portal

The portal module provides interactive visualization tools for exploring Nenya latent spaces and imagery.

Classes

class nenya.portal.Image(image, Us, DT, lat=None, lon=None)

Class representing an image and its metadata.

Parameters:

  • image (numpy.ndarray) – The image data

  • Us (tuple) – UMAP coordinates (U0, U1)

  • DT (float) – Temperature difference value

  • lat (float, optional) – Latitude (optional)

  • lon (float, optional) – Longitude (optional)

copy()

Create a deep copy of the Image object.

Returns:

Copied Image object

Return type:

Image

class nenya.portal.OSSinglePortal(table_file, input_Image=None, init_Us=None)

Main class for the interactive portal.

Parameters:

  • table_file (str) – Path to the UMAP table file

  • input_Image (Image, optional) – Optional input image to start with

  • init_Us (list, optional) – Initial UMAP coordinates if no image is provided

__call__(doc)

Add the portal layout to a Bokeh document.

Parameters:

doc (bokeh.document.Document) – Bokeh document

open_files()

Open the HDF5 files containing images.

load_images(tbl_idx)

Load images from disk based on table indices.

Parameters:

tbl_idx (list) – List of table indices to load

Returns:

Tuple of (images, titles)

Return type:

tuple

set_matched(radius)

Find images within radius in UMAP space.

Parameters:

radius (float) – Radius for matching

find_closest_U(Us)

Find the index of the closest point in UMAP space.

Parameters:

Us (tuple) – UMAP coordinates (U0, U1)

Returns:

Index of the closest point

Return type:

int

set_primary_by_objID(obj_ID)

Set the primary image by object ID.

Parameters:

obj_ID (int) – Object ID in the table

reset_from_primary()

Reset the interface based on the primary image.

get_im_empty()

Get an empty image of the correct size.

Returns:

Empty image

Return type:

numpy.ndarray

Example Usage

import os
from nenya.portal import OSSinglePortal, Image
from bokeh.server.server import Server
import numpy as np

# Path to UMAP table
table_file = os.path.join(os.getenv('OS_SST'), 'MODIS_L2', 'Nenya', 'Tables', 'MODIS_Nenya_96clear_v5_DT15.parquet')

# Optional: Create an Image object
img = np.zeros((64, 64))  # Example image
input_Image = Image(img, Us=(0.5, 0.5), DT=2.0, lat=35.0, lon=-120.0)

# Create a session
def get_session(doc):
    sess = OSSinglePortal(table_file, input_Image=input_Image)
    return sess(doc)

# Start Bokeh server
server = Server({'/': get_session}, num_procs=1)
server.start()
print('Opening Bokeh application on http://localhost:5006/')

server.io_loop.add_callback(server.show, "/")
server.io_loop.start()

Portal Components

The portal interface includes:

  1. Primary Image View: - Displays the current selected image - Shows DT, U0, U1 values - Controls for color mapping

  2. UMAP Plot: - Shows the 2D embedding of all images - Highlights the current image and matches - Color-coded by selected metric (LL, DT, etc.)

  3. Gallery: - Displays multiple images from matched set - Navigation controls for viewing more matches

  4. Data Table: - Shows metadata for matched images - Allows selection for setting primary image

  5. Geographic View: - Map showing geographic locations - Points colored by the same metric as UMAP

Technical Details

The portal uses:

  • Bokeh for interactive visualization

  • HDF5 for efficient image access

  • Pandas for table manipulation

  • NumPy for numerical operations

Data Organization:

  • Images are stored in HDF5 files referenced by the table

  • The table contains metadata and UMAP coordinates

  • Images are loaded on-demand to minimize memory usage

Customization:

  • Color schemes can be adjusted

  • Matching radius and transparency are configurable

  • Metrics for coloring can be selected through the UI