Skip to content

Discovering and Reusing Past Work

Once connected to an artifact store (it can be an individual or shared one), we can query existing artifacts, like so:

lineapy.artifact_store()

which would print a list looking as the following:

iris_preprocessed:0 created on 2022-09-29 01:22:39.612871
iris_preprocessed:1 created on 2022-09-29 01:22:41.336159
iris_preprocessed:2 created on 2022-09-29 01:22:43.511112
iris_model:0 created on 2022-09-29 01:22:45.381132
iris_model:1 created on 2022-09-29 01:22:46.786414
iris_model:2 created on 2022-09-29 01:22:47.990517
iris_model:3 created on 2022-09-29 01:22:49.366484
toy_artifact:0 created on 2022-09-29 01:22:50.189060
toy_artifact:1 created on 2022-09-29 01:22:50.676276
toy_artifact:2 created on 2022-09-29 01:22:51.084704

Each line contains three pieces of information about an existing artifact: its name, version, and time of creation. Hence, for an artifact named iris_model, we have four versions created at different times.

Now, say we are interested in reusing the first version of this artifact. We can retrieve the desired artifact as follows:

model_artifact = lineapy.get("iris_model", version=0)

Note that what has been retrieved and saved in model_artifact is not the model itself; it is the model artifact, which contains more than the model itself, e.g., the code that was used to generate the model. Hence, to resuse the model, we need to extract the artifact's value:

model = model_artifact.get_value()

However, we actually do not fully know how to reuse this model as we are missing the memory (or knowledge, if the artifact was created by someone else) of its context such as input details. Thankfully, the artifact also stores the code that was used to generate its value, so we can check it out:

print(data_artifact.get_code())

which prints:

import lineapy
from sklearn.linear_model import LinearRegression

art_df_processed = lineapy.get("iris_preprocessed", version=2)
df_processed = art_df_processed.get_value()
mod = LinearRegression()
mod.fit(
    X=df_processed[["petal.width", "d_versicolor", "d_virginica"]],
    y=df_processed["sepal.width"],
)

With this, we now know the source and shape of the data that was used to train this model, which enables us to adapt and reuse the model in our context. Specifically, we can check out the training data by loading the corresponding artifact, like so:

art_df_processed = lineapy.get("iris_preprocessed", version=2)
df_processed = art_df_processed.get_value()
print(df_processed)

Based on the values in the data, we would have a more concrete understanding of the model and its job, which would enable us to make new predictions, like so:

import pandas as pd

# Create data to make predictions on
df = pd.DataFrame({
    "petal.width": [1.3, 5.2, 0.3, 1.5, 4.9],
    "d_versicolor": [1, 0, 0, 1, 0],
    "d_virginica": [0, 1, 0, 0, 1],
})

# Make new predictions
df["sepal.width.pred"] = model.predict(df)

This example illustrates the benefit of LineaPy's unified storage framework: encapsulating both value and code as well as other metadata, LineaPy's artifact store enables the user to explore the history and relations among different works, hence rendering them more reusable.


Was this helpful?

Help us improve docs with your feedback!