Method fit

nlp_class2/rntn_tensorflow.py:77–151

(self, trees, lr=1e-2, mu=0.9, reg=1e-1, epochs=5)

        self.params = [self.We, self.W11, self.W22, self.W12, self.W1, self.W2, self.Wo]

    def fit(self, trees, lr=1e-2, mu=0.9, reg=1e-1, epochs=5):
        train_ops = []
        costs = []
        predictions = []
        all_labels = []
        i = 0
        N = len(trees)
        print("Compiling ops")
        for t in trees:
            i += 1
            sys.stdout.write("%d/%d\r" % (i, N))
            sys.stdout.flush()
            logits = self.get_output(t)
            labels = get_labels(t)
            all_labels.append(labels)

            cost = self.get_cost(logits, labels, reg)
            costs.append(cost)

            prediction = tf.argmax(logits, 1)
            predictions.append(prediction)

            train_op = tf.train.MomentumOptimizer(lr, mu).minimize(cost)
            train_ops.append(train_op)

        # save for later so we don't have to recompile if we call score
        self.predictions = predictions
        self.all_labels = all_labels

        self.saver = tf.train.Saver()
        init = tf.initialize_all_variables()
        actual_costs = []
        per_epoch_costs = []
        correct_rates = []
        with tf.Session() as session:
            session.run(init)

            for i in range(epochs):
                train_ops, costs, predictions, all_labels = shuffle(train_ops, costs, predictions, all_labels)
                epoch_cost = 0
                n_correct = 0
                n_total = 0
                j = 0
                N = len(train_ops)
                for train_op, cost, prediction, labels in zip(train_ops, costs, predictions, all_labels):
                    _, c, p = session.run([train_op, cost, prediction])
                    epoch_cost += c
                    actual_costs.append(c)
                    n_correct += np.sum(p == labels)
                    n_total += len(labels)

                    j += 1
                    if j % 10 == 0:
                        sys.stdout.write("j: %d, N: %d, c: %f\r" % (j, N, c))
                        sys.stdout.flush()

                    if np.isnan(c):
                        exit()