Skip to content

Latest commit

 

History

History
188 lines (96 loc) · 6.08 KB

CHANGELOG.md

File metadata and controls

188 lines (96 loc) · 6.08 KB
Raw

openproblems-v2 0.1.0

common

NEW FUNCTIONALITY

  • extract_scores: Summarise a metrics output tsv.

  • dataset_concatenate: Concatenate N AnnData datasets.

  • Created test data resources_test/pancreas with src/common/resources_test_scripts/pancreas.sh.

  • list_git_shas: create list of latest commit hashes of all files in repo.

  • get_api_info: extract api info from tasks

  • get_method_info: extract method info from config yaml

  • get_metric_info: extract metric info from config yaml

  • check_migration_status: compare git shas from methods with v1

  • get_results: extract benchmark scores

  • get_task_info: extract task info

  • unit_test: Common unit test that can be used by all tasks

  • check_dataset_schema: check if the dataset used has the required fields defined in the api anndata_*.yaml files

datasets

NEW FUNCTIONALITY

  • workflows/process_openproblems_v1: Fetch and process legacy OpenProblems v1 datasets, whilst adding extra information to the .uns.

  • normalization/log_cpm: A log CPM normalization method.

  • normalization/log_scran_pooling: A log scran pooling normalization method.

  • normalization/sqrt_cpm: A sqrt CPM normalization method.

  • normalization/l1_sqrt: A scaled L1 sqrt normalization. extracted from Alra method in the denoising task from v1

  • subsample: Subsample an h5ad file. Allows keeping observations from specific batches and celltypes, also allows keeping certain features.

V1 MIGRATION

  • loaders/openproblems_v1: Fetch a dataset from OpenProblems v1, whilst adding extra information to the .uns.

  • loaders/openproblems_v1_multimodal: Fetch a multimodal dataset from OpenProblems v1, whilst adding extra information to the .uns.

label_projection

NEW FUNCTIONALITY

  • api/anndata_*: Created a file format specifications for the h5ad files throughout the pipeline.

  • api/comp_*: Created an api definition for the split, method and metric components.

  • split_dataset: Added a component for splitting raw datasets into task-ready dataset objects.

  • resources_test/label_projection/pancreas with src/label_projection/resources_test_scripts/pancreas.sh.

V1 MIGRATION

  • methods/knn: Migrated from v1.

  • methods/logistic_regression: Migrated from v1.

  • methods/mlp: Migrated from v1.

  • methods/scanvi: Migrated and adapted from v1.

  • methods/seurat_transferdata: Migrated and adapted from v1.

  • methods/xgboost: Migrated from v1.

  • control_methods/majority_vote: Migrated from v1.

  • control_methods/random_labels: Migrated from v1.

  • control_methods/true_labels: Migrated from v1.

  • metric/accuracy: Migrated from v1.

  • metric/f1: Migrated from v1.

denoising

NEW FUNCTIONALITY

  • api/anndata_*: Created a file format specifications for the h5ad files throughout the pipeline.

  • api/comp_*: Created an api definition for the split, method and metric components.

  • split_dataset: Added a component for splitting raw datasets into task-ready dataset objects.

  • resources_test/denoising/pancreas with src/denoising/resources_test_scripts/pancreas.sh.

V1 MIGRATION

  • control_methods/no_denoising: Migrated from v1. Extracted from baseline method

  • control_methods/perfect_denoising: Migrated from v1.Extracted from baseline method

  • methods/alra: Migrated from v1. Changed from python to R and uses lg_cpm normalised data instead of L1 sqrt

  • methods/dca: Migrated and adapted from v1.

  • methods/knn_smoothing: Migrated and adapted from v1.

  • methods/magic: Migrated from v1.

  • metrics/mse: Migrated from v1.

  • metrics/poisson: Migrated from v1.

Changes from V1

  • Anndata layers are used to store data instead of obsm

  • extended the use of sparse data in methods unless it was not possible

  • split_dataset also removes unnecessary data from train and test datasets not needed by the methods and metrics.

Dimensionality reduction

New functionality

  • api/anndata_*: Created a file format specifications for the h5ad files throughout the pipeline.

  • api/comp_*: Created an api definition for the split, control method, method and metric components.

  • split_dataset: Added a component for splitting raw datasets into task-ready dataset objects.

  • control_methods: Added a component for baseline methods specifically.

  • resources_test/dimensionality_reduction/pancreas with src/dimensionality_reduction/resources_test_scripts/pancreas.sh.

  • Added variant key to config files to store variants (different input parameters) of every component.

V1 migration

  • control_methods/true_features: Migrated from v1. Extracted from baseline method True Features.

  • control_methods/random_features: Migrated from v1. Extracted from baseline method Random Features.

  • methods/umap: Migrated from v1.

  • methods/ivis: Migrated from v1.

  • methods/tsne: Migrated and adapted from v1.

  • methods/densmap: Migrated and adapted from v1.

  • methods/phate: Migrated from v1.

  • methods/pca: Migrated from v1.

  • methods/neuralee: Migrated from v1.

  • metrics/rmse: Migrated from v1, but will likely be removed.

  • metrics/trustworthiness: Migrated from v1, but will likely be removed.

  • metrics/density_preservation: Migrated from v1.

  • metrics/coranking: Migrated from v1. This script originally called nn_ranking.py and written in Python.

Changes from V1

  • Raw counts and normalized expression data is stored in .layers["counts"] and .layers["normalized"], respectively, instead of in .X.

  • A split_dataset has been implemented to make a distinction between the data a method is allowed to see (here called the train data) and what a metric is allowed to see (here called the test data).

  • methods/ivis had originally been removed from the v1 (temporarily) but has been added back to the v2.

  • The metrics as defined in the nn_ranking.py script have been documented and refactored into an R component metrics/coranking.

  • metrics/rmse should be removed because RMSE metrics don't really make sense here.

  • metrics/trustworthiness should be removed because it is already included in metrics/coranking.