Interpretation of learning curve (training & validation) I know what a learning curve representative of an "ideal" overfitting looks like. However, I am not 100% sure how to interpret the learning curve shown below. Why does the model sometimes seem to "slip" ?
It is a small experiment of a binary classification problem with slightly imbalanced classes (2:1 ratio) which was implemented in keras. The loss is calculated as binary crossentropy. I accounted for the imbalance by weighing one class more as described here.
P.s.: Not exactly sure whether this post belongs here or rather on data science stackexchange...

 A: Assuming that your validation data is separate from your training data and you have enough samples in both sets, I can see no sign of overfitting in these curves. 
However it is possible to overfit on the validation set too, if you train your model over and over again to find the best hyper-parameters. Your safest bet would be testing the generalization of your model on a separate hold-out test set (besides the validation set).
The slip you refer to is sometimes an indication of a small number of samples. For example if you have 9 samples in your validation set, missing one would result in a steep drop. This does not seem to be the case, however, judging by the small shifts in some epochs. Another explanation could be that you have a small number of samples in one of the under-represented classes, which would be weighted by a large class weight. If that were the case, the effect of missing one of those samples would be magnified by the large class weight.
In any case you should evaluate on a test set on more metrics that just the accuracy. I'd recommend computing a confusion matrix which would give you a clear view of how your model performs for each class.
