Answer project 5 question 3 MIRA.

p5_classification/mira.py

@@ -60,8 +60,47 @@ class MiraClassifier:
         datum is a counter from features to values for those features
         representing a vector of values.
         """
-        "*** YOUR CODE HERE ***"
-        util.raiseNotDefined()
+
+        def scaleCounter(counter, scalar):
+            counter = counter.copy()
+            scalar = float(scalar)
+            for key in counter:
+                counter[key] *= scalar
+            return counter
+
+        def updateWeights(weights, expectedLabel, guessedLabel, datum, c):
+            weightExpected = weights[expectedLabel]
+            weightGuessed = weights[guessedLabel]
+            tau = ((weightExpected - weightGuessed) * datum + 1.0) / ((datum * datum) * 2.0)
+            tau = min(c, tau)
+            weights[expectedLabel] = weights[expectedLabel] + scaleCounter(datum, tau)
+            weights[guessedLabel] = weights[guessedLabel] - scaleCounter(datum, tau)
+
+        def evaluateWeights(weights):
+            correct = 0
+            for datum, expectedLabel in zip(validationData, validationLabels):
+                guessedLabel = guessLabel(weights, datum)
+                if guessedLabel != expectedLabel:
+                    correct += 1
+            return correct / float(len(validationData))
+
+        def guessLabel(weights, datum):
+            vectors = util.Counter()
+            for l in self.legalLabels:
+                vectors[l] = weights[l] * datum
+            return vectors.argMax()
+
+        allWeights = [] 
+        for c in Cgrid:
+            weights = self.weights.copy()
+            for iteration in range(self.max_iterations):
+                for datum, expectedLabel in zip(trainingData, trainingLabels):
+                    guessedLabel = guessLabel(weights, datum)
+                    if guessedLabel != expectedLabel:
+                        updateWeights(weights, expectedLabel, guessedLabel, datum, c)
+            accuracy = evaluateWeights(weights)
+            allWeights.append((accuracy, weights))
+        self.weights = max(allWeights)[1]
 
     def classify(self, data ):
         """