Skip to main content

Utilities for nolearn.lasagne

Project description

# nolearn-utils

[![Build Status](](

Iterators and handlers for nolearn.lasagne to allow efficient real-time image augmentation and training progress monitoring

## Real-time image augmentation

- `ShuffleBatchIteratorMixin` to shuffle training samples
- `ReadImageBatchIteratorMixin` to transform image file path into image as color or as gray, and with specified image size
- `RandomFlipBatchIteratorMixin` to randomly (uniform) flip the image horizontally or verticaly
- `AffineTransformBatchIteratorMixin` to apply affine transformation (scale, rotate, translate) to randomly selected images from the given transformation options - `BufferedBatchIteratorMixin` to perform transformation in another thread automatically and put the result in a buffer (default size = 5)
- `LCNBatchIteratorMixin` to perform local contrast normalization to images
- `MeanSubtractBatchIteratorMixin` to subtract samples from the pre-calculated mean

Example of using iterators as below:

train_iterator_mixins = [
TrainIterator = make_iterator('TrainIterator', train_iterator_mixins)

train_iterator_kwargs = {
'buffer_size': 5,
'batch_size': batch_size,
'read_image_size': (image_size, image_size),
'read_image_as_gray': False,
'read_image_prefix_path': './data/train/',
'flip_horizontal_p': 0.5,
'flip_vertical_p': 0,
'affine_p': 0.5,
'affine_scale_choices': np.linspace(0.75, 1.25, 5),
'affine_translation_choices': np.arange(-3, 4, 1),
'affine_rotation_choices': np.arange(-45, 50, 5)
train_iterator = TrainIterator(**train_iterator_kwargs)

The `BaseBatchIterator` is also modified from `nolearn.lasagne` to provide a progress bar for training process for each iteration

## Handlers

- `EarlyStopping` stops training when loss stop improving
- `StepDecay` to gradually reduce a parameter (e.g. learning rate) over time
- `SaveTrainingHistory` to save training history (e.g. training loss)
- `PlotTrainingHistory` to plot out training loss and validation accuracy
over time after each iteration with matplotlib

## Examples

Example code requires `scikit-learn`


`example/mnist/` should produce a model of about 99.5% accuracy in less than 50 epoch.

MNIST data can be downloaded from

### CIFAR10

CIFAR10 images can be downloaded from [Kaggle]( Place the downloaded data as follows:

├── data
│   ├── train
│ | ├── 1.png
│ | ├── 2.png
│ | ├── 3.png
│ | ├── ...
│   └── trainLabels.csv

`example/cifat10/` should produce a model at about 85% accuracy at 100 epoch. Images are read from disk and augmented at training time (from another thread)


- [ ] Embarrassingly parallelize transform

## License


Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release. See tutorial on generating distribution archives.

Built Distribution

nolearn_utils-0.2.dev0-py2.py3-none-any.whl (14.3 kB view hashes)

Uploaded py2 py3

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page