Lab documentation

Appearance

Upgrade guide to version 3.0.0 ​

The release of version 3.0.0 of pyhectiqlab package comes with many new features. In addition, we have conserved much of the old API, making the upgrade as smooth as possible. In this section, we show how to migrate your code.

Upgrading ​

To upgrade the pyhectiqlab package, use pip and specify the latest versions >=3.0.0:

bash
pip install --upgrade pyhectiqlab==3.0.0

Run ​

To create runs, you can simply replace the name argument by title:

python
from pyhectiqlab import Run

run = Run(
    title="My first run", 
    name="My first run", 
    project="lab/demo",
)
python
from pyhectiqlab import Run

run = Run(
    name="My first run",
    project="lab/demo",
)
python
from pyhectiqlab import Run

run = Run(
    title="My first run",
    project="lab/demo",
)

Also, note that you can now remove the project argument from the run creation, as it can be fixed externally using the Project namespace or specifying using the environment variable HECTIQ_LAB_PROJECT:

python
from pyhectiqlab import Project, Run

Project.set("lab/demo") 
# or
os.environ["HECTIQ_LAB_PROJECT"] = "lab/demo"
run = Run(
    title="My first run", 
    name="My first run", 
    project="lab/demo", 
)
python
from pyhectiqlab import Project, Run

run = Run(
    name="My first run",
    project="lab/demo",
)
python
import os
from pyhectiqlab import Project, Run

Project.set("lab/demo")
# or
os.environ["HECTIQ_LAB_PROJECT"] = "lab/demo"

run = Run(title="My first run")

Previously, the runs were identified using a universally unique identifier (UUID) system, where the UUIDs were given by the name of the runs. We changed this system to give less importance to the name of the run, hence we now work with a simple integer rank ID system. Now, to retrieve a run, you now have to specify the rank of the run:

python
from pyhectiqlab import Run
from pyhectiqlab import Project 
Project.set("lab/demo") 
retrieved_run = Run(
    rank=14, 
    name="An old run", 
    project="lab/demo", 
)
python
from pyhectiqlab import Run

retrieved_run = Run(
    name="An old run",
    project="lab/demo",
)
python
from pyhectiqlab import Run, Project

Project.set("lab/demo")
retrieved_run = Run(rank=14)

You can find the rank ID of a specific run on the web application.

WARNING

The Run class is now a singleton object, hence all instances of the class share the same information at all times. As a result, you cannot initialize multiple independent runs.

Config ​

The configs have had a facelift in the last upgrade. We change the run API for retrieving configs, which now must be performed through the Run class:

python
from pyhectiqlab import Run, Config

config = Config.download(run="My run", project="lab/demo") 
config = Run.retrieve_config(run=14, project="lab/demo")
python
from pyhectiqlab import Config

config = Config.download(run="My run", project="lab/demo")
python
from pyhectiqlab import Run

config = Run.retrieve_config(run=14, project="lab/demo")

Managing versions ​

The version management system has been simplified to the bare minimum for you to track the git repos that are attached to your experiments. To push all the system and python related information and attach it to a given run, simply run the track_versions.

python
from pyhectiqlab import Run

run = Run(
    title="My run", 
    name="My run", 
    project="lab/demo"
)
run.track_version(["my_repos"]) 
run.add_package_repo_state("my_repos") 
run.add_package_versions(globals())
python
from pyhectiqlab import Run

run = Run(
    name="My run", 
    project="lab/demo"
)
run.add_package_repo_state("my_repos") 
run.add_package_versions(globals())
python
from pyhectiqlab import Run

run = Run(
    title="My run",
    project="lab/demo"
)
run.track_versions(["my_repos"])

Logs and messages ​

As of yet, the logs and messages system have been removed from version 3.0.0. We recommend using standard logging loggers. We will reimplement those features in future versions.

To track the logs of a run, use the steps system. For example:

python
import logging
logger = logging.getLogger()
from pyhectiqlab import Step, Run
run = Run(title="My run", project="lab/demo")

run.add_logger() 

with Step(name="Step 1") as step:  
    logger.info("This message is logged.")
python
import logging
logger = logging.getLogger()
from pyhectiqlab import Step, Run
run = Run(title="My run", project="lab/demo")
run.add_logger()
python
import logging
logger = logging.getLogger()
from pyhectiqlab import Step, Run
run = Run(title="My run", project="lab/demo")
with Step(name="Step 1") as step:
    logger.info("This message is logged.")

Models and datasets ​

We have revamped the way datasets and models are uploaded and pushed to the lab. First, you no longer have to download or upload them using runs; it can be done using the functional or object-oriented API, as shown in the tutorial, or using the command-line interface.

python
from pyhectiqlab import Dataset 
from pyhectiqlab import Run 

Dataset.create( 
    name="my_dataset", 
    version="1.0.0", 
    source="path/to/the/data", 
    upload=True
) 

run = Run(name="upload of dataset", project="lab/demo") 
run.add_dataset( 
    source_path="path/to/the/data", 
    name="my_dataset", 
    version="1.0.0", 
    push_dir=True
)
python
from pyhectiqlab import Run

run = Run(name="upload of dataset", project="lab/demo")
run.add_dataset(
    source_path="path/to/the/data",
    name="my_dataset",
    version="1.0.0",
    push_dir=True
)
python
from pyhectiqlab import Run
from pyhectiqlab import Dataset

Dataset.create(
    name="my_dataset",
    version="1.0.0",
    source="path/to/the/data",
    upload=True
)
bash
hectiq-lab Dataset.create --source path/to/the/data --name my_dataset --version 1.0.0

A similar API is used for uploading models:

python
from pyhectiqlab import Model 
from pyhectiqlab import Run 

Model.create( 
    path="path/to/the/model/data", 
    name="my_model", 
    version="1.0.0", 
) 

run = Run(name="upload of model", project="lab/demo") 
run.add_mlmodel( 
    source_path="path/to/the/model/data", 
    name="my_model", 
    version="1.0.0", 
    push_dir=True
)
python
from pyhectiqlab import Run

run = Run(name="upload of model", project="lab/demo")
run.add_mlmodel(
    source_path="path/to/the/model/data",
    name="my_model",
    version="1.0.0",
)
python
from pyhectiqlab import Run
from pyhectiqlab import Model

Model.create(
    path="path/to/the/model/data",
    name="my_model",
    version="1.0.0"
)
bash
hectiq-lab Model.create --path path/to/the/model/data --name my_model --version 1.0.0

For downloading datasets and models, you can also use the Dataset and Model classes:

python
from pyhectiqlab import Model 
from pyhectiqlab import Run 

Model.download( 
    name="my_model", 
    version="1.0.0", 
    path="path/to/the/model/data", 
) 

run = Run(name="upload of model", project="lab/demo") 
run.download_dataset( 
    name="my_model", 
    version="1.0.0", 
    save_path="path/to/the/model/data", 
)
python
from pyhectiqlab import Run
from pyhectiqlab.mlmodels import download_mlmodel


run = Run(name="upload of model", project="lab/demo")
path = run.download_mlmodel(
    name="my_model",
    version="1.0.0",
    save_path="path/to/the/model/data",
)
# or 
download_mlmodel(
    mlmodel_name="my_model",
    version="1.0.0",
    project_path="lab/demo",
    save_path="path/to/the/model/data",
)
python
from pyhectiqlab import Run
from pyhectiqlab import Model

Model.download(
    name="my_model",
    version="1.0.0",
    path="path/to/the/model/data",
)
bash
hectiq-lab Model.download --name my_model --version 1.0.0 --path path/to/the/model/data

Artifacts ​

The artifact has not changed.

Metrics ​

The metrics have not changed.