Query single attribute values

Python package: neptune-query

To fetch configs, scores, or other single values from experiments, use the fetch_experiments_table() function. The returned data frame mimics the runs table of the web app.

If series attributes are included, only the last logged value is returned.

loss_and_config_df = nq.fetch_experiments_table(
    experiments=["kittiwake_week-1", "kittiwake_week-2", "kittiwake_week-3"],
    attributes=r"loss | config",
)

Sample output

                   loss  config/n_layer  config/learning_rate
experiment
kittiwake_week-1   0.22              12                  0.01
kittiwake_week-2   0.28              12                  0.02
kittiwake_week-3   0.24              12                  0.01

To specify experiments or attributes to include, pass a list of exact names or a regular expression to the experiments or attributes argument.

To provide more complex criteria or join multiple conditions together, use filter objects. For a guide, see Define filters.

Target runs instead of experiments

To specify runs by ID instead of experiments by name, use the fetch_runs_table() function from the runs module:

import neptune_query.runs as nq_runs


runs_df = nq_runs.fetch_runs_table(
    runs=["glowing-database-60704895-9ps29"],  # list of run IDs, or regex to match
    attributes=r"loss | config",
)

Sample output

                                          loss  config/n_layer  config/learning_rate
run
glowing-database-60704895-9ps29           0.02              12                  0.01