Quickstart

To get a quick feel for OpsML, run the following code in a new terminal. The following uses Mlflow as a UI interface and local storage and sqlite.

Start Local Server

$ opsml-uvicorn-server

...
<span style="color: green;">INFO</span>:     [INFO] Started server process
<span style="color: green;">INFO</span>:     [INFO] Waiting for application startup

...
<span style="color: green;">INFO</span>:     [INFO] Application startup complete
<span style="color: green;">INFO</span>:     [INFO] Uvicorn running on http://0.0.0.0:8888

Next, open a new terminal and run the following python script. Make sure to set the OPSML_TRACKING_URI which tells OpsML where to log experiments.

Run Initial Python Script

export OPSML_TRACKING_URI=${YOUR_TRACKING_URI}

# imports
from sklearn.linear_model import LinearRegression
from opsml import (
    CardInfo,
    CardRegistries,
    DataCard,
    DataSplit,
    ModelCard,
    PandasData,
    SklearnModel,
)
from opsml.helpers.data import create_fake_data


info = CardInfo(name="linear-regression", repository="opsml", contact="user@email.com")
registries = CardRegistries()


#--------- Create DataCard ---------#

# create fake data
X, y = create_fake_data(n_samples=1000, task_type="regression")
X["target"] = y

# Create data interface
data_interface = PandasData(
    data=X,
    data_splits=[
        DataSplit(label="train", column_name="col_1", column_value=0.5, inequality=">="),
        DataSplit(label="test", column_name="col_1", column_value=0.5, inequality="<"),
    ],
    dependent_vars=["target"],
)

# Create and register datacard
datacard = DataCard(interface=data_interface, info=info)
registries.data.register_card(card=datacard)

#--------- Create ModelCard ---------#

# split data
data = datacard.split_data()

# fit model
reg = LinearRegression()
reg.fit(data["train"].X.to_numpy(), data["train"].y.to_numpy())

# create model interface
interface = SklearnModel(
    model=reg,
    sample_data=data["train"].X.to_numpy(),
    task_type="regression",  # optional
)

# create modelcard
modelcard = ModelCard(
    interface=interface,
    info=info,
    to_onnx=True,  # lets convert onnx
    datacard_uid=datacard.uid,  # modelcards must be associated with a datacard
)
registries.model.register_card(card=modelcard)

Opsml UI

Next, navigate to OPSML_TRACKING_URI and you should see the following:

Click on the linear-regression card and you should see the following:

Click on the DataCard button and you should see the following:

Download your model via CLI

Try downloading your model via the CLI:

opsml-cli download-model --name 'linear-regression' --version '1.0.0'

Here we are downloading the model with name linear-regression and version 1.0.0. You could also provide the uid of the model instead of the name and version. By default, the cli command will download objects to the models directory. You will see both the model joblib file as well as the model's associated metadata. If you'd wish to download the onnx version of the model, you can add the --onnx flag to the command.

opsml-cli download-model --name 'linear-regression' --version '1.0.0' --onnx