practice_sdoml package

Subpackages

• practice_sdoml.modeling package
  • Submodules
  • practice_sdoml.modeling.evaluate module
    • evaluate_and_generate_figures()
  • practice_sdoml.modeling.model module
    • SimpleNet
      • SimpleNet.forward()
  • practice_sdoml.modeling.predict module
    • main()
  • practice_sdoml.modeling.train module
    • train()
  • Module contents

Submodules

practice_sdoml.app module

Deployment and Interactive Demonstrations for SDOML. Built with Gradio, leveraging core modules from practice_sdoml.

practice_sdoml.app.generate_feature_plot(feature_name: str)

practice_sdoml.app.get_data_summary()

practice_sdoml.app.load_dataset_sample()

Loads exploratory dataset sample prioritizing packaged real data.

practice_sdoml.app.run_evaluation()

Evaluates SimpleNet exclusively on the 20% test partition (3,000 samples).

practice_sdoml.app.train_interactive(epochs: int, lr: float, batch_size: int, progress=<gradio.helpers.Progress object>)

Train SimpleNet uploading Gradio loading bar

practice_sdoml.config module

practice_sdoml.dataset module

Data ingestion and preprocessing module with train/test partitioning support.

class practice_sdoml.dataset.DiabetesDataset(csv_path: Path | str | None = None, split: str = 'train', test_size: float = 0.2, random_state: int = 42)

Bases: Dataset

PyTorch Dataset for diabetes_risk.csv supporting stratified train/test splits.

Parameters:

• csv_path (Path | str | None) – Optional explicit path to the CSV file.

• split (str) – Target data partition (‘train’, ‘test’, or ‘all’).

• test_size (float) – Fraction of samples allocated to the test subset.

• random_state (int) – Random seed for deterministic reproducibility.

practice_sdoml.dataset.get_dataloader(batch_size: int = 32, shuffle: bool = True, split: str = 'train') → Tuple[DataLoader, int, int]

Factory helper function to instantiate DiabetesDataset and build a DataLoader.

Parameters:

• batch_size (int) – Number of samples per batch.

• shuffle (bool) – Whether to shuffle samples at every epoch.

• split (str) – Partition identifier (‘train’, ‘test’, or ‘all’).

Returns:

Tuple containing (DataLoader, num_features, num_classes).

Return type:

Tuple[DataLoader, int, int]

practice_sdoml.features module

practice_sdoml.features.main(input_path: Path = PosixPath('/home/runner/work/practice_sdoml_jn/practice_sdoml_jn/data/processed/dataset.csv'), output_path: Path = PosixPath('/home/runner/work/practice_sdoml_jn/practice_sdoml_jn/data/processed/features.csv'))

practice_sdoml.plots module

Graphic utilities for model evaluation module

practice_sdoml.plots.plot_calibration_curve(targets: ndarray, probs: ndarray, num_classes: int, output_path: Path) → None

Creates and saves fiability diagram.

Parameters:

• targets – True labels.

• probs – Predicted probabilities.

• num_classes – Class number.

• output_path – FIle route for saving images.

practice_sdoml.plots.plot_confusion_matrix(targets: ndarray, preds: ndarray, output_path: Path) → None

Creates and saves confusion matrix

Parameters:

• targets – True labels.

• preds – Model predictions.

• output_path – FIle route for saving images.

practice_sdoml.plots.plot_top_loss_samples(sample_losses: ndarray, output_path: Path, top_k: int = 10) → None

Creates a barplot with the worst classifications.

Parameters:

• sample_losses – Loss array per sample.

• output_path – FIle route for saving images.

• top_k – Sample number to show.

Module contents

Paquete principal de practice_sdoml.

class practice_sdoml.DiabetesDataset(csv_path: Path | str | None = None, split: str = 'train', test_size: float = 0.2, random_state: int = 42)

Bases: Dataset

PyTorch Dataset for diabetes_risk.csv supporting stratified train/test splits.

Parameters:

• csv_path (Path | str | None) – Optional explicit path to the CSV file.

• split (str) – Target data partition (‘train’, ‘test’, or ‘all’).

• test_size (float) – Fraction of samples allocated to the test subset.

• random_state (int) – Random seed for deterministic reproducibility.

class practice_sdoml.SimpleNet(input_dim: int, num_classes: int)

Bases: Module

Feedfordward neural network for tabular classification.

This architecture is created of intermediate lineal layers combined with non-linear activation functions (ReLU) to process input characteristics and predict class probability.

Parameters:

• input_dim (int) – Feature number.

• num_classes (int) – Total number of classes to predict.

forward(x)

Executes fordwars pass of the neural network.

It takes an input tensor batch and spread it over the sequential layers in order to generate non-normalized logits.

Parameters:

x (torch.Tensor) – Input tensor with dimensions (batch_size, input_dim)

Returns:

Output logits with dimensions (batch_size, num_classes)

Return type:

torch.Sensor

practice_sdoml.get_dataloader(batch_size: int = 32, shuffle: bool = True, split: str = 'train') → Tuple[DataLoader, int, int]

Factory helper function to instantiate DiabetesDataset and build a DataLoader.

Parameters:

• batch_size (int) – Number of samples per batch.

• shuffle (bool) – Whether to shuffle samples at every epoch.

• split (str) – Partition identifier (‘train’, ‘test’, or ‘all’).

Returns:

Tuple containing (DataLoader, num_features, num_classes).

Return type:

Tuple[DataLoader, int, int]

practice_sdoml.train(epochs: int = 3, lr: float = 0.001)

Trains SimpleNet model using provided data by DiabetesDataset