JKtheSlacker made me do it

serie2/rapport.md

@@ -166,7 +166,7 @@ We see the error is slowly decreasing :
 ```
 ![cost](output10.png)
 ### Summarize best learning rate
-We see that the learning rate of 0.05 is the best because it is the one that converges the fastest. Indeed, the error is decreasing the fastest and the cost is the lowest.
+We see that the learning rate of 0.05 is the best because it is the one that has the best value for the test set. Effectively, with this learning rate the error is descreasing faster that 0.001 and the cost is also the lowest for the test set.
 
 
 ### Analyse the dependency of the final error rate on the number of epochs