declearn.dataset.tensorflow.TensorflowDataset
Bases: Dataset
Dataset subclass to wrap up 'tensorflow.data.Dataset' instances.
Source code in declearn/dataset/tensorflow/_tensorflow.py
43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 |
|
__init__(dataset, buffer_size=None, batch_mode='default', seed=None)
Wrap up a 'tensorflow.data.Dataset' into a declearn Dataset.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
dataset |
tf.data.Dataset
|
A tensorflow Dataset instance to be wrapped for declearn use. The dataset is expected to yield sample-level records, made of one to three (tuples of) tensorflow tensors: model inputs, target labels and/or sample weights. |
required |
buffer_size |
Optional[int]
|
Optional buffer size denoting the number of samples to pre-fetch
and shuffle when sampling from the original dataset. The higher,
the better the shuffling, but also the more memory costly.
If None, use context-based |
None
|
batch_mode |
BatchMode
|
Flag specifying how to batch inputs. Use "padded" or "ragged" to
batch up variable-dimension samples (e.g. sequences of tokens),
using either |
'default'
|
seed |
Optional[int]
|
Optional seed for the random number generator based on which the dataset is (optionally) shuffled when generating batches. Note that successive batch-generating calls will not yield the same results, as the seeded state is not reset on each call. |
None
|
Notes
The wrapped tensorflow.data.Dataset
:
- must have a fixed length (with TensorFlow <2.13) / should
have an established
cardinality
(TensorFlow >=2.13). - should return sample-level (unbatched) elements, as either:
- (inputs,)
- (inputs, labels)
- (inputs, labels, weights) where each element may be a (nested structure of) tensor(s).
- when using
declearn.model.tensorflow.TensorflowModel
:- inputs may be a single tensor or list of tensors
- labels may be a single tensor or None (usually, not None)
- weights may be a single tensor or None
Source code in declearn/dataset/tensorflow/_tensorflow.py
47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 |
|