brails.processors.occupancy_classifier.occupancy_classifier module

class brails.processors.occupancy_classifier.occupancy_classifier.OccupancyClassifier(input_data: dict | None = None)

Bases: ImageClassifier

The OccupancyClassifier classifier attempts to predict occupancy types of buildings as 1 of 2 types: Residential or Other. The classification is done by the ImageClassifier class. This class is a wrapper that just sets up the inputs for that class.

Variables

model_path: str Path to the model file

Methods:
predict(ImageSet): To return the predictions for the set of images

provided

predict(images: ImageSet) → dict

The method that predicts building occupancy type.

Args
images: ImageSet containing the set of images for which a

prediction is required

Returns
dict: Dictionary with keys set to image names in ImageSet.images

and values set to the predicted occupancy types

retrain(dataDir, batchSize=8, nepochs=100, plotLoss=True)

Retrain existing model using training dataset and hyperparameters.

Parameters__ - model_path (str): Path to the pre-trained model to be fine-tuned.

Default is ‘tmp/models/trained_model.pth’.

  • train_data_dir (str): Directory containing the training and

    validation datasets. Default is ‘tmp/hymenoptera_data’.

  • model_inp_size (int): Input size for the model, used for resizing

    images in the dataset. Default is 384.

  • batch_size (int): Batch size for data loading. Default is 32.

  • nepochs (Union[int, List[int]]): Number of epochs for training.

    Should be an integer for retraining. Default is 100.

  • es_tolerance (int): Early stopping tolerance; the number of epochs

    without improvement before stopping. Default is 10.

  • plot_accuracy (bool): Whether to plot the validation accuracy over

    epochs. Default is True.

Returns__ - None

Raises__ - ValueError: If nepochs is not provided as an integer during

retraining.

Procedure__ 1. Loads the model from the specified path. 2. Applies data augmentation and normalization to the training dataset

and normalization to the validation dataset.

  1. If the default training data directory is used, downloads a sample

    dataset.

  2. Prepares PyTorch DataLoader objects for the training and

    validation datasets.

  3. Sends the model to the GPU and fine-tunes it using Stochastic

    Gradient Descent (SGD) optimization.

  4. After training, the model is saved at the specified path.

  5. If plot_accuracy is True, plots the validation accuracy versus

    training epochs.

Example__ >>> classifier = ImageClassifier() >>> classifier.retrain(model_path=’model.pth’,

train_data_dir=’my_data’, nepochs=50)