finished toolbox

learning_curve.py

@@ -7,20 +7,24 @@
 from sklearn.linear_model import LogisticRegression
 
 data = load_digits()
-print data.DESCR
-num_trials = 10
+#print data.DESCR
+num_trials = 100
+
 train_percentages = range(5,95,5)
 test_accuracies = numpy.zeros(len(train_percentages))
 
-# train a model with training percentages between 5 and 90 (see train_percentages) and evaluate
-# 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
-
-# TODO: your code here
+def trainer(percent, num_trials):
+    results = []
+    model = LogisticRegression(C=10**-4)
+    for i in range(num_trials):
+        X_train, X_test, y_train, y_test = train_test_split(data.data, data.target, train_size=(percent/float(100)))
+        model.fit(X_train, y_train)
+        results.append(model.score(X_test,y_test))
+    return sum(results)/float(num_trials)
 
+results = [trainer(percent, num_trials) for percent in train_percentages]
 fig = plt.figure()
-plt.plot(train_percentages, test_accuracies)
+plt.plot(train_percentages, results)
 plt.xlabel('Percentage of Data Used for Training')
 plt.ylabel('Accuracy on Test Set')
 plt.show()

questions.txt

@@ -0,0 +1,4 @@
+1) The bigger the training set, the more accurate the model is on the test set.
+2) The lower end of the curve is noisy, because accuracy is highly dependent on how representative the training set data is of the test set. Regression to the mean causes larger sets to be inherently more "stable" in their evaluation.
+3)100 trials produces a much smoother curve.
+4)As the inverse regularization coefficient increases, the model tries to regress to more and more accurate levels, creating a more accurate model. (In rough terms). This causes a much more representative accuracy graph.