slick-dnn - GitLite

KO | EN

gitlite — search

#hacktoberfest #typescript #android #machine-learning #rust #docker #nodejs #cli #vue #awesome-list #pytorch #security #ios #openai

slick-dnn

(★ 120)

Tiny and elegant deep learning library

파일 탐색기

최종 버전 다운로드 (.zip)

- - autograd.rst
  - conf.py
  - index.rst
  - optimizers.rst
- Makefile
- requirements.txt
- mnist_cnn.py
- mnist_linear.py
- - __init__.py
  - activations.py
  - losses.py
  - mathematical.py
  - tensor_modifications.py
- - __init__.py
  - example_datasets.py
- __init__.py
- module.py
- optimizers.py
- variable.py
.gitignore
LICENSE
Pipfile
Pipfile.lock
README.md
setup.py

# 설치 가이드

1. 코드 내려받기

git clone https://github.com/kaszperro/slick-dnn

깃허브에서 프로젝트 코드 전체를 내 컴퓨터로 내려받습니다.

cd slick-dnn

방금 내려받은 프로젝트 폴더 안으로 이동합니다.

2. Python

쉬움 추천

사전 준비물

Git GitHub에서 프로젝트 코드를 내려받으려면 필요합니다.
Python 3 설치 시 'Add Python to PATH' 옵션을 꼭 체크하세요 (Windows).

pipenv install

이 명령어를 터미널에 그대로 입력해 실행하세요.

python <실행할 파일명>.py # README에서 정확한 실행 파일명을 확인하세요

파이썬 스크립트(또는 모듈)를 실행합니다.

✅ 에러 메시지 없이 실행되고 터미널에 안내 문구가 출력되면 정상입니다.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Slick-dnn

Deep learning library written in python just for fun.

It uses numpy for computations. API is similar to PyTorch's one.

### Docs:

https://slick-dnn.readthedocs.io/en/latest/

### Includes:

1. Activation functions:
    * ArcTan
    * ReLU
    * Sigmoid
    * Softmax
    * Softplus
    * Softsign
    * Tanh
    
2. Losses:
    * MSE
    * Cross Entropy

3. Optimizers:
    * SGD
    * Adam

4. Layers:
    * Linear
    * Conv2d
    * Sequential
    
5. Autograd operations:
    * Reshape
    * Flatten
    * SwapAxes
    * Img2Col
    * MaxPool2d
    * AvgPool2d
    * MatMul
    * Mul
    * Sub
    * Add

### Examples:

* In examples directory there is a MNIST linear classifier, which scores over 96% accuracy on test set.

* In examples directory there is also MNIST CNN classifier, which scored 99.19% accuracy on test set. One epoch of training takes about 290 seconds. It took 7 epochs to reach 99.19% accuracy (~30 min). Time measured on i5-4670k

* Sequential model creation:
```python
from slick_dnn.module import Linear, Sequential
from slick_dnn.autograd.activations import Softmax, ReLU
my_model = Sequential(
    Linear(28 * 28, 300),
    ReLU(),
    Linear(300, 300),
    ReLU(),
    Linear(300, 10),
    Softmax()
    )
```
* Losses:
```python
from slick_dnn.module import Linear
from slick_dnn.autograd.losses import CrossEntropyLoss, MSELoss
from slick_dnn.variable import Variable
import numpy as np

my_model = Linear(10, 10)

loss1 = CrossEntropyLoss()
loss2 = MSELoss()

good_output = Variable(np.zeros((10,10)))
model_input = Variable(np.ones((10,10)))
model_output = my_model(model_input)

error = loss1(good_output, model_output)

# now you can propagate error backwards:
error.backward()
```

* Optimizers:

```python
from slick_dnn.module import Linear
from slick_dnn.autograd.losses import CrossEntropyLoss, MSELoss
from slick_dnn.variable import Variable
from slick_dnn.autograd.optimizers import SGD
import numpy as np

my_model = Linear(10, 10)

loss1 = CrossEntropyLoss()
loss2 = MSELoss()

optimizer1 = SGD(my_model.get_variables_list())

good_output = Variable(np.zeros((10,10)))
model_input = Variable(np.ones((10,10)))
model_output = my_model(model_input)

error = loss1(good_output, model_output)

# now you can propagate error backwards:
error.backward()

# and then optimizer can update variables:
optimizer1.zero_grad()
optimizer1.step()

```