API Documentation¶
A Python package for assisting neural network production with TensorFlow.
Batch Loaders¶
Value JSON:¶
| Name | Type | Required | Description |
|---|---|---|---|
dataset |
column 2 | column 3 | column 4 |
| body row 2 | … | … |
-
class
nnetmaker.IndependentBatchLoader(in_names, target_names, loader_args_dict, **kwargs)¶ A loader for iterating over batches of independent examples stored as TFRecord files in the specified data directory.
-
class
nnetmaker.ContinuousSequenceBatchLoader(in_names, target_names, loader_args_dict, **kwargs)¶ A loader for iterating over batches of subsequences selected from large continuous sequences of data. Each
TFRecordfile stores a sequence of contiguous data chunks that together comprise a sequence containing no boundaries within. Each file is considered an independent source, and batch subsequences are selected by concatenating all chunks along the first dimension then selecting windows of the concatenated data. All features for each chunk must have the same length along the first dimension.
-
class
nnetmaker.DiscreteSequenceBatchLoader(in_names, target_names, loader_args_dict, **kwargs)¶ A loader for iterating over batches of subsequences selected from large discrete sequences of data. Each TFRecord file stores a sequence of contiguous subsequences that each represent a single unit and together comprise a longer sequence. Each file is considered an independent source, and batch subsequences are selected by concatenating all unit subsequences along the first dimension then selecting a number of contiguous units, aligned at unit subsequence boundaries. Unit subsequence features may have different lengths along the first dimension.