Interactive Experiments Compare

Concatenate data from multiple experiment runs

Use the concat_experiments_on_channel helper function from neptune-contrib to do that.

[7]:
from neptunecontrib.api.utils import concat_experiments_on_channel

exp_df = concat_experiments_on_channel(experiments, channel_name)
exp_df.head()
[7]:
x unet_0 epoch_val iout loss id
0 0.0 0.612894 SAL-1134
1 1.0 0.679910 SAL-1134
2 2.0 0.690255 SAL-1134
3 3.0 0.707496 SAL-1134
4 4.0 0.713793 SAL-1134

Plot an interactive Altair chart.

Use channel_curve_compare from neptunecontrib.viz.

[ ]:
from neptunecontrib.viz.experiments import channel_curve_compare

channel_curve_compare(exp_df, width=400, heights=[50,200])

Note You may need to change the rendering method depending on your machine.

import altair as alt
alt.renderers.enable('jupyterlab')

Because Vega-Lite visualizations keep all the chart data in the HTML the visualizations can consume huge amounts of memory if not handled properly. That is why, by default the hard limit of 5000 rows is set to the len of dataframe. That being said, you can disable it by adding the following line in the notebook or code.

import altair as alt
alt.data_transformers.enable('default', max_rows=None)