def __init__(self, gan, config):
    """
    Initializes a gan component based on a `gan` and a `config` dictionary.
    Different components require different config variables.  
    A `ValidationException` is raised if the GAN component configuration fails to validate.
    """
    self.gan = gan
    self.config = hc.Config(config)
    errors = self.validate()
    if errors != []:
        raise ValidationException(self.__class__.__name__+": " +"\n".join(errors))
    self.create_ops(config)

def add_noise(self, net):
    config = self.config
    if not config.noise:
        return net
    print("[discriminator] adding noise", config.noise)
    net += tf.random_normal(net.get_shape(), mean=0, stddev=config.noise, dtype=tf.float32)
    return net

def biases(self):
    """
        Biases of the GAN component.
    """
    return self.ops.biases

def build(self, net):
    config = self.config
    gan = self.gan
    ops = self.ops
    layers = config.layers
    depth_increase = config.depth_increase
    activation = ops.lookup(config.activation)
    final_activation = ops.lookup(config.final_activation)
    x, g = self.split_batch(net)
    net = self.add_noise(net)
    if layers > 0:
        initial_depth = max(ops.shape(net)[3], config.initial_depth or 64)
        if config.skip_layer_filters and 0 in config.skip_layer_filters:
            pass
        else:
            net = self.layer_filter(net)
        net = config.block(self, net, initial_depth, filter=config.initial_filter or 3)
    for i in range(layers):
        xg = None
        is_last_layer = (i == layers-1)
        filters = ops.shape(net)[3]
        net = activation(net)
        net = self.layer_regularizer(net)
        if config.skip_layer_filters and (i+1) in config.skip_layer_filters:
            pass
        else:
            net = self.layer_filter(net)
            print("[hypergan] adding layer filter", net)
        net = self.progressive_enhancement(config, net, xg)
        depth = filters + depth_increase
        net = config.block(self, net, depth)
        print('[discriminator] layer', net)
    for i in range(config.extra_layers or 0):
        output_features = int(int(net.get_shape()[3]))
        net = activation(net)
        net = self.layer_regularizer(net)
        net = ops.conv2d(net, 3, 3, 1, 1, output_features//(config.extra_layers_reduction or 1))
        print('[discriminator] extra layer', net)
    k=-1
    if config.relation_layer:
        net = activation(net)
        net = self.layer_regularizer(net)
        net = self.relation_layer(net)
    #net = tf.reshape(net, [ops.shape(net)[0], -1])
    if final_activation or (config.fc_layers or 0) > 0:
        net = self.layer_regularizer(net)
    for i in range(config.fc_layers or 0):
        net = self.layer_regularizer(net)
        net = activation(net)
        net = ops.reshape(net, [ops.shape(net)[0], -1])
        net = ops.linear(net, config.fc_layer_size or 300)
        net = self.layer_regularizer(net)
    if final_activation:
        net = final_activation(net)
    print("[discriminator] output", net)
    return net

def create(self, net=None, x=None, g=None):
    config = self.config
    gan = self.gan
    ops = self.ops
    if net is None:
        if x is None:
            x = gan.inputs.x
        if g is None:
            g = gan.generator.sample
        x, g = self.resize(config, x, g)
        net = tf.concat(axis=0, values=[x, g])
        net = self.layer_filter(net)
    net = self.build(net)
    self.sample = net
    return net

def create_ops(self, config):
    """
    Create the ops object as `self.ops`.  Also looks up config
    """
    if self.gan is None:
        return
    if self.gan.ops_backend is None:
        return
    self.ops = self.gan.ops_backend(config=self.config, device=self.gan.device)
    self.config = self.gan.ops.lookup(config)

def fully_connected_from_list(self, nets):
    results = []
    ops = self.ops
    for net, net2 in nets:
        net = ops.concat([net, net2], axis=3)
        shape = ops.shape(net)
        bs = shape[0]
        net = ops.reshape(net, [bs, -1])
        features = ops.shape(net)[1]
        net = ops.linear(net, features)
        #net = self.layer_regularizer(net)
        net = ops.lookup('lrelu')(net)
        #net = ops.linear(net, features)
        net = ops.reshape(net, shape)
        results.append(net)
    return results

def layer_filter(self, net):
    config = self.config
    gan = self.gan
    ops = self.ops
    if 'layer_filter' in config and config.layer_filter is not None:
        print("[discriminator] applying layer filter", config['layer_filter'])
        stacks = ops.shape(net)[0] // gan.batch_size()
        filters = []
        for stack in range(stacks):
            piece = tf.slice(net, [stack * gan.batch_size(), 0,0,0], [gan.batch_size(), -1, -1, -1])
            filters.append(config.layer_filter(gan, self.config, piece))
        layer = tf.concat(axis=0, values=filters)
        net = tf.concat(axis=3, values=[net, layer])
    return net

def layer_regularizer(self, net):
    symbol = self.config.layer_regularizer
    op = self.gan.ops.lookup(symbol)
    if op:
        net = op(self, net)
    return net

def permute(self, nets, k):
    return list(itertools.permutations(nets, k))

def progressive_enhancement(self, config, net, xg):
    if 'progressive_enhancement' in config and config.progressive_enhancement and xg is not None:
        net = tf.concat(axis=3, values=[net, xg])
    return net

def relation_layer(self, net):
    ops = self.ops
    #hack
    shape = ops.shape(net)
    input_size = shape[1]*shape[2]*shape[3]
    netlist = self.split_by_width_height(net)
    permutations = self.permute(netlist, 2)
    permutations = self.fully_connected_from_list(permutations)
    net = ops.concat(permutations, axis=3)
    #hack
    bs = ops.shape(net)[0]
    net = ops.reshape(net, [bs, -1])
    net = ops.linear(net, input_size)
    net = ops.reshape(net, shape)
    return net

def required(self):
    return "activation layers block depth_increase initial_depth".split()

def resize(self, config, x, g):
    if(config.resize):
        # shave off layers >= resize 
        def should_ignore_layer(layer, resize):
            return int(layer.get_shape()[1]) > config['resize'][0] or \
                   int(layer.get_shape()[2]) > config['resize'][1]
        xs = [px for px in xs if not should_ignore_layer(px, config['resize'])]
        gs = [pg for pg in gs if not should_ignore_layer(pg, config['resize'])]
        x = tf.image.resize_images(x,config['resize'], 1)
        g = tf.image.resize_images(g,config['resize'], 1)
    else:
        return x, g

def reuse(self, net=None, x=None, g=None):
    config = self.config
    gan = self.gan
    ops = self.ops
    if net is None:
        if x is None:
            x or gan.inputs.x
        if g is None:
            g or gan.generator.sample
        x, g = self.resize(config, x, g)
        net = self.combine_filter(config, x, g)
    self.ops.reuse()
    net = self.build(net)
    self.ops.stop_reuse()
    return net

def split_batch(self, net, count=2):
    """ 
    Discriminators return stacked results (on axis 0).  
    
    This splits the results.  Returns [d_real, d_fake]
    """
    ops = self.ops or self.gan.ops
    s = ops.shape(net)
    bs = s[0]
    nets = []
    net = ops.reshape(net, [bs, -1])
    start = [0 for x in ops.shape(net)]
    for i in range(count):
        size = [bs//count] + [x for x in ops.shape(net)[1:]]
        nets.append(ops.slice(net, start, size))
        start[0] += bs//count
    return nets

def split_by_width_height(self, net):
    elems = []
    ops = self.gan.ops
    shape = ops.shape(net)
    bs = shape[0]
    height = shape[1]
    width = shape[2]
    for i in range(width):
        for j in range(height):
            elems.append(ops.slice(net, [0, i, j, 0], [bs, 1, 1, -1]))
    return elems

def validate(self):
    """
    Validates a GANComponent.  Return an array of error messages. Empty array `[]` means success.
    """
    errors = []
    required = self.required()
    for argument in required:
        if(self.config.__getattr__(argument) == None):
            errors.append("`"+argument+"` required")
    if(self.gan is None):
        errors.append("GANComponent constructed without GAN")
    return errors

def variables(self):
    """
        All variables associated with this component.
    """
    return self.ops.variables()

def weights(self):
    """
        The weights of the GAN component.
    """
    return self.ops.weights