Visdom: how to plot multiple line seperately on one window

[lixinsu]

I use visdom to plot train loss, I start many processes in different configurations to train model, and I want to plot multiple curve of train loss in one window for comparing different model to choose the best model parameters,Could I use visdom to do this?

[JackUrb]

Hi @lixinsu - do you mean you want to plot multiple lines separately in different plots in the same window? (something like matlab's subplots?) Or plot multiple traces on the same plot in a window?

We don't support the former currently, but to get the desired behavior for that I'd suggest using multiple windows with a tag in the titles, this way you can filter by that tag and see them all.

To plot multiple curves on the same axes you can determine the window name you want to use, and then use viz.line(... win=<window_name>, update='append', name=<trace_name>) for each trace.

Alternately, after https://github.com/facebookresearch/visdom/pull/283 is in you can plot all the relevant data for each experiment in its own env and then use the compare features to show the data on the same plot without needing to plot from different places into the same window.

[lixinsu]

thanks, I just ask for ploting multiple curves on the same axes, but I found that the code
below
viz = Visdom(port=8093)

win = viz.line(
X=np.column_stack([np.arange(0, 1) for i in range(10)]),
Y=np.column_stack([np.arange(0, 1) for i in range(10)]),
win="test"
)

viz.line(
X=np.arange(1, 38),
Y=np.random.randn(37),
win="test",
name='6',
update='append',
)

viz.line(
X=np.arange(1, 38),
Y=np.random.randn(37),
win="test",
name='11',
update='append',
)

generate the axes as following

the second line ,I name it as 11, but in the graph it names 1?

[lixinsu]

is it a bug ,or I make a mistake in the usage

[JackUrb]

Are you using the pypi version of visdom or the one installed from source here? The naming issue is fixed in version 0.1.7.1

[JackUrb]

That version hasn't been released yet as it breaks some behavior for proxies and I'm waiting to fix it before I release. You can use easy_install . or python setup.py install after cloning the git repo to install from source.

[lixinsu]

I use visdom (0.1.7) installed from master branch

[lixinsu]

thanks

[lixinsu]

I install 0.1.7.2 from master branch, but the program still give the above result

[JackUrb]

As in it's giving the second trace the name "1"? What happens if you enable a legend? Can you try swapping the order of the appends and seeing if you get "11" and "1"?