Method fit

ann_class2/tf_with_save.py:54–97

(self, X, Y, Xtest, Ytest)

  def fit(self, X, Y, Xtest, Ytest):
    N, D = X.shape
    K = len(set(Y))

    # hyperparams
    max_iter = 30
    lr = 1e-3
    mu = 0.9
    regularization = 1e-1
    batch_sz = 100
    n_batches = N // batch_sz

    cost = self.build(D, K)
    l2_penalty = regularization*tf.reduce_mean(self.W**2) / 2
    cost += l2_penalty
    train_op = tf.train.MomentumOptimizer(lr, momentum=mu).minimize(cost)

    costs = []
    init = tf.global_variables_initializer()
    with tf.Session() as session:
        session.run(init)

        for i in range(max_iter):
            for j in range(n_batches):
                Xbatch = X[j*batch_sz:(j*batch_sz + batch_sz),]
                Ybatch = Y[j*batch_sz:(j*batch_sz + batch_sz),]

                session.run(train_op, feed_dict={self.inputs: Xbatch, self.targets: Ybatch})
                if j % 200 == 0:
                    test_cost = session.run(cost, feed_dict={self.inputs: Xtest, self.targets: Ytest})
                    Ptest = session.run(self.predict_op, feed_dict={self.inputs: Xtest})
                    err = error_rate(Ptest, Ytest)
                    print("Cost / err at iteration i=%d, j=%d: %.3f / %.3f" % (i, j, test_cost, err))
                    costs.append(test_cost)

        # save the model
        self.saver.save(session, self.savefile)

    # save dimensions for later
    self.D = D
    self.K = K

    plt.plot(costs)
    plt.show()


  def predict(self, X):