All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- RasterMath get_image_as_array() now supports mask.
- Update groups management for sklearn>=0.25
- Fixed bug with raster/vector datasets (mtb.datasets.load_historical_data())
- Fixed bug with get_image_as_array() from RasterMath (completely rewrite this part)
- Requirements is not directly written inside setup.py due to bugs.
- Fix bug in setup.py using requirements.txt instead of ./requirements.txt
- Adding psutil to depency
- get_image_as_array function for RasterMath
- train_test_split supports now groups=None
- Support list for cross-validation in order to give an unready unfolded cv.
New features provided by @marclang for the charts module :
- Allows to display both F1 and accuracy or mean metrics
- Allows to display accuracy after have been displaying mean (and vice versa)
- Allows to display float matrix
- Fix path separator to access tutorial dataset
- Final version for JOSS (paper.md and paper.bib updated thanks to @kbarnhart)
- RasterMath use available memory to speed up process and manage now several cores (n_jobs)
- train_test_split in cross_validation module
- Enhance mask management for RasterMath
- Move FlushCache to optimize RasterMath
- RasterMath get_random_block returns only block which are not totally unmasked
- charts.PlotConfusionMatrix has a default argument (zero_is_min=True)
- Fix bug when in RasterMath when input is only one band
- Fix bug in RasterMath with mask and list
- n_jobs for RasterMath (thanks to Helene @HTDBD and Arthur @ArthurDfs, two great students)
- function write_block and generally a most intuitive way to use RasterMath (with the help of @HTDBD and @ArthurDfs)
- new branch spatial added
- Added this line
- SequentialFeatureSelection parameters order Changed. scoring is now before standardize.
- Update doc for load_historical_data()
- Fix bug in get_block() and get_random_block() which returned the same block each time due to new method.
- Fix bug with nodata in RasterMath when output is of float type
- RasterMath made a lot of improvements using block reading and writing. For example, the default block size is now 256x256 (you can keep the default block size by choosing block_size=False), and Museo ToolBox automatic detect if the geotiff will be tiled or not (it depends on the block size).
- Some folders have Changed name :
- raster_tools and vector_tools to processing
- learn_tools to ai
- some functions have Changed name :
- getSamplesFromROI to extract_values
- historicalMap to load_historical_data
- getDistanceMatrix to get_distance_matrix
- classes now always begin with a capital case :
- learnAndPredict to SuperLearner
- rasterMath to RasterMath
- sequentialFeatureSelection to SequentialFeatureSelection
- bug #7 : getSamplesFromROI (nowd extract_ROI) now extracts ROI values using by default memory. If it fails, it will create a temporary raster on disk then delete it when finished.
- Remove command lines (cli)
- getSamplesFromROI return list of available fields if wrong field given.
- rasterMath convert np.nan value to nodata value (if numpy >= 1.17)
- Minor fix when using learnAndPredict with an outside customized function
- Better management fo cross-validation in learnAndPredict
- Fix minor bug using False or None value with cv in learnAndPredict
- Add an option to use SFS without writing each best model on the disk.
- Move some functions from vector_tools to raster_tools, functions are anyway still available from vector_tools
- learnAndPredict manages int value for cross-validation by using RandomStratifiedKFold
- Enhance blocksize management for rasterMath
- Move command line code in _cli folder
- Improvements of rasterMath
- customBlockSize defines now the same block size for window reading and for the output
- add seed parameter (to set a random generator) in getRandomBlock()
- add getRasterParameters() and customRasterParameters() function.
- update rasterMath to generate by default a 256*256 raster block size.
- update rasterMath to prevent bug if user has osgeo/gdal version is lower than 2.1.
- prevent bug when in rasterMath if processor has only 1 core.
- minor fixes