Python client library for accessing the Algorithmia API For API documentation, see the PythonDocs

The official Algorithmia python client is available on PyPi. Install it with pip:

pip install algorithmia

Build algorithmia client wheel:

python setup.py bdist_wheel

Install a wheel manually:

pip install --user --upgrade dist/algorithmia-*.whl

First, create an Algorithmia client and authenticate with your API key:

import Algorithmia

apiKey = '{{Your API key here}}'
client = Algorithmia.client(apiKey)

Now you're ready to call algorithms.

The following examples of calling algorithms are organized by type of input/output which vary between algorithms.

Note: a single algorithm may have different input and output types, or accept multiple types of input, so consult the algorithm's description for usage examples specific to that algorithm.

Call an algorithm with text input by simply passing a string into its pipe method. If the algorithm output is text, then the result field of the response will be a string.

algo = client.algo('demo/Hello/0.1.1')
response = algo.pipe("HAL 9000")
print response.result    # Hello, world!
print response.metadata  # Metadata(content_type='text',duration=0.0002127)
print response.metadata.duration # 0.0002127

Call an algorithm with JSON input by simply passing in a type that can be serialized to JSON: most notably python dicts and arrays. For algorithms that return JSON, the result field of the response will be the appropriate deserialized type.

algo = client.algo('WebPredict/ListAnagrams/0.1.0')
result = algo.pipe(["transformer", "terraforms", "retransform"]).result
# -> ["transformer","retransform"]

Call an algorithm with Binary input by passing a byte array into the pipe method. Similarly, if the algorithm response is binary data, then the result field of the response will be a byte array.

input = bytearray(open("/path/to/bender.png", "rb").read())
result = client.algo("opencv/SmartThumbnail/0.1").pipe(input).result
# -> [binary byte sequence]

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

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.

from Algorithmia.algorithm import OutputType
response = client.algo('util/echo').set_options(timeout=60, stdout=False)
print response.metadata.stdout

Note: stdout=True is only supported if you have access to the algorithm source.

The Algorithmia client also provides a way to manage both Algorithmia hosted data and data from Dropbox or S3 accounts that you've connected to you Algorithmia account.

Create directories by instantiating a DataDirectory object and calling create():

client.dir("data://.my/foo").create()
client.dir("dropbox://somefolder").create()

Upload files by calling put on a DataFile object, or by calling putFile on a DataDirectory object.

foo = client.dir("data://.my/foo")
foo.file("remote_file").putFile("/path/to/myfile")
foo.file("sample.txt").put("sample text contents")
foo.file("binary_file").put(some_binary_data)

Note: you can instantiate a DataFile by either client.file(path) or client.dir(path).file(filename)

Download files by calling getString, getBytes, getJson, or getFile on a DataFile object:

foo = client.dir("data://.my/foo")
sampleText = foo.file("sample.txt").getString()  # String object
binaryContent = foo.file("binary_file").getBytes()  # Binary data
tempFile = foo.file("myfile").getFile()   # Open file descriptor

Delete files and directories by calling delete on their respective DataFile or DataDirectory object. DataDirectories take an optional force parameter that indicates whether the directory should be deleted if it contains files or other directories.

foo = client.dir("data://.my/foo")
foo.file("sample.txt").delete()
foo.delete(true) // true implies force deleting the directory and its contents

Iterate over the contents of a directory using the iterated returned by calling list, files, or dirs on a DataDirectory object:

foo = client.dir("data://.my/foo")

# List files in "foo"
for file in foo.files():
    print file.path " at URL: " + file.url + " last modified " + file.last_modified

# List directories in "foo"
for file in foo.dirs():
    print dir.path " at URL: " + file.url

# List everything in "foo"
for entry in foo.list():
    print entry.path " at URL: " + entry.url

Directory permissions may be set when creating a directory, or may be updated on already existing directories.

from Algorithmia.acl import ReadAcl, AclType
foo = client.dir("data://.my/foo")
# ReadAcl.public is a wrapper for Acl(AclType.public) to make things easier
foo.create(ReadAcl.public)

acl = foo.get_permissions()  # Acl object
acl.read_acl == AclType.public  # True

foo.update_permissions(ReadAcl.private)
foo.get_permissions().read_acl == AclType.private # True

The main backwards incompatibility between 0.9.x and 1.0.0 is the result of an algorithm call. In 0.9.x the result of an algorithm call is just the algorithm's output (which is not the full spec returned by the API)

result = client.algo('util/echo').pipe('Hello, world!')
print result   # Hello, world!

In 1.0.x the result of an algorithm matches the API specification. The algorithm's output is nested in an attribute 'result', and metadata can be accessed via the 'metadata' attribute.

result = client.algo('util/echo').pipe('Hello, world!')
print result.result     # Hello, world!
print result.metadata   # content_type, duration etc

Aside from that you should be able to drop in the newest version of the client. Another advantage of using the newest client is full access to the entire Data API specification.

export ALGORITHMIA_API_KEY={{Your API key here}}
cd Test
python acl_test.py
python algo_test.py
python datadirectorytest.py
python datafile_test.py
python utiltest.py