Python keras.backend.mean() Examples

def generate_pattern(layer_name, filter_index, size=150):
    # 过滤器可视化函数
    layer_output = model.get_layer(layer_name).output
    loss = K.mean(layer_output[:, :, :, filter_index])
    grads = K.gradients(loss, model.input)[0]
    grads /= (K.sqrt(K.mean(K.square(grads))) + 1e-5)
    iterate = K.function([model.input], [loss, grads])
    input_img_data = np.random.random((1, size, size, 3)) * 20 + 128.
    
    step = 1
    for _ in range(40):
        loss_value, grads_value = iterate([input_img_data])
        input_img_data += grads_value * step
    
    img = input_img_data[0]
    return deprocess_image(img) 

def call(self, x):
        mean = K.mean(x, axis=-1)
        std = K.std(x, axis=-1)

        if len(x.shape) == 3:
            mean = K.permute_dimensions(
                K.repeat(mean, x.shape.as_list()[-1]),
                [0,2,1]
            )
            std = K.permute_dimensions(
                K.repeat(std, x.shape.as_list()[-1]),
                [0,2,1] 
            )
            
        elif len(x.shape) == 2:
            mean = K.reshape(
                K.repeat_elements(mean, x.shape.as_list()[-1], 0),
                (-1, x.shape.as_list()[-1])
            )
            std = K.reshape(
                K.repeat_elements(mean, x.shape.as_list()[-1], 0),
                (-1, x.shape.as_list()[-1])
            )
        
        return self._g * (x - mean) / (std + self._epsilon) + self._b 

def gradient_penalty_loss(self, y_true, y_pred, averaged_samples):
        """
        Computes gradient penalty based on prediction and weighted real / fake samples
        """
        gradients = K.gradients(y_pred, averaged_samples)[0]
        # compute the euclidean norm by squaring ...
        gradients_sqr = K.square(gradients)
        #   ... summing over the rows ...
        gradients_sqr_sum = K.sum(gradients_sqr,
                                  axis=np.arange(1, len(gradients_sqr.shape)))
        #   ... and sqrt
        gradient_l2_norm = K.sqrt(gradients_sqr_sum)
        # compute lambda * (1 - ||grad||)^2 still for each single sample
        gradient_penalty = K.square(1 - gradient_l2_norm)
        # return the mean as loss over all the batch samples
        return K.mean(gradient_penalty) 

def test_helper(func, exponent, layer, lr, dropout_first, dropout_middle, 
                dropout_last, alpha, prefix='SWO GBP ', postfix='',
                with_comparison=False):
    print('Test %s, %s, %s, %s, %s %s %s' % (exponent, layer, lr, dropout_first,
                                       dropout_middle, dropout_last, alpha))
    model = func(exponent=exponent, lr=lr, layers=layer, 
                 dropout_first=dropout_first, dropout_middle=dropout_middle,
                 dropout_last=dropout_last, prefix=prefix, postfix=postfix, 
                 alpha=alpha)
    model.train(200)
    val_loss = np.mean(model.history['history']['val_loss'][-5:])
    
#    if with_comparison:
#        swo = inst.get_swaptiongen(inst.hullwhite_analytic)
#        _, values = swo.compare_history(model, dates=dates)
#        
    
    return (val_loss, layer, exponent, lr, dropout_first, dropout_middle, 
            dropout_last, alpha) 

def audio_discriminate_loss2(gamma=0.1,beta = 2*0.1,num_speaker=2):
    def loss_func(S_true,S_pred,gamma=gamma,beta=beta,num_speaker=num_speaker):
        sum_mtr = K.zeros_like(S_true[:,:,:,:,0])
        for i in range(num_speaker):
            sum_mtr += K.square(S_true[:,:,:,:,i]-S_pred[:,:,:,:,i])
            for j in range(num_speaker):
                if i != j:
                    sum_mtr -= gamma*(K.square(S_true[:,:,:,:,i]-S_pred[:,:,:,:,j]))

        for i in range(num_speaker):
            for j in range(i+1,num_speaker):
                #sum_mtr -= beta*K.square(S_pred[:,:,:,i]-S_pred[:,:,:,j])
                #sum_mtr += beta*K.square(S_true[:,:,:,:,i]-S_true[:,:,:,:,j])
                pass
        #sum = K.sum(K.maximum(K.flatten(sum_mtr),0))

        loss = K.mean(K.flatten(sum_mtr))

        return loss
    return loss_func 

def prepareSamples(self, cnum = 0, num = 1000): #8x8 images, bottom row is constant

        try:
            os.mkdir("Results/Samples-c" + str(cnum))
        except:
            x = 0

        im = self.im.get_class(cnum)
        e = self.GAN.E.predict(im, batch_size = BATCH_SIZE * k_images)

        mean = np.mean(e, axis = 0)
        std = np.std(e, axis = 0)

        n = noise(num)
        nc = nClass(num, mean, std)

        im = self.GAN.G.predict([n, nc], batch_size = BATCH_SIZE)

        for i in range(im.shape[0]):

            x = Image.fromarray(np.uint8(im[i]*255), mode = 'RGB')

            x.save("Results/Samples-c" + str(cnum) + "/im ("+str(i+1)+").png") 

def rpn_class_loss_graph(rpn_match, rpn_class_logits):
    """RPN anchor classifier loss.
    rpn_match: [batch, anchors, 1]. Anchor match type. 1=positive,
               -1=negative, 0=neutral anchor.
    rpn_class_logits: [batch, anchors, 2]. RPN classifier logits for FG/BG.
    """
    # Squeeze last dim to simplify
    rpn_match = tf.squeeze(rpn_match, -1)
    # Get anchor classes. Convert the -1/+1 match to 0/1 values.
    anchor_class = K.cast(K.equal(rpn_match, 1), tf.int32)
    # Positive and Negative anchors contribute to the loss,
    # but neutral anchors (match value = 0) don't.
    indices = tf.where(K.not_equal(rpn_match, 0))
    # Pick rows that contribute to the loss and filter out the rest.
    rpn_class_logits = tf.gather_nd(rpn_class_logits, indices)
    anchor_class = tf.gather_nd(anchor_class, indices)
    # Cross entropy loss
    loss = K.sparse_categorical_crossentropy(target=anchor_class,
                                             output=rpn_class_logits,
                                             from_logits=True)
    loss = K.switch(tf.size(loss) > 0, K.mean(loss), tf.constant(0.0))
    return loss 

def gen_adv_loss(logits, y, loss='logloss', mean=False):
    """
    Generate the loss function.
    """

    if loss == 'training':
        # use the model's output instead of the true labels to avoid
        # label leaking at training time
        y = K.cast(K.equal(logits, K.max(logits, 1, keepdims=True)), "float32")
        y = y / K.sum(y, 1, keepdims=True)
        out = K.categorical_crossentropy(y, logits, from_logits=True)
    elif loss == 'logloss':
        out = K.categorical_crossentropy(y, logits, from_logits=True)
    else:
        raise ValueError("Unknown loss: {}".format(loss))

    if mean:
        out = K.mean(out)
    # else:
    #     out = K.sum(out)
    return out 

def predict(self, x):
        r"""
        Predict quantiles of the conditional distribution P(y|x).

        Forward propagates the inputs in `x` through the network to
        obtain the predicted quantiles `y`.

        Arguments:

            x(np.array): Array of shape `(n, m)` containing `n` m-dimensional inputs
                         for which to predict the conditional quantiles.

        Returns:

             Array of shape `(n, k)` with the columns corresponding to the k
             quantiles of the network.

        """
        predictions = np.stack(
            [m.predict((x - self.x_mean) / self.x_sigma) for m in self.models])
        return np.mean(predictions, axis=0) 

def posterior_mean(self, x):
        r"""
        Computes the posterior mean by computing the first moment of the
        estimated posterior CDF.

        Arguments:

            x(np.array): Array of shape `(n, m)` containing `n` inputs for which
                         to predict the posterior mean.
        Returns:

            Array containing the posterior means for the provided inputs.
        """
        y_pred, qs = self.cdf(x)
        mus = y_pred[-1] - np.trapz(qs, x=y_pred)
        return mus 

def sampling(args: tuple):
    """
    Reparameterization trick by sampling z from unit Gaussian
    :param args: (tensor, tensor) mean and log of variance of q(z|x)
    :returns tensor: sampled latent vector z
    """

    # unpack the input tuple
    z_mean, z_log_var = args

    # mini-batch size
    mb_size = K.shape(z_mean)[0]

    # latent space size
    dim = K.int_shape(z_mean)[1]

    # random normal vector with mean=0 and std=1.0
    epsilon = K.random_normal(shape=(mb_size, dim))

    return z_mean + K.exp(0.5 * z_log_var) * epsilon 

def optimizer(self):
        a = K.placeholder(shape=(None,), dtype='int32')
        y = K.placeholder(shape=(None,), dtype='float32')

        py_x = self.model.output

        a_one_hot = K.one_hot(a, self.action_size)
        q_value = K.sum(py_x * a_one_hot, axis=1)
        error = K.abs(y - q_value)

        quadratic_part = K.clip(error, 0.0, 1.0)
        linear_part = error - quadratic_part
        loss = K.mean(0.5 * K.square(quadratic_part) + linear_part)

        optimizer = RMSprop(lr=0.00025, epsilon=0.01)
        updates = optimizer.get_updates(self.model.trainable_weights, [], loss)
        train = K.function([self.model.input, a, y], [loss], updates=updates)

        return train

    # approximate Q function using Convolution Neural Network
    # state is input and Q Value of each action is output of network 

def optimizer(self):
        a = K.placeholder(shape=(None, ), dtype='int32')
        y = K.placeholder(shape=(None, ), dtype='float32')

        py_x = self.model.output

        a_one_hot = K.one_hot(a, self.action_size)
        q_value = K.sum(py_x * a_one_hot, axis=1)
        error = K.abs(y - q_value)

        quadratic_part = K.clip(error, 0.0, 1.0)
        linear_part = error - quadratic_part
        loss = K.mean(0.5 * K.square(quadratic_part) + linear_part)

        optimizer = RMSprop(lr=0.00025, epsilon=0.01)
        updates = optimizer.get_updates(self.model.trainable_weights, [], loss)
        train = K.function([self.model.input, a, y], [loss], updates=updates)

        return train

    # approximate Q function using Convolution Neural Network
    # state is input and Q Value of each action is output of network
    # dueling network's Q Value is sum of advantages and state value 

def rpn_class_loss_graph(rpn_match, rpn_class_logits):
    """RPN anchor classifier loss.

    rpn_match: [batch, anchors, 1]. Anchor match type. 1=positive,
               -1=negative, 0=neutral anchor.
    rpn_class_logits: [batch, anchors, 2]. RPN classifier logits for FG/BG.
    """
    # Squeeze last dim to simplify
    rpn_match = tf.squeeze(rpn_match, -1)
    # Get anchor classes. Convert the -1/+1 match to 0/1 values.
    anchor_class = K.cast(K.equal(rpn_match, 1), tf.int32)
    # Positive and Negative anchors contribute to the loss,
    # but neutral anchors (match value = 0) don't.
    indices = tf.where(K.not_equal(rpn_match, 0))
    # Pick rows that contribute to the loss and filter out the rest.
    rpn_class_logits = tf.gather_nd(rpn_class_logits, indices)
    anchor_class = tf.gather_nd(anchor_class, indices)
    # Cross entropy loss
    loss = K.sparse_categorical_crossentropy(target=anchor_class,
                                             output=rpn_class_logits,
                                             from_logits=True)
    loss = K.switch(tf.size(loss) > 0, K.mean(loss), tf.constant(0.0))
    return loss 

def profile_contrib(p):
    return kl.Lambda(lambda p:
                     K.mean(K.sum(K.stop_gradient(tf.nn.softmax(p, dim=-2)) * p, axis=-2), axis=-1)
                     )(p) 

def deprocess_image(x):
    # 将张量转换为有效图像的函数
    x -= x.mean()
    x /= (x.std() + 1e-5)
    x *= 0.1

    x += 0.5
    x = np.clip(x, 0, 1)

    x *= 255
    x = np.clip(x, 0, 255).astype('uint8')
    return x 

def norm(self, xs, norm_id):
    mu = K.mean(xs, axis=-1, keepdims=True)
    sigma = K.sqrt(K.var(xs, axis=-1, keepdims=True) + 1e-3)
    xs = self.gs[norm_id] * (xs - mu) / (sigma + 1e-3) + self.bs[norm_id]
    return xs 

def pair_loss(y_true, y_pred):
    y_true = tf.cast(y_true, tf.int32)
    parts = tf.dynamic_partition(y_pred, y_true, 2)
    y_pos = parts[1]
    y_neg = parts[0]
    y_pos = tf.expand_dims(y_pos, 0)
    y_neg = tf.expand_dims(y_neg, -1)
    out = K.sigmoid(y_neg - y_pos)
    return K.mean(out) 

def negative_binary_crossentropy(y_true, y_pred):
    """Instead of minimizing log(1-D), maximize log(D).

    Note that when using this loss function, you should not change the target.
    For example, if you want G -> 0 and D -> 1, then you should replace your
    binary_crossentropy loss with negative_binary_crossentropy loss without
    changing to G -> 1.
    """

    return -K.mean(K.binary_crossentropy(y_pred, 1 - y_true), axis=-1) 

def maximize(_, y_pred):
    """Maximizes y_pred, regardless of y_true."""

    return -K.mean(y_pred) 

def minimize(_, y_pred):
    """Minimizes y_pred, regardless of y_true."""

    return K.mean(y_pred) 

def cosine(a, b):
    """Cosine similarity. Maximum is 1 (a == b), minimum is -1 (a == -b)."""

    a = K.l2_normalize(a)
    b = K.l2_normalize(b)
    return 1 - K.mean(a * b, axis=-1) 

def geometric(a, b):
    """Geometric mean of sigmoid and euclidian similarity."""

    return sigmoid(a, b) * euclidean(a, b) 

def create_model(gpu):
    with tf.device(gpu):
        input = Input((1280, 1918, len(dirs)))
        x = Lambda(lambda x: K.mean(x, axis=-1, keepdims=True))(input)
        model = Model(input, x)
        model.summary()
    return model 

def bootstrapped_crossentropy(y_true, y_pred, bootstrap_type='hard', alpha=0.95):
    target_tensor = y_true
    prediction_tensor = y_pred
    _epsilon = _to_tensor(K.epsilon(), prediction_tensor.dtype.base_dtype)
    prediction_tensor = K.tf.clip_by_value(prediction_tensor, _epsilon, 1 - _epsilon)
    prediction_tensor = K.tf.log(prediction_tensor / (1 - prediction_tensor))

    if bootstrap_type == 'soft':
        bootstrap_target_tensor = alpha * target_tensor + (1.0 - alpha) * K.tf.sigmoid(prediction_tensor)
    else:
        bootstrap_target_tensor = alpha * target_tensor + (1.0 - alpha) * K.tf.cast(
            K.tf.sigmoid(prediction_tensor) > 0.5, K.tf.float32)
    return K.mean(K.tf.nn.sigmoid_cross_entropy_with_logits(
        labels=bootstrap_target_tensor, logits=prediction_tensor)) 

def online_bootstrapping(y_true, y_pred, pixels=512, threshold=0.5):
    """ Implements nline Bootstrapping crossentropy loss, to train only on hard pixels,
        see  https://arxiv.org/abs/1605.06885 Bridging Category-level and Instance-level Semantic Image Segmentation
        The implementation is a bit different as we use binary crossentropy instead of softmax
        SUPPORTS ONLY MINIBATCH WITH 1 ELEMENT!
    # Arguments
        y_true: A tensor with labels.

        y_pred: A tensor with predicted probabilites.

        pixels: number of hard pixels to keep

        threshold: confidence to use, i.e. if threshold is 0.7, y_true=1, prediction=0.65 then we consider that pixel as hard
    # Returns
        Mean loss value
    """
    y_true = K.flatten(y_true)
    y_pred = K.flatten(y_pred)
    difference = K.abs(y_true - y_pred)

    values, indices = K.tf.nn.top_k(difference, sorted=True, k=pixels)
    min_difference = (1 - threshold)
    y_true = K.tf.gather(K.gather(y_true, indices), K.tf.where(values > min_difference))
    y_pred = K.tf.gather(K.gather(y_pred, indices), K.tf.where(values > min_difference))

    return K.mean(K.binary_crossentropy(y_true, y_pred)) 

def mutual_info_loss(self, c, c_given_x):
        """The mutual information metric we aim to minimize"""
        eps = 1e-8
        conditional_entropy = K.mean(- K.sum(K.log(c_given_x + eps) * c, axis=1))
        entropy = K.mean(- K.sum(K.log(c + eps) * c, axis=1))

        return conditional_entropy + entropy 

def wasserstein_loss(self, y_true, y_pred):
        return K.mean(y_true * y_pred) 

def wasserstein_loss(self, y_true, y_pred):
        return K.mean(y_true * y_pred) 

def wasserstein_loss(self, y_true, y_pred):
        return K.mean(y_true * y_pred)