Developer Center

Resources to get you started with Algorithmia



Welcome to hosting your Tensorflow model on Algorithmia! This guide is designed as an introduction to hosting a Tensorflow model and publishing an algorithm even if you’ve never used Algorithmia before.


Before you get started hosting your model on Algorithmia there are a few things you’ll want to do first:

Train and save your model.

After training your Tensorflow model, you’ll want to save your variable checkpoints and the graph from your trained model so you can upload it to Algorithmia.

Create a Data Collection

Here you’ll want to create a data collection to host your graph and variable checkpoint data.

  • To use the Data API, log into your Algorithmia account and create a data collection via the Data Collections page.

  • Click on “Add Collection” under the “My Collections” section on your data collections page.

  • After you create your collection you can set the read and write access on your data collection. For more information check out: Data Collection Types

Create a data collection

Upload your Model into a Collection

Next, upload your pickled model to your newly created data collection.

  • Load model by clicking box “Drop files here to upload”

  • Note the path to your files: data://username/collections_name/pickled_model.pkl

Create a data collection

Create your Algorithm

Creating your algorithm is easy!

  • To add an algorithm, simply click “Add Algorithm” from the user profile icon.
  • Name your algorithm, select the language, choose permissions and make the code either open or closed source.

Note: There is also a checkbox for ‘Standard Execution Environment’ or ‘Advanced GPU’. For deep learning models you will want to check ‘Advanced GPU’.

Create your algorithm

Set your Dependencies

Now is the time to set your dependencies that your model relies on.

  • Click on the “Dependencies” button at the top right of the UI and list your packages under the required ones already listed and click “Save Dependencies” on the bottom right corner.

Set your dependencies

If you plan on using tensorflow with GPU support, make sure to use the tensorflow-gpu python package instead of the tensorflow one, with the version number 1.2.0. It can be written in the dependency file like this: tensorflow-gpu==1.2.0.

Load your Model

Here is where you load your graph and run your model which will be called by the apply() function. Our recommendation is to preload your model in a separate function before apply(). The reasoning behind this is because when your model is first loaded it can take some time to load depending on the file size. However, with all subsequent calls only the apply() function gets called which will be much faster since your model is already loaded!

Now to check out the code adapted from MNIST for Beginners tutorial from Tensorflow:

import Algorithmia
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf

client = Algorithmia.client()

def load_data():
    """Retrieve variable checkpoints and graph from user collection"""
    vc_uri = 'data://user_name/demos/variable_checkpoint_tensorflow.ckpt'
    checkpoint_file = client.file(vc_uri).getFile().name

    graph_uri = 'data://user_name/models/graph_model_tensorflow.pb'
    graph_file = client.file(graph_uri).getFile().name

    return (checkpoint_file, graph_file)
# Enable allow_growth if you still run into memory issues.
def generate_gpu_config(memory_fraction):
    config = tf.ConfigProto()
    # config.gpu_options.allow_growth = True
    config.gpu_options.per_process_gpu_memory_fraction = memory_fraction
    return config

# Get called once
saver = tf.train.Saver()
checkpoints, graph = load_data()

def inject_data(input):
    Finds the prediction and accuracy of digit image

    Prints accuracy and predictions on user input
    # Set your memory fraction equal to a value less than 1, 0.6 is a good starting point.
    # If no fraction is defined, the tensorflow algorithm may run into gpu out of memory problems.
    fraction = 0.6
    # Inject data into Tensor graph
    with tf.Session(graph=graph, config=generate_gpu_config(fraction)) as sess:
        # Load previously saved graph
        with tf.gfile.FastGFile(graph, 'rb') as f:
            graph_def = tf.GraphDef()
            tf.import_graph_def(graph_def, name='')
        # Map variables
        saver.restore(sess, checkpoints)
        y_ = sess.graph.get_tensor_by_name('Placeholder_1:0')
        y = sess.graph.get_tensor_by_name('Softmax:0')
        x = sess.graph.get_tensor_by_name('Placeholder:0')

        correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
        accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
        print(, feed_dict={
              x: input, y_: mnist.test.labels}))
        prediction = tf.argmax(y, 1)
        print(prediction.eval(feed_dict={x: input}))

def apply(input):
    Input would be an image file either from:

    data sources via using the Data API
    or as an http request using urllib
    output = inject_data(input)
    return output

If you are authoring an algorithm, avoid using the ‘.my’ pseudonym in the source code. When the algorithm is executed, ‘.my’ will be interpreted as the user name of the user who called the algorithm, rather than the author’s user name.

Publish your Algorithm

Last is publishing your algorithm. The best part of hosting your model on Algorithmia is that users can access it via an API that takes only a few lines of code to use! Here is what you can set when publishing your algorithm:

  • Set version permissions to public or private use

  • Set it to royalty free or set to per-call royalty

  • Set access permissions to have full access to the internet and ability to call other algorithms

If you want to have a better idea of how a finished tensorflow algorithm looks like, check out: InceptionNet

For more information and detailed steps: creating and publishing your algorithm

Publish your algorithm

That’s it for hosting your tensorflow model on Algorithmia!