Questions for next meeting

README.md

@@ -57,6 +57,14 @@ Next steps:
 
 ## Questions
 - what about the other datasets mentioned in paper ? 
-- How about other historical datasets with less/more alarams ? 
+- How about other historical datasets with less/more alarms ? 
+- Different hyperparams in code and report: 
+	- results super non-determinisitc (good test set -> accuracy : 90% - 60/70% bigger variability)
+	- why such a high dropout
+	- why such a big model (reduce growth rate (less neurons in ech layer)) -> same accuracy with 10th of the params (4500 -> 400)
+	- amount of neuron in every layer with regards to growth rate (factor) (we need the original paper of the architecture) 
+	- more parameters than data points 
+	- why undersample since we lose 30 points because we already have little points
+