Tracking items

This example illustrates how skore can be used to track some items using their history, for example tracking some ML metrics over time.

Creating and loading the skore project

We create and load the skore project in the current directory:

import skore


my_project = skore.Project("my_project")

Tracking an integer

Let us store several integer values for a same item called my_int, each storage being separated by 0.1 second:

import time

my_project.put("my_int", 4)

time.sleep(0.1)
my_project.put("my_int", 9)

time.sleep(0.1)
my_project.put("my_int", 16)

We retrieve the history of the my_int item:

history = my_project.get("my_int", version="all", metadata=True)

We can print the details of the first version of this item:

print(history[0])

{'value': 4, 'date': '2025-02-21T10:38:09.807602+00:00', 'note': None}

Let us construct a dataframe with the values and last updated times:

import numpy as np
import pandas as pd

list_value, list_created_at, list_updated_at = zip(
    *[(version["value"], history[0]["date"], version["date"]) for version in history]
)

df_track = pd.DataFrame(
    {
        "value": list_value,
        "created_at": list_created_at,
        "updated_at": list_updated_at,
    }
)
df_track.insert(0, "version_number", np.arange(len(df_track)))
df_track

	version_number	value	created_at	updated_at
0	0	4	2025-02-21T10:38:09.807602+00:00	2025-02-21T10:38:09.807602+00:00
1	1	9	2025-02-21T10:38:09.807602+00:00	2025-02-21T10:38:09.917174+00:00
2	2	16	2025-02-21T10:38:09.807602+00:00	2025-02-21T10:38:10.022275+00:00

Notice that the created_at dates are the same for all iterations because they correspond to the date of the first version of the item, but the updated_at dates are spaced by 0.1 second (approximately) as we used time.sleep(0.1) between each put().

We can now track the value of the item over time:

import plotly.express as px

fig = px.line(
    df_track,
    x="version_number",
    y="value",
    hover_data=df_track.columns,
    markers=True,
)
fig.update_layout(xaxis_type="category")
fig

Note

We can hover over the histories of the item to visualize the last update date for example.

Here, we focused on how to use skore’s tracking of history of items. But why track items?

• We could track some items such as machine learning scores over time to better understand which feature engineering works best.

• Avoid overwriting a useful metric by mistake. No results are can be lost.

• The last updated time can help us reproduce an iteration of a key metric.