package autodl

Get desktop application:
View/edit binary Protocol Buffers messages

Compressed value

Used in: Matrix

• optional bytes encoded_image = 1

  Image encoded with the util image_codec::CoderUtil and the image_codec::IMAGE_JPG encoding.

Metadata of autodl datasets. An input is a sequence of a set of real valued matrices (aka matrix bundle). An output is a real valued vector.

• repeated MatrixSpec matrix_spec = 1

  Describes the matrices in the matrix bundles.

• optional bool is_sequence = 2

  True if the input set of matrix bundles is a sequence (order matters).

• optional int32 output_dim = 3

  Number of labels in the output.

• map<string, int32> label_to_index_map = 4

  Mapping between int and human-readable label names.

• map<string, int32> feature_to_index_map = 5

  Mapping between int and human-readable feature names.

• map<string, int32> channel_to_index_map = 6

  Mapping between int and human-readable channel names.

• optional int32 sequence_size = 7

  Size of the sequence of matrix bundles. If the size of a sample is less than sequence_size it will be padded with a 0-value matrices. Any bundle after sequence_size will be dropped.

• optional int32 sample_count = 8

  Number of samples in this dataset.

Matrix values in a dense format. Column-major layout.

Used in: Matrix

• repeated float value = 1

  Matrix[i][j] is at index: i * col_count + j. The length of the array can be smaller than col_count * row_count. The missing values are considered null/zero.

An input is a sequence of matrix bundles. The matrix bundles should have the same structure. That is the nth matrices in each bundle should have the same specification.

Used in: Sample

• repeated MatrixBundle bundle = 1

• optional bool is_sequence = 2

  True if this input is part of a sequence, e.g. a frame in a video clip.

Used in: Output

• optional int32 index = 1

• optional float score = 2

Used in: MatrixBundle

• oneof values

  • SparseValue sparse = 1

  • DenseValue dense = 2

  • Compressed compressed = 5

• optional MatrixSpec spec = 3

• optional int32 bundle_index = 4

  Position of this matrix in a bundle.

A matrix bundle can be used to model a mix of dense and sparse matrices. It is for example useful to model an image and some extra information of the image (where it comes from). Matrices inside a matrix bundle can have different formats (e.g. different dimensions and dense/sparse).

Used in: Input

• repeated Matrix matrix = 1

• optional int32 sequence_index = 2

  Position of this bundle set in a sequence (from 0 to ...).

Metadata about a matrix

Used in: DataSpecification, Matrix

• optional int32 col_count = 1

• optional int32 row_count = 2

• optional bool is_sequence_col = 3

  True if input matrix columns/rows refer to sequences (e.g. for time series).

• optional bool is_sequence_row = 4

• optional bool has_locality_col = 5

  Sets if the column indices carry some structure/information based on closeness of indices, like in a time series where the col index is time.

• optional bool has_locality_row = 6

• optional MatrixSpec.Format format = 8

  A matrix is a Dense matrix unless specified.

• optional bool is_sparse = 7

• optional int32 num_channels = 9

  Number of channels. This is useful for RGB or sensor data

Used in: MatrixSpec

• DENSE = 0

• SPARSE = 1

• COMPRESSED = 2

Used in: Sample

• repeated Label label = 1

• optional int32 id = 1

• optional Input input = 2

• optional Output output = 3

Matrix entry in a sparse format.

Used in: SparseValue

• optional int32 row = 1

• optional int32 col = 2

• optional float value = 3

Matrix values in a sparse format.

Used in: Matrix

• repeated SparseEntry entry = 1