segger.training¶
training module for Segger.
Contains the implementation of the Segger model using Graph Neural Networks.
LitSegger ¶
LitSegger(**kwargs)
Bases: LightningModule
LitSegger is a PyTorch Lightning module for training and validating the Segger model.
Attributes¶
model : Segger The Segger model wrapped with PyTorch Geometric's to_hetero for heterogeneous graph support. validation_step_outputs : list A list to store outputs from the validation steps. criterion : torch.nn.Module The loss function used for training, specifically BCEWithLogitsLoss.
Initializes the LitSegger module with the given parameters.
Parameters¶
**kwargs : dict Keyword arguments for initializing the module. Specific parameters depend on whether the module is initialized with new parameters or components.
Source code in src/segger/training/train.py
32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 |
|
configure_optimizers ¶
configure_optimizers()
Configures the optimizer for training.
Returns¶
torch.optim.Optimizer The optimizer for training.
Source code in src/segger/training/train.py
207 208 209 210 211 212 213 214 215 216 217 |
|
forward ¶
forward(batch)
Forward pass for the batch of data.
Parameters¶
batch : SpatialTranscriptomicsDataset The batch of data, including node features and edge indices.
Returns¶
torch.Tensor The output of the model.
Source code in src/segger/training/train.py
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 |
|
from_components ¶
from_components(model)
Initializes the LitSegger module with existing Segger components.
Parameters¶
model : Segger The Segger model to be used.
Source code in src/segger/training/train.py
101 102 103 104 105 106 107 108 109 110 |
|
from_new ¶
from_new(num_tx_tokens, init_emb, hidden_channels, out_channels, heads, num_mid_layers, aggr, metadata)
Initializes the LitSegger module with new parameters.
Parameters¶
num_tx_tokens : int Number of unique 'tx' tokens for embedding (this must be passed here). init_emb : int Initial embedding size. hidden_channels : int Number of hidden channels. out_channels : int Number of output channels. heads : int Number of attention heads. aggr : str Aggregation method for heterogeneous graph conversion. num_mid_layers: int Number of hidden layers (excluding first and last layers). metadata : Union[Tuple, Metadata] Metadata for heterogeneous graph structure.
Source code in src/segger/training/train.py
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 |
|
training_step ¶
training_step(batch, batch_idx)
Defines the training step.
Parameters¶
batch : Any The batch of data. batch_idx : int The index of the batch.
Returns¶
torch.Tensor The loss value for the current training step.
Source code in src/segger/training/train.py
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 |
|
validation_step ¶
validation_step(batch, batch_idx)
Defines the validation step.
Parameters¶
batch : Any The batch of data. batch_idx : int The index of the batch.
Returns¶
torch.Tensor The loss value for the current validation step.
Source code in src/segger/training/train.py
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 |
|
Segger ¶
Segger(num_tx_tokens, init_emb=16, hidden_channels=32, num_mid_layers=3, out_channels=32, heads=3)
Bases: Module
Initializes the Segger model.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
num_tx_tokens |
int)
|
Number of unique 'tx' tokens for embedding. |
required |
init_emb |
int)
|
Initial embedding size for both 'tx' and boundary (non-token) nodes. |
16
|
hidden_channels |
int
|
Number of hidden channels. |
32
|
num_mid_layers |
int)
|
Number of hidden layers (excluding first and last layers). |
3
|
out_channels |
int)
|
Number of output channels. |
32
|
heads |
int)
|
Number of attention heads. |
3
|
Source code in src/segger/models/segger_model.py
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 |
|
decode ¶
decode(z, edge_index)
Decode the node embeddings to predict edge values.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
z |
Tensor
|
Node embeddings. |
required |
edge_index |
EdgeIndex
|
Edge label indices. |
required |
Returns:
Name | Type | Description |
---|---|---|
Tensor |
Tensor
|
Predicted edge values. |
Source code in src/segger/models/segger_model.py
81 82 83 84 85 86 87 88 89 90 91 92 |
|
forward ¶
forward(x, edge_index)
Forward pass for the Segger model.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
x |
Tensor
|
Node features. |
required |
edge_index |
Tensor
|
Edge indices. |
required |
Returns:
Name | Type | Description |
---|---|---|
Tensor |
Tensor
|
Output node embeddings. |
Source code in src/segger/models/segger_model.py
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 |
|
SpatialTranscriptomicsDataset ¶
SpatialTranscriptomicsDataset(root, transform=None, pre_transform=None, pre_filter=None)
Bases: InMemoryDataset
A dataset class for handling SpatialTranscriptomics spatial transcriptomics data.
Attributes:
Name | Type | Description |
---|---|---|
root |
str
|
The root directory where the dataset is stored. |
transform |
callable
|
A function/transform that takes in a Data object and returns a transformed version. |
pre_transform |
callable
|
A function/transform that takes in a Data object and returns a transformed version. |
pre_filter |
callable
|
A function that takes in a Data object and returns a boolean indicating whether to keep it. |
Initialize the SpatialTranscriptomicsDataset.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
root |
str
|
Root directory where the dataset is stored. |
required |
transform |
callable
|
A function/transform that takes in a Data object and returns a transformed version. Defaults to None. |
None
|
pre_transform |
callable
|
A function/transform that takes in a Data object and returns a transformed version. Defaults to None. |
None
|
pre_filter |
callable
|
A function that takes in a Data object and returns a boolean indicating whether to keep it. Defaults to None. |
None
|
Source code in src/segger/data/utils.py
516 517 518 519 520 521 522 523 524 525 |
|
processed_file_names
property
¶
processed_file_names
Return a list of processed file names in the processed directory.
Returns:
Type | Description |
---|---|
List[str]
|
List[str]: List of processed file names. |
raw_file_names
property
¶
raw_file_names
Return a list of raw file names in the raw directory.
Returns:
Type | Description |
---|---|
List[str]
|
List[str]: List of raw file names. |
download ¶
download()
Download the raw data. This method should be overridden if you need to download the data.
Source code in src/segger/data/utils.py
545 546 547 548 |
|
get ¶
get(idx)
Get a processed data object.
Parameters:
Name | Type | Description | Default |
---|---|---|---|
idx |
int
|
Index of the data object to retrieve. |
required |
Returns:
Name | Type | Description |
---|---|---|
Data |
Data
|
The processed data object. |
Source code in src/segger/data/utils.py
563 564 565 566 567 568 569 570 571 572 573 574 |
|
len ¶
len()
Return the number of processed files.
Returns:
Name | Type | Description |
---|---|---|
int |
int
|
Number of processed files. |
Source code in src/segger/data/utils.py
555 556 557 558 559 560 561 |
|
process ¶
process()
Process the raw data and save it to the processed directory. This method should be overridden if you need to process the data.
Source code in src/segger/data/utils.py
550 551 552 553 |
|