Neptune UI automatically proposes columns based on what is different between Runs.
Suggested columns are the right-most columns in the dashboard.
You can:
add a suggested column by clicking on a + icon
get a new column suggestion by clicking on a - icon
Suggested columns in Neptune
Adding aggregate columns for series
For Fields where you logged a series of values, you can choose an aggregate function and add it to the Runs table.
This is useful for training losses where you want to show the best (min or max) value of a loss during training.
You can display:
MIN: minimal value
MAX: maximal value
AVERAGE: average or mean value
LAST: the value that was logged last
VARIANCE: variance calculated on the logged values
Searching and filtering Runs
You can search and filter Runs in Neptune based on anyField.
To filter Runs, start typing the Fields' name into the filter box and select the field you are looking for. Select the most fitting operator and provide reference value if required by the chosen operator.
You can enter queries using the keyboard only. Use up and down arrows to move around and Enter to select and accept.
Each field type has a set of specific operators that lets you specify what you are looking for.
You can:
use both the full path of the Field and a custom column name to search columns
combine many filters together to narrow the search as needed
Examples:
Find models small enough to be used on mobile (model/size <= 50 MB) that have a decent accuracy (test/acc > 0.90)
Check all runs from this month (sys/creation_time > YYYY.MM.DD) that failed during training (sys/failed = True)
Find all my runs (sys/owner = "giulia") from my current task (sys/tags ∋ "exploration")
Search history
You can access all your previous search queries with the Recent searches.