Deep Learning with OneFlow made easy
Carefree?
carefree-learn
aims to provide CAREFREE usages for both users and developers.
User Side
🖼️
Computer Vision # MNIST classification task with LeNet
import cflow
import numpy as np
import oneflow.data as data
(x_train, y_train), (x_test, y_test) = data.load_mnist()
x_train, x_test = np.concatenate(x_train, axis=0), np.concatenate(x_test, axis=0)
y_train = np.concatenate(y_train, axis=0)[..., None]
y_test = np.concatenate(y_test, axis=0)[..., None]
data = cflow.cv.TensorData(x_train, y_train, x_test, y_test)
m = cflow.cv.CarefreePipeline(
"clf",
dict(
in_channels=1,
num_classes=10,
img_size=28,
latent_dim=128,
encoder1d="lenet",
),
fixed_epoch=5,
loss_name="cross_entropy",
metric_names=["acc", "auc"],
tqdm_settings={"use_tqdm": True, "use_step_tqdm": True},
)
m.fit(data, cuda=0)
Developer Side
This is a WIP section :D
Installation
carefree-flow
requires Python 3.6 or higher.
Pre-Installing OneFlow
carefree-flow
requires oneflow>=0.4.0
. Please refer to OneFlow for pre-installation.
pip installation
After installing OneFlow, installation of carefree-flow
would be rather easy:
git clone https://github.com/carefree0910/carefree-flow
cd carefree-flow
pip install -e .
Citation
If you use carefree-flow
in your research, we would greatly appreciate if you cite this library using this Bibtex:
@misc{carefree-flow,
year={2021},
author={Yujian He},
title={carefree-flow, Deep Learning with OneFlow made easy},
howpublished={\url{https://https://github.com/carefree0910/carefree-flow/}},
}
License
carefree-flow
is MIT licensed, as found in the LICENSE
file.