stelliferous

Releases (2)

• 0.0.3.0 invalid

  expected a fresh constructor name (duplicate declaration: Text)

  at checkpoint_pickle.bend:14

  8 files do not load

  • checkpoint_pickle.bend
  • tensor_batch_norm.bend
  • tensor_checkpoint.bend
  • tensor_f32.bend
  • tensor_layer.bend
  • tensor_loss.bend
  • tensor_optimizer.bend
  • tensor_training.bend
• 0.0.2.0 indexed

Definitions (2,086)

tensor_operations.bend

• def apply_context
• def without_contextA mapper without a context, as the Unit-context case.
• def apply_indexed_when
• def holds
• def apply_indexedWrites f(index, element) for a contiguous input view into output[0..count).
• def small
• def pool_geometryThe window geometry is exact and every sample address stays in U32.
• def sweptThe unchecked sweep of two owners; the callers establish its bounds.
• def sweep_when
• def sweep_geometryThe window geometry of a whole axis is exact and every sample address stays in U32.
• def sweep_input
• def sweep_output
• def sweepWindows along one axis of outer contiguous blocks; see Access.Sweep.
• def pool_finishedThe input and the output of a pool around its column-swept planes.
• def pool_rows
• def pool_lines
• def poolWindows over the last two axes of channels contiguous planes: a sweep along the columns of every row into middle, [channels,height,out_width], then a sweep along the rows into output.
• def reduce_buffer
• def reduce_when
• def reduce

tensor_select.bend

• def dense
• def expanded
• type BandPlan
• def planned
• def plan
• def common
• def selected_leaf
• def selected_bands
• def dispatch
• def matching
• def shapes
• def masked_shape
• def values_shape
• def status
• def select

tensor_shape.bend

• type Shape
• def rank
• def extent
• def same_extents
• def same
• def has_zero
• def checked_multiply
• def product
• def count_nonempty
• def numelScalar [] has one element.
• def broadcast_extentNumPy broadcasting of one axis: equal extents, or one of them is 1.
• def prepend_extent
• def broadcast_same_rank
• def with_rankThe extents with leading axes of extent 1 added up to rank.
• def broadcastThe common extents of two operands under NumPy broadcasting: the extents are aligned at the last axis, a missing axis counts as 1, and on each axis the extents are equal or one of them is 1.
• def without_axisThe extents without one axis.
• def sum_when
• def checked_add
• def checked_add_to
• def paddedEach extent with before elements added in front and after elements behind.

tensor_view.bend

• type Axis
• type View
• def address_whenSuccessful addresses are below logical_length.
• def finish_address
• def advance_product
• def advance_inside
• def advance
• def locateThe recursive step consumes one axis and one coordinate; no flattening/copy.
• def address
• def axis_extents
• def shape
• def numel
• def rank
• def extent
• def build_axes
• def contiguous_count
• def contiguous
• def contiguous_product
• def contiguous_axis
• def contiguous_sizeValidate strides from the innermost axis outward.
• def has_count
• def contiguous_nonempty
• def is_contiguous
• def last_coordinates
• def found
• def fits_nonempty
• def axis_extents_ofForward strides are nonnegative: the last coordinate is the maximum address.
• def fits
• def contains_index
• def unique_axes
• def prepend_axis
• def reorder
• def permute_when
• def permuteOrder must contain each axis exactly once, including singleton axes.
• def slice_step
• def sliced_axis
• def add_offset
• def slice_axes
• def sliceExplicit start/count/positive step.
• def stretched_axisOne axis read as extent elements: unchanged when the extents agree, the single element repeated (stride 0) when the axis has extent 1.
• def stretched_prepend
• def stretch_axes
• def with_rankThe axes with leading axes of extent 1 added up to rank; the same elements.
• def stretched_view
• def stretchThe view read with the extents target under NumPy broadcasting.