Turi Create  4.0
turi::activity_classification::data_iterator::batch Struct Reference

#include <toolkits/activity_classification/ac_data_iterator.hpp>

Classes

struct  chunk_info
 

Public Attributes

neural_net::shared_float_array features
 
neural_net::shared_float_array labels
 
neural_net::shared_float_array weights
 
neural_net::shared_float_array labels_per_row
 
std::vector< chunk_infobatch_info
 

Detailed Description

Defines the output of a data_iterator.

Definition at line 76 of file ac_data_iterator.hpp.

Member Data Documentation

◆ batch_info

std::vector<chunk_info> turi::activity_classification::data_iterator::batch::batch_info

Metadata for each valid (non-padded) row in the batch.

The size of this vector is at most requested_batch_size. The info at index i describes the array at features[i].

Definition at line 134 of file ac_data_iterator.hpp.

◆ features

neural_net::shared_float_array turi::activity_classification::data_iterator::batch::features

An array with shape: (requested_batch_size, 1, prediction_window * predictions_in_chunk, num_feature_columns)

Each row is a chunk of feature values from one session.

Definition at line 97 of file ac_data_iterator.hpp.

◆ labels

neural_net::shared_float_array turi::activity_classification::data_iterator::batch::labels

An array with shape: (requested_batch_size, 1, predictions_in_chunk, 1)

Each row is the sequence of class label (indices) from one chunk (labels picked after majority voting).

If no target was specified, then this value is default constructed.

Definition at line 107 of file ac_data_iterator.hpp.

◆ labels_per_row

neural_net::shared_float_array turi::activity_classification::data_iterator::batch::labels_per_row

An array with shape: (requested_batch_size, 1, prediction_window * predictions_in_chunk, 1)

Each row is the sequence of raw class labels (indices) for each individual sample.

Definition at line 126 of file ac_data_iterator.hpp.

◆ weights

neural_net::shared_float_array turi::activity_classification::data_iterator::batch::weights

An array with shape: (requested_batch_size, 1, predictions_in_chunk, 1)

Each row is a sequence of 0 or 1 values indicating whether the corresponding label is padding (0) or refers to actual data (1).

If no target was specified, then this value is default constructed.

Definition at line 117 of file ac_data_iterator.hpp.


The documentation for this struct was generated from the following file: