add_n_broadcast(tensors[, name]) |
Add zero or many tensors with broadcasting. |
assert_rank(x, ndims[, message, name]) |
Assert the rank of x is ndims. |
assert_rank_at_least(x, ndims[, message, name]) |
Assert the rank of x is at least ndims. |
assert_scalar_equal(a, b[, message, name]) |
Assert 0-d scalar a == b. |
assert_shape_equal(x, y[, message, name]) |
Assert the shape of x equals to y. |
bits_per_dimension(log_p, value_size[, …]) |
Compute “bits per dimension” of x. |
broadcast_concat(x, y, axis[, name]) |
Broadcast x and y, then concat them along axis. |
broadcast_to_shape(x, shape[, name]) |
Broadcast x to match shape. |
broadcast_to_shape_strict(x, shape[, name]) |
Broadcast x to match shape. |
classification_accuracy(y_pred, y_true[, name]) |
Compute the classification accuracy for y_pred and y_true. |
convert_to_tensor_and_cast(x[, dtype]) |
Convert x into a tf.Tensor, and cast its dtype if required. |
depth_to_space(input, block_size[, …]) |
Wraps tf.depth_to_space(), to support tensors higher than 4-d. |
flatten_to_ndims(x, ndims[, name]) |
Flatten the front dimensions of x, such that the resulting tensor will have at most ndims dimensions. |
log_mean_exp(x[, axis, keepdims, name]) |
Compute \(\log \frac{1}{K} \sum_{k=1}^K \exp(x_k)\). |
log_sum_exp(x[, axis, keepdims, name]) |
Compute \(\log \sum_{k=1}^K \exp(x_k)\). |
maybe_clip_value(x[, min_val, max_val, name]) |
Maybe clip the elements of x. |
pixelcnn_2d_sample(fn, inputs, height, width) |
Sample output from a PixelCNN 2D network, pixel-by-pixel. |
prepend_dims(x[, ndims, name]) |
Prepend [1] * ndims to the beginning of the shape of x. |
reshape_tail(input, ndims, shape[, name]) |
Reshape the tail (last) ndims into specified shape. |
shift(input, shift[, name]) |
Shift each axis of input according to shift, but keep identical size. |
smart_cond(cond, true_fn, false_fn[, name]) |
Execute true_fn or false_fn according to cond. |
softmax_classification_output(logits[, name]) |
Get the most possible softmax classification output for each logit. |
space_to_depth(input, block_size[, …]) |
Wraps tf.space_to_depth(), to support tensors higher than 4-d. |
transpose_conv2d_axis(input, …[, name]) |
Ensure the channels axis of input tensor to be placed at the desired axis. |
transpose_conv2d_channels_last_to_x(input, …) |
Ensure the channels axis (known to be the last axis) of input tensor to be placed at the desired axis. |
transpose_conv2d_channels_x_to_last(input, …) |
Ensure the channels axis of input tensor to be placed at the last axis. |
unflatten_from_ndims(x, static_front_shape, …) |
The inverse transformation of flatten(). |