See the attached image with grouped runs and range set to samples. The actual plots (faint lines) have a much higher granularity then the grouped plot. This decrease in granularity is particularly visible when x-axis is set to log-scale.
Hi @vyasnikhil96 thank you for writing in! The issue you’re seeing is related to a combination of bucketing and sampling that takes place in order to improve the UI performance. Indeed the issue is even more obvious when working in log scale.
Our engineers are actively working on this feature to allow users to dynamically adjust the total number of points. I have now increased the requests, and will keep you posted here once there are any updates regarding this!
Would this work for you? alternatively, you could download the data using our API after the run is finished, in order to create the chart externally and log the generated plot or image.
Yes I think adjusting the number of points would be a great solution with the caveat that sampling is based on whether the plot is in log scale or not. If it is in log scale points should be weighted by 1/step.
Hi @vyasnikhil96 thanks for the additional information. I have already reported this UI behavior you’re seeing to our engineers and we will reach out to you as soon as there are any updates.