Keras model you want to save the weights for
	Optional The number of batches after which to save
param after_epochs:
	Optional The number of epochs after which to save (defaults to 1)

# Example of how to automatically save a model using a callback
zoo_callback = zoo.make_callback(your_model)
your_model.fit(X_train, Y_train, # ...
               callbacks=[zoo_callback])

Using Tensorflow with Gradientzoo¶

This doc will demonstrate how to save and load your trained neural network in Python using Tensorflow.

Initialization¶

from gradientzoo import TensorflowGradientzoo

This is the client that you’ll create to save and load your model to the service.

param model_slug:
param username:	The gradientzoo username of the project you want to connect to
	The slug (short url-safe name) of the model
param api_base:	Optional URL prefix to specify where the gradientzoo API is located, which is especially useful if you are running your own instance of gradientzoo
param auth_token_id:
	Your authentication token as provided by the service
param default_dir:
	Optional When a download directory is asked for later, this is what it will default to if you pass None

Note: For convenience and aesthetics, you can pass both the username and model_slug as username/model_slug to the username parameter.

# Examples of how to instantiate the client
zoo = TensorflowGradientzoo('exampleuser/modelslug')
zoo = TensorflowGradientzoo('exampleuser', 'modelslug', auth_token_id='12345af')
zoo = TensorflowGradientzoo('exampleuser/modelslug', api_base='https://api2.gradientzoo.com')

Examples

The next few examples will assume zoo, an instantiated TensorflowGradientzoo client:

from gradientzoo import TensorflowGradientzoo
zoo = TensorflowGradientzoo('exampleuser/modelslug')

Loading a Model¶

If you want to load a model from the remote cloud, that can be done by calling the load method on the client.

returns file_model:
param session:	The Tensorflow session you want to load the weights into
param filename:	Optional The filename of the thing you would like to download
param id:	Optional If there is a specific version of the file you want to download, provide the file id (as seen on the website) here
param dir:	Optional The directory to download the file to
	A dictionary with metadata about the file you downloaded

# Example of how to load a Tensorflow session
with tf.Session() as sess:
    # Load latest weights from Gradientzoo
    zoo.load(sess)

Saving a Model¶

If you want to save your moel to the remote cloud, simply call the upload_file method on the client.

param session:	The Tensorflow session you want to save the weights for
param filename:	Optional The name of the file you’re uploading
param metadata:	Optional A bag of key/value pairs to be associated with this file (must be JSON encodable)
param dir:	Optional The temporary directory to save the model weights to before uploading them

# Example of how to save a Tensorflow session
zoo.save(sess)

Using Lasagne with Gradientzoo¶

This doc will demonstrate how to save and load your trained neural network in Python using Lasagne.

Initialization¶

from gradientzoo import LasagneGradientzoo

This is the client that you’ll create to save and load your model to the service.

param model_slug:
param username:	The gradientzoo username of the project you want to connect to
	The slug (short url-safe name) of the model
param api_base:	Optional URL prefix to specify where the gradientzoo API is located, which is especially useful if you are running your own instance of gradientzoo
param auth_token_id:
	Your authentication token as provided by the service
param default_dir:
	Optional When a download directory is asked for later, this is what it will default to if you pass None

Note: For convenience and aesthetics, you can pass both the username and model_slug as username/model_slug to the username parameter.

# Examples of how to instantiate the client
zoo = LasagneGradientzoo('exampleuser/modelslug')
zoo = LasagneGradientzoo('exampleuser', 'modelslug', auth_token_id='12345af')
zoo = LasagneGradientzoo('exampleuser/modelslug', api_base='https://api2.gradientzoo.com')

Examples

The next few examples will assume zoo, an instantiated LasagneGradientzoo client:

from gradientzoo import LasagneGradientzoo
zoo = LasagneGradientzoo('exampleuser/modelslug')

Loading a Model¶

If you want to load a model from the remote cloud, that can be done by calling the load method on the client.

returns file_model:
param network:	The Lasagne network you want to load the weights into
param filename:	Optional The filename of the thing you would like to download
param id:	Optional If there is a specific version of the file you want to download, provide the file id (as seen on the website) here
param dir:	Optional The directory to download the file to
	A dictionary with metadata about the file you downloaded

# Example of how to load a Lasagne network
zoo.load(your_lasagne_network)

Saving a Model¶

If you want to save your moel to the remote cloud, simply call the upload_file method on the client.

param network:	The Lasagne network you want to save the weights for
param filename:	Optional The name of the file you’re uploading
param metadata:	Optional A bag of key/value pairs to be associated with this file (must be JSON encodable)
param dir:	Optional The temporary directory to save the model weights to before uploading them

# Example of how to save a Lasagne network
zoo.save(your_lasagne_network)

Using plain old Python with Gradientzoo¶

This doc will demonstrate how to save and load your trained neural network in Python. This is useful either for writing your own library integration, or for doing something outside of what the standard integrations expect.

Initialization¶

from gradientzoo import Gradientzoo

This is the client that you’ll create to save and load your model to the service.

param model_slug:
param username:	The gradientzoo username of the project you want to connect to
	The slug (short url-safe name) of the model
param api_base:	Optional URL prefix to specify where the gradientzoo API is located, which is especially useful if you are running your own instance of gradientzoo
param auth_token_id:
	Your authentication token as provided by the service
param default_dir:
	Optional When a download directory is asked for later, this is what it will default to if you pass None

Note: For convenience and aesthetics, you can pass both the username and model_slug as username/model_slug to the username parameter.

# Examples of how to instantiate the client
zoo = Gradientzoo('exampleuser/modelslug')
zoo = Gradientzoo('exampleuser', 'modelslug', auth_token_id='12345af')
zoo = Gradientzoo('exampleuser/modelslug', api_base='https://api2.gradientzoo.com')

Examples

The next few examples will assume zoo, an instantiated Gradientzoo client:

from gradientzoo import Gradientzoo
zoo = Gradientzoo('exampleuser/modelslug')

Loading a Model¶

If you want to load a model from the remote cloud, that can be done by calling the download_file method on the client.

param chunk_size:
param filename:	The filename of the thing you would like to download
param id:	Optional If there is a specific version of the file you want to download, provide the file id (as seen on the website) here
param dir:	Optional The directory to download the file to
	Optional The size of the chunks to read and flush to disk
returns file_model:
	A dictionary with metadata about the file you downloaded

# Example of how to download a file
filepath, file_model = zoo.download_file('model.npz')

with open(filepath, 'r') as f:
    data = f.read()

# Now load the data into your model however your framework or library works

Saving a Model¶

If you want to save your moel to the remote cloud, simply call the upload_file method on the client.

param filename:	The name of the file you’re uploading
param f:	A file object that can be read to upload to the server
param metadata:	Optional A bag of key/value pairs to be associated with this file (must be JSON encodable)

# Example of how to upload a file, first by saving it out from your library
filename = '/tmp/model.npz'
numpy.savez(filename, your_model_tensor)

# Then sending it up to gradientzoo
with open(filename, 'r') as f:
    zoo.upload_file(filename, f, {'loss': your_model_loss})

Welcome to Gradientzoo’s documentation!¶

Using Keras with Gradientzoo¶

Initialization¶

Loading a Model¶

Saving a Model¶

Creating a Keras callback that will save automatically¶

Using Tensorflow with Gradientzoo¶

Initialization¶

Loading a Model¶

Saving a Model¶

Using Lasagne with Gradientzoo¶

Initialization¶

Loading a Model¶

Saving a Model¶

Using plain old Python with Gradientzoo¶

Initialization¶

Loading a Model¶

Saving a Model¶

Indices and tables¶