Datasets

API Documentation

siapy.datasets

The datasets module provides structured containers and utilities for transforming spectral image data into formats optimized for analysis and machine learning. It bridges the gap between raw spectral data and analytical workflows.

import numpy as np
from rich import print

from siapy.datasets.tabular import TabularDataset
from siapy.entities import Shape, SpectralImage, SpectralImageSet

# Create two mock spectral images with dimensions 100x100 pixels and 10 bands
rng = np.random.default_rng(seed=42)
image1 = SpectralImage.from_numpy(rng.random((100, 100, 10)))
image2 = SpectralImage.from_numpy(rng.random((100, 100, 10)))

# Define a region of interest as a rectangular shape (coordinates in pixel space)
# This will select pixels from position (50,50) to (80,80) in both images
rectangle = Shape.from_rectangle(x_min=50, y_min=50, x_max=80, y_max=80)
image1.geometric_shapes.append(rectangle)
image2.geometric_shapes.append(rectangle)

# Combine the images into a SpectralImageSet for batch processing
image_set = SpectralImageSet([image1, image2])

# Initialize the TabularDataset with our image set and process the data
# This extracts pixel data from the regions defined by our shapes
dataset = TabularDataset(image_set)
dataset.process_image_data()

# The dataset is now iterable - we can access each entity (processed region)
for entity in dataset:
    print(f"Processing entity: {entity}")

# Generate tabular data from our processed dataset
# Setting mean_signatures=False preserves individual pixel values instead of averaging them
dataset_data = dataset.generate_dataset_data(mean_signatures=False)

# Convert the tabular data to a pandas DataFrame with multi-level indexing
# This makes it easy to analyze and manipulate the data
df = dataset_data.to_dataframe_multiindex()

print(f"dataset_data: \n{dataset_data}")
print(f"df: \n{df}")