TensorFlow实现的L2-SVM二分类模型

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

__version__ = '0.3.7'
__author__ = 'Abien Fred Agarap'

import numpy as np
import os
import sys
import tensorflow as tf
import time


class Svm:
    """Implementation of L2-Support Vector Machine using TensorFlow"""

    def __init__(self, alpha, batch_size, svm_c, num_classes, num_features):
        """Initialize the SVM class
        Parameter
        ---------
        alpha : float
          The learning rate for the SVM model.
        batch_size : int
          Number of batches to use for training and testing.
        svm_c : float
          The SVM penalty parameter.
        num_classes : int
          Number of classes in a dataset.
        num_features : int

          Number of features in a dataset.
        """
        self.alpha = alpha
        self.batch_size = batch_size
        self.svm_c = svm_c
        self.num_classes = num_classes
        self.num_features = num_features

        def __graph__():
            """Building the inference graph"""

            learning_rate = tf.placeholder(dtype=tf.float32, name='learning_rate')

            with tf.name_scope('input'):
                # [BATCH_SIZE, SEQUENCE_LENGTH]
                x_input = tf.placeholder(dtype=tf.float32, shape=[None, self.num_features], name='x_input')

                # [BATCH_SIZE, N_CLASSES]
                y_input = tf.placeholder(dtype=tf.uint8, shape=[None], name='y_input')

                y_onehot = tf.one_hot(indices=y_input, depth=self.num_classes, on_value=1, off_value=-1,
                                      name='y_onehot')

            with tf.name_scope('training_ops'):
                with tf.name_scope('weights'):
                    weight = tf.get_variable(name='weights',
                                             initializer=tf.random_normal([self.num_features, self.num_classes],
                                                                          stddev=0.01))
                    self.variable_summaries(weight)
                with tf.name_scope('biases'):