deepBlink v.0.0.5

Overview

“It works fine on MY computer”

Those days are over as it's testing time. While the test-suite is not fully complete it gives us some more confidence over our commits.
Additional changes in this release include:

• Change from prediction matrix to coordinate list as default dataset component. Accompanying changes include the use of cell_size in the dataset class and a dynamic Resnet architecture.
• Scipy requirement for a new deepblink.metrics.f1_cutoff_score way of calculating the F1 score directly from coordinates.
• Depreciation of deepblink.metrics.compute_score and deepblink.metrics.weighted_f1_coordinates.
• Addition of deepblink.optimizers.amsgrad function.
• Movement of functions:
  • train_valid_split to util
  • load_image to io

The next release will include benchmarking scripts and is most likely the last release before the first minor 0.1.0 and non-pre-release.

---

Authors

• @BBQuercus
• @zhanyinx