tinyms 0.1.0

TinyMS is an Easy-to-Use deep learning development toolkit.

TinyMS

English | 查看中文

TinyMS is an Easy-to-Use deep learning development toolkit based on MindSpore, designed to providing quick-start guidelines for machine learning beginners.

Installation

Please checkout the install document to quickly install or upgrade TinyMS project.

Quick start

Have no idea what to do with TinyMS❓ See the Quick Start to implement the image classification application in one minutes❗

Besides, here are some use cases listed to demonstrate how TinyMS simplifies the code flow for users.

Data loading and preprocess

from tinyms.data import MnistDataset, download_dataset from tinyms.vision import mnist_transform data_path = download_dataset('mnist') mnist_ds = MnistDataset(data_path, shuffle=True) mnist_ds = mnist_transform.apply_ds(mnist_ds)

Network construction

from tinyms.model import lenet5 net = lenet5(class_num=10)

Model train/evaluation

from tinyms.model import Model model = Model(net) model.compile(loss_fn=net_loss, optimizer=net_opt, metrics=net_metrics) model.train(epoch_size, train_dataset) model.save_checkpoint('./checkpoint_lenet.ckpt') ··· model.load_checkpoint('./checkpoint_lenet.ckpt') model.eval(eval_dataset)

Model prediction

from PIL import Image import tinyms as ts from tinyms.model import Model, lenet5 from tinyms.vision import mnist_transform img = Image.open(img_path) img = mnist_transform(img) net = lenet5(class_num=10) model = Model(net) model.load_checkpoint('./checkpoint_lenet.ckpt') input = ts.expand_dims(ts.array(img), 0) res = model.predict(input).asnumpy() print("The label is:", mnist_transform.postprocess(res))

API documentation

If you are interested in learning TinyMS API, please find TinyMS Python API in API Documentation.

Community

For any developers who are not familiar with how TinyMS community works, please find the Contributing Guidelines to get started.

Release Notes

The release notes, see our RELEASE.

License

This work is licensed under Apache License 2.0.