Forum Discussion
Scatterplot trend line does not go through data
Sure. Here is a table of log-linear data that I created with python:
| x | y |
| 1 | 1000 |
| 2 | 784.76 |
| 3 | 615.8482 |
| 4 | 483.293 |
| 5 | 379.269 |
| 6 | 297.6351 |
| 7 | 233.5721 |
| 8 | 183.2981 |
| 9 | 143.845 |
| 10 | 112.8838 |
| 11 | 88.58668 |
| 12 | 69.51928 |
| 13 | 54.55595 |
| 14 | 42.81332 |
| 15 | 33.59818 |
| 16 | 26.36651 |
| 17 | 20.69138 |
| 18 | 16.23777 |
| 19 | 12.74275 |
| 20 | 10 |
When I plot this in PowerBI with y on a linear scale, the trend line is correct. When I plot with y on a log scale, the trend line no longer goes through the data.
Linear scale:
Log scale:
The data were created so that the relationship is perfectly log-linear, and the line should go exactly through the datapoints. Here is a plot I made with the data in python with the y data plotted as log10(y):
Hi dtan ,
I can reproduce your scenario.
I build a table from your sample data and add a calculated column based y column, trend line works well as the result that your python code generated.
logY = LOG10([y])
Regards,
Xiaoxin Sheng
Thanks -- unfortunately with this fix (as in my Python demo) is that the y-axis is now in log-units.
The ideal scenario we're trying for is to have the y-axis in their original units (e.g., 1, 10, 100), but with their spacing on a log scale, and the trend line reflecting the linearized relationship *after* the log transform.
I can't find a clear way to plot the log-transformed datapoints with the values on the y-axis in the original scale, but with the trend line showing the now linearized relationship in PowerBI.
In scripting-based plotting tools you can generally apply values to the y-axis for display purposes (e.g., have 1, 10, and 100, equispaced along the y-axis), while plotting the log-transformed data with the appropriate trendline. Is this possible in PowerBI? Is there a hacky workaround that can be done to display the y-axis in the original units, while plotting the log transformed data with trend line?
More generally, it seems less than ideal to have a trendline *not* go through data after the axis is changed to a log scale (independent of my particular problem -- this is a platform-level issue).