def forward()

in models.py [0:0]

• 30 lines of code
• 7 McCabe index (conditional complexity)

    def forward(self, inp, weights, reuse=False, scope='', stop_grad=False, label=None, stop_at_grad=False, stop_batch=False):
        weights = weights.copy()
        batch = tf.shape(inp)[0]

        act = tf.nn.leaky_relu

        if not FLAGS.cclass:
            label = None

        if stop_grad:
            for k, v in weights.items():
                if type(v) == dict:
                    v = v.copy()
                    weights[k] = v
                    for k_sub, v_sub in v.items():
                        v[k_sub] = tf.stop_gradient(v_sub)
                else:
                    weights[k] = tf.stop_gradient(v)

        # Make sure gradients are modified a bit
        inp = smart_conv_block(inp, weights, reuse, 'c1_pre', use_stride=False)

        hidden1 = smart_res_block(inp, weights, reuse, 'res_optim', adaptive=False, label=label, act=act)
        hidden2 = smart_res_block(hidden1, weights, reuse, 'res_1', stop_batch=stop_batch, downsample=False, adaptive=False, label=label, act=act)
        hidden3 = smart_res_block(hidden2, weights, reuse, 'res_2', stop_batch=stop_batch, label=label, act=act)

        if FLAGS.use_attention:
            hidden4 = smart_atten_block(hidden3, weights, reuse, 'atten', stop_at_grad=stop_at_grad, label=label)
        else:
            hidden4 = smart_res_block(hidden3, weights, reuse, 'res_3', adaptive=False, downsample=False, stop_batch=stop_batch, label=label, act=act)

        hidden5 = smart_res_block(hidden4, weights, reuse, 'res_4', stop_batch=stop_batch, adaptive=False, label=label, act=act)
        compact = hidden6 = smart_res_block(hidden5, weights, reuse, 'res_5', adaptive=False, downsample=False, stop_batch=stop_batch, label=label)
        hidden6 = tf.nn.relu(hidden6)
        hidden5 = tf.reduce_sum(hidden6, [1, 2])

        hidden6 = smart_fc_block(hidden5, weights, reuse, 'fc5')

        energy = hidden6

        return energy