questions.txt

1. The general trend in the curve is upwards - the more data you are using for the training, the better the accuracy of
 the test set is!
2. The small end of the curve (low percentage of data used for training) is very noisy because you are dealing with a
random set so the training may not reflect the actual characteristics of the letter well (in other words, you may have
outliers in the set of letters used for training when the percentage of characters used for training is low).
3. In order to get a smooth (not super spiky) curve, I tried 100 trials, which helped a lot but wasn't enough to reduce
all of the "spikiness"; at 1000 trials, the curve is reasonably smooth
4. As I try different values of C, the curve changes. Using 100 trials each, the curve at C=1**-10 is extemely smooth,
C=10**-10 is fairly "lumpy/spikey" (linear in smaller portions of the graph) and at C=100**-10 much more so (linear in
large portions of the graph).