Developer Center

Resources to get you started with Algorithmia

Python

Updated

Available on GitHub.

The Algorithmia Python client provides a native Python interface to the Algorithmia API, letting developers manage and call algorithms, work with data in object stores using Algorithmia Data Sources, and access other features of the Algorithmia platform.

This guide will cover setting up the client, calling an algorithm using direct user input, calling an algorithm that accesses data through Algorithmia Data Sources, and using Algorithmia’s Hosted Data service. For complete details about the Algorithmia API, please refer to the API Docs.

The code in this guide can be run directly in a Python interpreter or used in your own scripts.

Set Up the Client

The official client is available on PyPi, and can be installed with pip:

pip3 install algorithmia

If you need to install the client from source, please see the additional installation instructions in the client README.

To use the client you’ll need an API key, which Algorithmia uses for fine-grained authentication across the platform. For this example, we’ll use the default-key that was created along with your account, which has a broad set of permissions. Log in to Algorithmia and navigate to Home > API Keys to find your key, or read the API keys documentation for more information.

Once the client is installed, you can import it into your code and instantiate the client object:

import Algorithmia

# Authenticate with your API key
apiKey = "YOUR_API_KEY"
# Create the Algorithmia client object
client = Algorithmia.client(apiKey)

Calling an Algorithm

Algorithms take three basic types of input whether they are invoked directly through the API or by using a client library: strings, JSON, and binary data. In addition, individual algorithms might have their own I/O requirements, such as using different data types for input and output, or accepting multiple types of input, so consult the input and output sections of an algorithm’s documentation for specifics.

The first algorithm we’ll call is a demo version of the algorithm used in the Algorithm Development Getting Started guide, which is available at demo/Hello. Looking at the algorithm’s documentation, it takes a string as input and returns a string.

In order to call an Algorithm from Python, we need to first create an algorithm object. With the client already instantiated, we can run the following code to create an object:

algo = client.algo('demo/Hello') 

Then, we can use the .pipe() method to call the algorithm. We’ll provide out input as the argument to the function, and then print the output using the result attribute:

response = algo.pipe("Mr. Bond")
print(reponse.result)

Which should print the phrase, Hello Mr. Bond

JSON and Python

The Python client provides some ease of use abstractions for working with algorithms with JSON inputs and outputs. When passing a Python array or dict into the .pipe() function, the library will automatically serialize it to JSON. Algorithms will return a JSON type and the result field of the response will contain an array or dict, as appropriate.

Let’s look at an example using JSON and the nlp/LDA algorithm. The algorithm docs tell us that the algorithm takes a list of documents and returns a number of topics which are relevant to those documents. The documents can be a list of strings, a Data API file path, or a URL. We’ll call this algorithm using a list of strings, following the format in the algorithm documentation:

algoJSON = client.algo('nlp/LDA/1.0.0')
input = {
    "docsList": [
        "It's apple picking season",
        "The apples are ready for picking"
    ]
}
response = algoJSON.pipe(input)
print(reponse.result)

The output will be [{'picking': 2}, {'apple': 1}, {'apples': 1, 'ready': 1}, {'season': 1}], which is the list of relavent words and the number of occurrances.

You might have noticed that in this example we included a version number when instantiating the algorithm. Pinning your code to a specific version of the algorithm can be especially important in a production environment where the underlying implementation might change from version to version.

Request Options

The client exposes options that can configure algorithm requests. This includes support for changing the timeout or indicating that the API should include stdout in the response. In the following example, we set the timeout to 60 seconds and disabling stdout in the response:

algo.set_options(timeout=60, stdout=False)

You can find more details in API Docs > Invoke an Algorithm.

Error Handling

To be able to better develop across languages, Algorithmia has created a set of standardized errors that can be returned by either the platform or by the algorithm being run. In Python, API errors and Algorithm exceptions will result in calls to pipe throwing an AlgoException:

client.algo('util/whoopsWrongAlgo').pipe('Hello, world!')
# Algorithmia.algo_response.AlgoException: algorithm algo://util/whoopsWrongAlgo not found

You can read more about Error Handling in the Algorithm Development section of the dev center.

Limits

Your account can make up to 80 Algorithmia requests at the same time (this limit can be raised if needed).

Working with Algorithmia Data Sources

For some algorithms, passing input to the algorithm at request time is sufficient, while others might have larger data requirements or need to preserve state between calls. Application developers can use Algorithmia’s Hosted Data to store data as text, JSON, or binary, and access it via the Algorithmia Data API.

The Data API defines connectors to a variety of storage providers, including Algorithmia Hosted Data, Amazon S3, Google Cloud Storage, Azure Storage Blobs, and Dropbox. After creating a connection in Data Sources, you can use the API to create, update, and delete directories and files and manage permissions across providers by making use of Data URIs in your code.

In this example, we’ll upload an image to Algorithmia’s Hosted Data storage provider, and use the dlib/FaceDetection algorithm to detect any faces in the image. The algorithm will create a new copy of the image with bounding boxes drawn around the detected faces, and then return a JSON object with details about the dimensions of the bounding boxes and a URI where you can download the resulting image.

Create a Data Collection

The documentation for “Face Detection” says that it takes a URL or a Data URI of the image to be processed, and a Data URI where the algorithm can store the result. We’ll create a directory to host the input image, then update its permissions so that it’s publicly accessible:

from Algorithmia.acl import ReadAcl, AclType
# Instantiate a DataDirectory object, set your data URI and call create
img_directory = client.dir("data://YOUR_USERNAME/img_directory")
# Create your data collection if it does not exist
if img_directory.exists() is False:
    img_directory.create()
# Change permissions on your data collection to public
img_directory.update_permissions(ReadAcl.public)

Instead of your username you can also use ‘.my’ when calling algorithms. For more information about the ‘.my’ pseudonym check out the Hosted Data Guide.

Upload Data to your Data Collection

Now we’re ready to upload an image file for processing. For this example, we’ll use this photo of a group of friends. Download the image and save it locally as friends.jpg.

Next, create a variable that holds the location where you would like to upload the image as a URI:

img_file = "data://.my/img_directory/friends.jpg"

Then upload your local file to the data collection using the .putFile() method:

if client.file(img_file).exists() is False:
    # Upload local file
    client.file(img_file).putFile("/your_local_path_to_file/friends.jpg")

This endpoint will replace a file if it already exists. If you wish to avoid replacing a file, check if the file exists before using this endpoint.

Confirm that the file was created by navigating to Algorithmia’s Hosted Data Source and finding your data collection and file.

You can also upload your data through the UI on Algorithmia’s Hosted Data Source. For instructions on how to do this go to the Hosted Data Guide.

Call the Algorithm

Once the file has been uploaded, you are ready to call the algorithm. Create the algorithm object, then pass the required inputs. An image URI (which is stored in img_file in the code above), a URI for the image output, and the size of the pixels to be used, to algo.pipe():

# Create the algorithm object
algoCV = client.algo('dlib/FaceDetection')
input = [
    "data://.my/img_directory/friends.jpg",
    "data://.algo/temp/people_detected.png"
]
# Invoke algorithm with error handling
try:
    # Get the summary result of your file's contents
    response = algoCV.pipe(img_file)
except Exception as error:
    # Algorithm error if, for example, the input is not correctly formatted
    print(error)

Once the algorithm has completed, response.result will contain information about the dimensions of the bounding boxes for any detected faces, the size of the image, and the URI for the resulting file, which you can then download (or provide as input to another algorithm in a pipeline).

Algorithms can create and store data in folders named with the algorithm name in the Algorithm Data collection. To access this folder from within an executing algorithm, the .algo shortcut can be used, as in the input example above. When accessing data from a client context, the algorithm author and name can be used along with the .algo shortcut to download data, in the format data://.algo/author/algoName/folder/fileName.

Download the resulting file

The URI included in the algorithm output uses the .algo shortcut, so we’ll need to modify it slighly to download the file by adding the algorithm name and author:

output_uri = response.result['images'][0]['output']
download_uri = output_uri[:13] + `dlib/FaceDetection` + output_uri[13:]

Verify that the file that you want to download exists, and try downloading it to a new local file:

# Download the file
if client.file(download_uri).exists() is True:
    local_file = client.file(download_uri).getFile()

This copies the file from your data collection and saves it as a file on your local machine, storing the filename in the variable local_file.

Alternately, if you just need the binary content of the file to be stored in a variable, you can retrieve the remote file’s content without saving the actual file:

# Download contents of file as a string
if client.file(download_uri).exists() is True:
    image_data = client.file(download_uri).getBytes()

This will get the image as binary data, saving it to the variable image_data, which might be useful when writing algorithms that are part of an image processing pipeline. For more image-manipulation tutorials, see the Computer Vision Recipes.

If the file was text (an image, etc), you could instead use the function .getString() to retrieve the file’s content as a string. For more methods on how to get a file using the Data API from a data collection go to the API Specification.

Additional Functionality

In addition to the functionality covered in this guide, the Python Client Library provides a complete interface to the Algorithmia platform, including managing algorithms, administering organizations, and working with source control. You can also visit the API Docs to view the complete API specification.

Next Steps

If you’re a data scientist or developer who will be building and deploying new algorithms, you can move on to the Algorithm Development > Getting Started guide.