Skip to content

Train 2D images

Short Description

Train an autoencoder (a deep learning model) on a given dataset for selected marker/markers. The autoencoder learns to compress the dataset images into a lower-dimensional encoding and then reconstruct them from this encoding. To train the aeTrain or aeTrainMulti model, simply direct the function to the dataset_dir folder.

Function

aeTrain(dataset_dir, outModelPath, input_dim=256, encoding_dim=64, max_epoch_num=100, batch_size=8, num_workers=10, prefetch_factor=8)

Parameters:

Name Type Description Default
dataset_dir str

The file path leading to the directory that holds the training data.

 required
outModelPath str

The file path where the trained model will be saved.

 required
input_dim int

The input dimension of the model, the input images assumed to be square, i.e., size x size. Default input dimensio is 256, which corresponds to an input image size of 16x16 pixels.

 256
encoding_dim int

The size of the encoding layer, representing the dimensionality of the feature encoding learned by the autoencoder.

64
max_epoch_num int

The maximum number of epochs to run during training.

100
batch_size int

The number of images in each batch of dataloader.

 8
num_workers int

The number of worker subprocesses for loading the data. A value of 0 means that the data will be loaded in the main process.

 10
prefetch_factor int

The number of batches to load in advance by each worker.

8

Returns:

Name Type Description
model model

The function saves the trained model to the specified path.
Example 
import spatialae as sa

# Define the parameters
input_dim = 256 # 16*16

# Replace with the path to your dataset directory
dataset_dir="/n/scratch/users/r/roh6824/Results/CRC12image_update/SpatialAE/SinglePatch/DNA1/"

# Replace with your desired output model file path
outModelPath = '/n/scratch/users/r/roh6824/Results/CRC12image_update/SpatialAE/ln_autoencoder_DNA_validate_300_model.pth'

# Train the autoencoder
sa.aeTrain(dataset_dir, outModelPath,  input_dim, encoding_dim,  max_epoch_num, batch_size)
Source code in spatialae/models/aeTrain.py 
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
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
def aeTrain(dataset_dir,
            outModelPath,
            input_dim = 256,
            encoding_dim = 64,
            max_epoch_num = 100,
            batch_size = 8,
            num_workers = 10,
            prefetch_factor = 8
            ):
    """
Parameters:
    dataset_dir (str):
        The file path leading to the directory that holds the training data.

    outModelPath (str):
        The file path where the trained model will be saved.

    input_dim (int, optional):
        The input dimension of the model, the input images assumed to be square, i.e., size x size. Default input dimensio is 256, which corresponds to an input image size of 16x16 pixels.

    encoding_dim (int, optional):
        The size of the encoding layer, representing the dimensionality of the feature encoding learned by the autoencoder. 

    max_epoch_num (int, optional):
        The maximum number of epochs to run during training. 

    batch_size (int, optional):
        The number of images in each batch of dataloader.

    num_workers (int, optional):
        The number of worker subprocesses for loading the data. A value of 0 means that the data will be loaded in the main process.

    prefetch_factor (int, optional):
        The number of batches to load in advance by each worker. 

Returns:
    model (model): 
        The function saves the trained model to the specified path.

Example:
    ```python

    import spatialae as sa

    # Define the parameters
    input_dim = 256 # 16*16

    # Replace with the path to your dataset directory
    dataset_dir="/n/scratch/users/r/roh6824/Results/CRC12image_update/SpatialAE/SinglePatch/DNA1/"

    # Replace with your desired output model file path
    outModelPath = '/n/scratch/users/r/roh6824/Results/CRC12image_update/SpatialAE/ln_autoencoder_DNA_validate_300_model.pth'

    # Train the autoencoder
    sa.aeTrain(dataset_dir, outModelPath,  input_dim, encoding_dim,  max_epoch_num, batch_size)

    ```

    """
    model = LitAutoEncoder(input_dim, encoding_dim)
    # Load the dataset
    train_dataset = SpatialImageDataset(dataset_dir, get_train = True)
    train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers = num_workers, prefetch_factor = prefetch_factor)

    validate_dataset = SpatialImageDataset(dataset_dir, get_validate = True)
    validate_loader = DataLoader(validate_dataset, batch_size=batch_size, shuffle=False, num_workers = num_workers, prefetch_factor = prefetch_factor)
    # train the model
    trainer = pl.Trainer(max_epochs=max_epoch_num)
    trainer.fit(model, train_loader, validate_loader)

    # save the trained model
    torch.save(model.state_dict(), outModelPath)

aeTrainMulti(dataset_dir, outModelPath, channels, input_dim=256, encoding_dim=64, max_epoch_num=100, batch_size=8, num_workers=10, prefetch_factor=8)

Parameters:

Name Type Description Default
dataset_dir str

The file path to the directory containing the training dataset. Each subdirectory within is expected to represent a marker.

 required
outModelPath str

The file path where the trained model will be saved.

 required
channels list

A list of strings representing the markers/channels to be used for model training. If a user wants to train the model on specific channels, they can specify them by name (e.g., ['CD3D', 'CD4']). Only data corresponding to these channels will be used for training.

 required
input_dim int

The input dimension of the model, the input images assumed to be square, i.e., size x size. Default input dimensio is 256, which corresponds to an input image size of 16x16 pixels.

 256
encoding_dim int

The size of the encoding layer, representing the dimensionality of the feature encoding learned by the autoencoder. Default is 64.

 64
max_epoch_num int

The maximum number of epochs to run during training. Default is 100.

 100
batch_size int

The number of images in each batch of dataloader. Default is 8.

 8
num_workers int

The number of worker subprocesses for loading the data. Default is 10. A value of 0 means that the data will be loaded in the main process.

 10
prefetch_factor int

The number of batches to load in advance by each worker. Default is 8.

 8

Returns:

Name Type Description
model model

The function saves the trained model to the specified path.
Example 
import spatialae as sa

# Define the parameters
channels = ["DNA1", "CD3", "KERATIN", "CD20", "CD68","CD8A", "CD163","ECAD", "CD31"]
input_dim=256
encoding_dim=64
batch_size=32
max_epoch_num=300

# Replace with the path to your dataset directory
dataset_dir=/n/scratch/users/r/roh6824/Results/CRC12image_update/SpatialAE/SinglePatch/

# Replace with your desired output model file path
outModelPath=/n/scratch/users/r/roh6824/Results/CRC12image_update/SpatialAE/ln_autoencoder_multi_validate_300_model_dim32.pth

# Train the autoencoder
sa.aeTrainMulti(dataset_dir, outModelPath, channels, input_dim, encoding_dim, max_epoch_num, batch_size)
Source code in spatialae/models/aeTrain.py 
 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
def aeTrainMulti(dataset_dir,
                 outModelPath,
                 channels,
                 input_dim = 256,
                 encoding_dim = 64,
                 max_epoch_num = 100,
                 batch_size = 8,
                 num_workers = 10,
                 prefetch_factor = 8
                 ):
    """
Parameters:
    dataset_dir (str):
        The file path to the directory containing the training dataset.
        Each subdirectory within is expected to represent a marker.

    outModelPath (str):
        The file path where the trained model will be saved.

    channels (list):
        A list of strings representing the markers/channels to be used for model training.
        If a user wants to train the model on specific channels, they can specify them
        by name (e.g., ['CD3D', 'CD4']). Only data corresponding to these channels will be
        used for training.

    input_dim (int, optional):
        The input dimension of the model, the input images assumed to be square, i.e., size x size.
        Default input dimensio is 256, which corresponds to an input image size of 16x16 pixels.

    encoding_dim (int, optional):
        The size of the encoding layer, representing the dimensionality of the 
        feature encoding learned by the autoencoder. Default is 64.

    max_epoch_num (int, optional):
        The maximum number of epochs to run during training. Default is 100.

    batch_size (int, optional):
        The number of images in each batch of dataloader. Default is 8.

    num_workers (int, optional):
        The number of worker subprocesses for loading the data. Default is 10.
        A value of 0 means that the data will be loaded in the main process.

    prefetch_factor (int, optional):
        The number of batches to load in advance by each worker. Default is 8.

Returns:
    model (model): 
        The function saves the trained model to the specified path.


Example:
    ```python

    import spatialae as sa

    # Define the parameters
    channels = ["DNA1", "CD3", "KERATIN", "CD20", "CD68","CD8A", "CD163","ECAD", "CD31"]
    input_dim=256
    encoding_dim=64
    batch_size=32
    max_epoch_num=300

    # Replace with the path to your dataset directory
    dataset_dir=/n/scratch/users/r/roh6824/Results/CRC12image_update/SpatialAE/SinglePatch/

    # Replace with your desired output model file path
    outModelPath=/n/scratch/users/r/roh6824/Results/CRC12image_update/SpatialAE/ln_autoencoder_multi_validate_300_model_dim32.pth

    # Train the autoencoder
    sa.aeTrainMulti(dataset_dir, outModelPath, channels, input_dim, encoding_dim, max_epoch_num, batch_size)

    ```
    """
    input_dim = input_dim*len(channels)
    model = LitAutoEncoder(input_dim, encoding_dim)
    # Load the dataset
    train_dataset = MultiChannelImageDataset(dataset_dir, channels, get_train = True)
    train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers = num_workers, prefetch_factor = prefetch_factor)

    validate_dataset = MultiChannelImageDataset(dataset_dir, channels, get_validate = True)
    validate_loader = DataLoader(validate_dataset, batch_size=batch_size, shuffle=False, num_workers = num_workers, prefetch_factor = prefetch_factor)
    # train the model
    trainer = pl.Trainer(max_epochs=max_epoch_num)
    trainer.fit(model, train_loader, validate_loader)
    # save the trained model
    torch.save(model.state_dict(), outModelPath)