distmetrics v1.0.6

Fixed

• Resolves #71 - median drift in cusum was incorrect.

distmetrics v1.0.5

Added

• Function to organize burst time series into windowed data

Changed

• Removed view/permute in favor of more readable einops to ensure self-documenting reshaping, transposing, etc

Fixed

• Notebook's curation of data for burst application.
• Proper model validation for mahalanobis.py pydantic models - removes warnings during test time.
• Update environment for gpu to utilize regex for cuda compatibility.

distmetrics v1.0.4

Fixed

• Fixed asf search for notebook examples:
  • filters out dates with single polarization
  • allows users to select which dual polarization to use (HH+HV vs. VV+VH)
• Baseline curation matches dist-s1.

Added

• Add models for v0 and v1 that use 32 x 32 input size.
• Dynamically update input_size (i.e. the size of the input image) for inference using the corresponding model attribute.
• Expose fill value of model inference (via fill_value) that was reminant of old logit transformation and is now explicitly set and can be modified

distmetrics v1.0.3

Changed

• Latest weights for transformer_anniversary_trained_optimized_fine and transformer_anniversary_trained_optimized the v1 datasets.

distmetrics v1.0.2

Added

• New models from v1 dataset.

distmetrics v1.0.1

Changed

• Load model config and obtain maximum sequence length (context length) of model.
• Decouple loading of config and weights.

Fixed

• Changelog action

Added

• tests for external model loading

distmetrics v1.0.0

Added

• Nodata interpolation for interior datapoints (i.e. within pixels collected during acquisition) which are np.nan including:
  • bilinear - an iterative bilinear-like interpolation
  • nearest - nearest neighbor interpolation
• dependabot
• Several models for easier testing
• notebooks that can do both immediate and anniversary selection to establish baseline

Changed

• Removes logit transformation from normal param estimation
• 4 new models for testing
• estimate_normal_params_of_logits --> estimate_normal_params
• Reorganization of transformer metric code into loading model(s), inference, torch model, and metric.
• Updated notebook so we only download necessary arrays.

Removed

• Notebook for multiple burst analysis (will add back later)

distmetrics v0.0.14

Fixed

• Version/Changelog

distmetrics v0.0.13

Added

• 'external' option for model_token parameter of load_transformer_model.
• And the associated cfg and wts file paths.

distmetrics v0.0.12

Added

• Test to ensure cpu and None interface work correctly for transformer metric.

Fixed

• The device argument for transformer needs to be correctly passed throughout the transformer estimation process and test suite.