IGTD
The Image Generator for Tabular Data (IGTD) method transforms tabular data into images by arranging features according to their similarity. Initially, it creates a similarity matrix using techniques such as Pearson or Spearman correlation to evaluate the relationships between features. Subsequently, it computes a matrix that represents the spatial distances between pixel positions in the image. Finally, IGTD rearranges the features to ensure that the layout of the similarity matrix closely matches the spatial arrangement in the image, refining the placement iteratively.
Import IGTD
To import IGTD model use:
>>> from TINTOlib.igtd import IGTD
>>> model = IGTD()
Inherited base functionality
IGTD inherits from MappingMethod (see Base classes), so it also provides:
Feature coordinates: a feature-to-pixel mapping exported to
features_positions.csvafterfit/fit_transform.Programmatic access:
_get_features_mapping()returns the feature coordinates as a DataFrame once the model is fitted.Shared utilities:
saveHyperparameters/loadHyperparametersand the standardfit/transform/fit_transformworkflow.
Hyperparameters & Configuration
When creating the IGTD class, some parameters can be modified. The parameters are:
Parameters |
Description |
Default value |
Valid values |
|---|---|---|---|
|
The type of problem, defining how the images are grouped. |
‘classification’ |
[‘classification’, ‘unsupervised’, ‘regression’] |
|
Preprocessing transformations like scaling, normalization, etc. |
MinMaxScaler() |
Scikit Learn transformers or custom implementation inheriting CustomTransformer. |
|
Show execution details in the terminal. |
False |
[True, False] |
|
The number of pixel rows and columns in the image. The product (rows x columns) must be greater than or equal to the number of features. |
[6, 6] |
list of two positive integers |
|
Method to calculate pairwise distances between features. |
‘Pearson’ |
[‘Pearson’, ‘Spearman’, ‘set’, ‘Euclidean’] |
|
Method to calculate distances between pixels in the image. |
‘Euclidean’ |
[‘Euclidean’, ‘Manhattan’] |
|
Function to evaluate differences between feature and pixel distance rankings. |
‘squared’ |
[‘squared’, ‘abs’] |
|
Maximum number of iterations for the algorithm if it does not converge. |
1000 |
integer |
|
Number of steps to check gain on the objective function for convergence. |
50 |
integer |
|
Threshold for error change rate to determine if switching features should occur. |
0 |
integer |
|
Minimum improvement in the objective function to continue optimization. |
0.00001 |
float |
|
Multiplication factor determining the size of the saved image relative to the original size. |
1 |
integer > 0 |
|
Output format using images with matplotlib with [0,255] range for pixel or using npy format. |
‘png’ |
[‘png’, ‘npy’] |
|
Color map to use with matplotlib. |
‘gray’ |
‘viridis’, ‘plasma’, ‘inferno’, ‘magma’, ‘cividis’, ‘Greys’, etc. |
|
Seed for reproducibility. |
1 |
integer |
Code example:
>>> model = IGTD(scale=[3,3], error="abs", val_step=60, cmap='viridis')
All the parameters that aren’t specifically set will have their default values.
Functions
IGTD has the following functions:
Function |
Description |
Output |
|---|---|---|
|
Allows to save the defined parameters. |
.pkl file with the configuration |
|
Load IGTD configuration previously saved with
|
|
|
Trains the model on the tabular data. This optimizes the feature arrangement matrix based on the chosen distance methods and error function.
|
|
|
Generates and saves synthetic images in a specified folder. Requires the model to be fitted first.
|
Folders with synthetic images |
|
Combines the training and image generation steps. Fits the model to the data and generates synthetic images in one step.
|
Folders with synthetic images |
The model must be fitted before using the transform method. If the model isn’t fitted, a RuntimeError will be raised.
Citation
Paper: https://doi.org/10.1038/s41598-021-90923-y
Code Repository: https://github.com/zhuyitan/igtd