"Ignore outliers in chart scaling" made me think my early training data was lost

I started retraining a model from scratch after changing its implementation, and I wanted to compare my results from the first few epochs between the two versions of the model. I went on TensorBoard for the original model and realized I couldn't see the accuracy for the first 3-4 epochs at all.

I ended up believing that the first few epochs had been lost somehow until @reedwm suggested unchecking "Ignore outliers in chart scaling." Once I did, I magically got all my data back.

I understand the rationale for having this feature is to zoom in on the more recent data which the user is probably more interested in, rather than having to zoom out to show the initial steep climb/drop. I wouldn't mind if it were simply a default zoom option, where you could still zoom out to see the rest of your data. However, if you aren't aware of this checkbox, your "outlier" data is completely inaccessible.

cc @chihuahua