Client library used to communicate with the ease.ml service.
This is the Python implementation of the ease.ml client.
This package is available on PyPI.
pip install easemlclient
Establishing a connection
To use the client API we first need to create a connection object that we will be using to target the running easeml instance. The connection must be inialized with a host name (here we use localhost) and either the API key or a username and password.
from easemlclient.model import Connection connection = Connection(host="localhost:8080", api_key="some-api-key")
Then we can query all the running jobs. To do that we need to create a
JobQuery instance which we use to specify the parameters of our query. For example, we can query all completed jobs. To get the result we call the
run() method of the query object and pass the connection instance.
from easemlclient.model import JobQuery query = JobQuery(status="completed") result, next_query = query.run(connection)
The result will contain a list of
Job objects taht satisfy our query. Results are paginated to limit the size of each request. If there are more pages to be loaded, then the
next_query variable will contain a
JobQuery instance that we can run and return the next page. The full pattern for loading all jobs is the following:
from easemlclient.model import JobQuery result, query = , JobQuery(status="completed") next_result, next_query = , query while next_query is not None: next_result, next_query = query.run(connection) result.extend(next_result)
We can take the first completed job and get a list of its tasks.
job = result tasks = job.tasks
tasks list actually contains "shallow" instances of the
Task class. This means that each instance contains only the task's
id field and no other fields. This is normal because the
Job object has only references to tasks, not entire tasks. To get a full version of a task given a "shallow" instance, we use the
task = tasks.get(connection)
Querying Specific Objects
Task object can also be used to query tasks by their ID. We simply create a new "shallow" instance using a task ID and call the
from easemlclient.model import Task task = Task(id="some-task-id").get(connection)
We have the ability to create certain objects, such as
Job. We do this by initializing an instance of that object, assigning values to relevant fields and calling the
post() method. Here is an example of creating a dataset object along with uploading of a dataset.
from easemlclient.model import Dataset, DatasetSource, DatasetStatus dataset = Dataset.create(id="test_dataset_1", source=DatasetSource.UPLOAD, name="Test Dataset 1").post(connection) with open("test_dataset_1.tar", "rb") as f: dataset.upload(connection=connection, data=f) # Once the dataset upload finishes, we need to update the status of the dataset to "transferred". dataset.status = DatasetStatus.TRANSFERRED # Once we assign values to fields, we use the patch command # to apply updates to the dataset object on the server. dataset.patch(connection)
Starting a new training Job and monitoring it
Here we show a slightly more complex example that demonstrates how to start a model selection and tuning job given a previously uploaded dataset.
We will first fetch the dataset object in order to be able to access its schema.
from easemlclient.model import Dataset dataset = Dataset(id="test_dataset_1").get(connection)
Then we query all models that are applicable to the given dataset. We use the
ModuleQuery class for this.
from easemlclient.model import ModuleQuery, ModuleType query = ModuleQuery(type=ModuleType.MODEL, status=ModuleStatus.ACTIVE, schema_in=dataset.schema_in, schema_out=dataset.schema_out) # We assume that the result does not contain more than one page. models, _ = query.run(connection)
We do the same for objectives.
from easemlclient.model import ModuleQuery, ModuleType query = ModuleQuery(type=ModuleType.OBJECTIVE, status=ModuleStatus.ACTIVE, schema_in=dataset.schema_in, schema_out=dataset.schema_out) objectives, _ = query.run(connection) # We will simply pick the first objective here. objective = objectives
Then we are ready to create a job.
from easemlclient.model import Job job = Job(dataset=dataset, objective=objective, models=models, max_tasks=20).post(connection)
max_tasks we specify the number of tasks to run before a job's status will become
completed. We can keep querying the job to check the status.
from time import sleep from easemlclient.model import JobStatus while job.get(connection).status != JobStatus.COMPLETED: time.sleep(10)
Once the job is completed, we can get the task with the best result.
from easemlclient.model import TaskQuery, ApiQueryOrder tasks, _ = TaskQuery(job=job, order_by="quality", order=ApiQueryOrder.DESC).run(connection) best_task = tasks.get(connection)
Finally, we can download the Docker image of the best task and save it as a tar file.
image = best_task.get_image(connection) open("/output/path/to/image.tar", "wb").write(image)
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size easemlclient-0.1.4-py3-none-any.whl (18.1 kB)||File type Wheel||Python version py3||Upload date||Hashes View hashes|
|Filename, size easemlclient-0.1.4.tar.gz (14.1 kB)||File type Source||Python version None||Upload date||Hashes View hashes|
Hashes for easemlclient-0.1.4-py3-none-any.whl