Overview

Click on a nut name for more details.

nuts-ml is based on nuts-flow, which provides additional nuts, see nuts-flow overview.

All nutsflow functions can be imported from nutsml as well, e.g. from nutsflow import Collect or from nutsml import Collect both work.

Network wrapping

• KerasNetwork : wrapper for Keras networks.

• PytorchNetwork : wrapper for Pytorch networks.

• LasagneNetwork : wrapper for Lasagne networks.

Data reading

• ReadLabelDirs : read file paths from label directories.

• ReadPandas : read data via Pandas from file system.

• ReadNumpy : load numpy array from file system.

• ReadImage : load image as numpy array from file system.

Data writing

• WriteImage : write images to file system.

Data viewing

• ViewImage : display image in window.

• ViewImageAnnotation : display image and annotation in window.

Data printing (from nuts-flow)

• Print : print data to console.

• PrintType : print data type

• PrintColType : print column data, eg. tuples

• PrintProgress : print progress on iterable.

Sample processing

• ConvertLabel : convert between string labels and integer class ids.

• CheckNaN : raise exception if data contains NaNs.

• PartitionByCol : partition samples depending on column value.

• SplitRandom : randomly split iterable into partitions, e.g. training, validation, test.

• SplitLeaveOneOut : split iterable into leave-one-out train and test sets.

• Stratify : stratifies samples by down-sampling or up-sampling.

Transforming & Augmenting

• AugmentImage : augment images using random transformations, e.g. rotation.

• ImageAnnotationToMask : return bit mask for geometric image annotation.

• ImageChannelMean : compute per-channel means over images and subtract from images.

• ImageMean : compute mean over images and subtract from images.

• ImagePatchesByAnnotation : randomly sample patches from image based on geometric annotation.

• ImagePatchesByMask : randomly sample patches from image based on annotation mask.

• RandomImagePatches : extract patches at random locations from images.

• RegularImagePatches : extract patches in a regular grid from images.

• TransformImage : transform images, e.g. crop, translate, rotate.

• Mixup : mixup augmentation, see mixup: Beyond Empirical Risk Minimization

Boosting

• Boost : boost samples with high confidence for incorrect class.

Batching

• BuildBatch : build batches for GPU-based training.

Plotting

• PlotLines : plot lines for selected data columns, e.g. accuracy, loss.

Logging

• LogToFile : log sample columns to file.