diff --git a/learning_curve.py b/learning_curve.py
index 2364f2c..35edf7e 100644
--- a/learning_curve.py
+++ b/learning_curve.py
@@ -8,7 +8,7 @@
 
 data = load_digits()
 print data.DESCR
-num_trials = 10
+num_trials = 100
 train_percentages = range(5,95,5)
 test_accuracies = numpy.zeros(len(train_percentages))
 
@@ -16,11 +16,32 @@
 # the resultant accuracy.
 # You should repeat each training percentage num_trials times to smooth out variability
 # for consistency with the previous example use model = LogisticRegression(C=10**-10) for your learner
+#result_data_train = []
+result_data_test = []
+for percentage in train_percentages:
+    n = 0
+    #train_data = []
+    test_data = []
+    while n <= num_trials :
+        n = n + 1
+        X_train, X_test, y_train, y_test = train_test_split(data.data, data.target, train_size= percentage)
+        model = LogisticRegression(C=10**-10)
+        model.fit(X_train, y_train)
+        #model_score_train =model.score(X_train,y_train)
+        model_score_test = model.score(X_test,y_test)
+       # train_data.append(model_score_train)
+        test_data.append(model_score_test) # this is a list of floats
+    #train_ave= numpy.mean(train_data)
+    test_ave = numpy.mean(test_data)
+    #result_data_train.append(train_ave)
+    result_data_test.append(test_ave)
+    test_accuracies = result_data_test
+
 
-# TODO: your code here
 
 fig = plt.figure()
-plt.plot(train_percentages, test_accuracies)
+plt.plot(train_percentages, test_accuracies, label = 'testing data')
 plt.xlabel('Percentage of Data Used for Training')
 plt.ylabel('Accuracy on Test Set')
+plt.legend()
 plt.show()
diff --git a/questions.txt b/questions.txt
new file mode 100644
index 0000000..795f118
--- /dev/null
+++ b/questions.txt
@@ -0,0 +1,10 @@
+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).
\ No newline at end of file