menu

neptune-community

A place where neptune.ai users and developers come together to make things work

Channels
Chat
view-forward
# All channels
view-forward
# General
view-forward
# Blog
view-forward
# Bugs
view-forward
# Feature requests
view-forward
# Random
view-forward
# What's new
view-forward
Team

neptune-community watercooler

July 11, 2019 at 12:49pm
Show previous messages

December 14, 2019 at 3:19pm
You can log metrics for precison_class1, precision_class2 ... to separate channels and later combine them into one channel via chart sets. I've done something like that in chart set auc with channels train_iter_auc valid_iter_auc. https://ui.neptune.ml/neptune-ml/credit-default-prediction/e/CRED-80/charts Does that help?
Edited
  • reply
  • like
Thanks . That can work but it's very tedious as I have 35 classes. Would make more sense to add a "series" param to log_metric e.g. neptune.log_metirc(metric='AP', series='cat', value=0.8)
  • reply
  • like
It is a good proposal. As a matter of fact we are actually working on chart sets right now. I think may talk to about that soon.
  • reply
  • like
There is an alternative option.
You can fetch your experiment via query api. Create charts and log it. Something like:
from neptune.sessions import Session
project = Session().get_project('neptune-ml/credit-default-prediction')
exp = project.get_experiments(id=['CRED-55'])[0]
fig, ax = plt.subplots()
for name in class_names:
df = exp.get_numeric_channels_values('precision_{}'.format(name))
ax.plot(df.x, df['precision_{}'.format(name)], label=name)
plt.legend()
fig.savefig('precision_per_class.png')
exp.log_image('precision_per_class', `precision_per_class.png')
There is also send_figure helper method in neptune-contrib to make it a bit simpler.
Edited
  • reply
  • like
I hope this helps
  • reply
  • like

January 14, 2020 at 3:50pm
Hi I am new in neptune. Now I wonder can we create 3d plots in neptune? for example, I have a 2D position (x, y) and 'z' which is the result of my network.
  • reply
  • like

January 15, 2020 at 7:32pm
Hi, I have a quick question about running xgboost models in neptune using R. For params, can I say model = "xgbTree" and metric = 'tweedie-nloglik@1.2'? The experiment always failed. Any examples of this kind of algorithm that I can take a look? Thanks!
  • reply
  • like
Hi, I have a quick question about running xgboost models in neptune using R. For params, can I say model = "xgbTree" and metric = 'tweedie-nloglik@1.2'? The experiment always failed. Any examples of this kind of algorithm that I can take a look? Thanks!
Could you please paste the code that produces errors? Or the error message itself?
  • reply
  • like
I just noticed there is a typo in R-support docs:
params = list(metric="Accuracy",
tuneLength=100,
model="rf",
searchMethod="random",
cvMethod="repeatedcv",
cvFolds=2,
cvRepeats=1)
# Create experiment
neptune$create_experiment(params=paramsd)
it should be
neptune$create_experiment(params=params)
Is that it by any chance?
  • reply
  • like

January 16, 2020 at 5:00pm
Could you please paste the code that produces errors? Or the error message itself?
Hi Probably not because of the typo. I used neptune$create_experiment(name='training on mtcars', params=params, properties=list( data_version=digest(dataset)), tags=c('mtcars', 'xgboost'), upload_source_files=list('train_random_forest.R') instead of neptune$create_experiment(params=params)
The code doesn't produce any error in R, but the experiment failed on neptune, where can I find the error details on neptune?
Edited
  • reply
  • like

January 17, 2020 at 1:18pm
hey thanks for reaching out
  • reply
  • like
You can look for an error message in the stdout and stderr in the monitoring tab in the experiment. Take a look at the screen below
Edited
  • reply
  • like
  • reply
  • like
Hi Here is the screenshot in my monitoring tab. What does this error means?
  • reply
  • like
Hey -> on the screen you posted, there is (unfortunately) no signs of errors. Let's move further discussion to the direct conversation.
  • reply
  • like