old_Home - BD-SEARCH/MLtutorial GitHub Wiki

์–ด์„œ์™€ ๋จธ์‹ ๋Ÿฌ๋‹์€ ์ฒ˜์Œ์ด์ง€?

๋ชฉ์ฐจ

0. Python ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ ์ •๋ฆฌ

์ธ๊ณต์ง€๋Šฅ, ๋”ฅ๋Ÿฌ๋‹
  • Tensorflow
  • Keras
  • PyTorch
  • Scikit-Learn
์ˆ˜ํ•™/๊ณผํ•™
  • Scipy, Numpy
์ด๋ฏธ์ง€ ์ฒ˜๋ฆฌ
  • OpenCV
  • Pillow
DB, ๋ฐ์ดํ„ฐ ์ฒ˜๋ฆฌ
  • SQLite3
  • Elasticsearch
ํ…์ŠคํŠธ ์ฒ˜๋ฆฌ
  • nltk
  • gensim
  • KoLNPy
  • Afinn
  • VaderSentiment
๋„คํŠธ์›Œํฌ, ์ธํ„ฐ๋„ท, ์›น ๋“ฑ
  • Requests
  • BeautifulSoup
  • lxml
  • Flask
  • Scrapy
๊ธฐํƒ€
  • pip
  • Anaconda
  • Six
  • Python-Future
  • Cython

1. ๊ธฐ๊ณ„ํ•™์Šต์ด๋ž€?

๊ฐœ์š”

  • ๋จธ์‹ ๋Ÿฌ๋‹์˜ ํ•ต์‹ฌ ์š”์†Œ
  • ๋จธ์‹ ๋Ÿฌ๋‹์˜ ๋ถ„๋ฅ˜

๊ธฐ๊ณ„ ํ•™์Šต์ด ํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒƒ

ํšŒ๊ท€๋ถ„์„(Regression)
  • ๋กœ์ง€์Šคํ‹ฑ ํšŒ๊ท€ (Logistic Regression)
๋ถ„๋ฅ˜ (Classification)
  • KNN Classification
๊ตฐ์ง‘ํ™”(Clustering)
  • K-Means Clustering

3. ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ๋ณธ ๊ฐœ๋…

(์—ฌ๊ธฐ ์ˆœ์„œ ์กฐ๊ธˆ ์ˆ˜์ • ํ•„์š”ํ• ๋“ฏ)

๋”ฅ๋Ÿฌ๋‹์˜ ๋‚œ๊ด€ 3๊ฐ€์ง€
๋”ฅ๋Ÿฌ๋‹ ๊ธฐ๋ณธ ๊ฐœ๋…
  • Fully-Connected Neural Network
    • Layer์˜ ๊ตฌ์กฐ
    • ํ™œ์„ฑํ™” ํ•จ์ˆ˜
  • ๋ชจ๋ธ ์„ค๊ณ„ ์š”๋ น ๋ฐ ์œ ์˜์ 
    • Overfitting๊ณผ Underfitting ํ”ผํ•˜๊ธฐ
Awesome / ์ฐธ๊ณ  ์ž๋ฃŒ ๋งํฌ
  • AI ๊ด€๋ จ

4. ์กฐ๊ธˆ ์–ด๋ ค์šด ๋”ฅ๋Ÿฌ๋‹

๋‹ค์–‘ํ•œ ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ๋ฒ•
  • CNN
  • RNN
  • Autoencoder
  • GAN
  • U-net

- ์ธ๊ณต์ง€๋Šฅ์œผ๋กœ ์ด๋ฏธ์ง€ ๋ถ„์„ํ•˜๊ธฐ

๊ธฐ๋ณธ ๊ฐœ๋…
  • ์ตํ˜€์•ผ ํ•  ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ์ˆ 
  • ImageNet
์ „์ฒ˜๋ฆฌ
์ด๋ฏธ์ง€ ๋ถ„์„์˜ ์ฃผ์š” ๋ชจ๋ธ
  • VGG
  • ResNet
์žฌ๋ฏธ์žˆ๋Š” ๋…ผ๋ฌธ/ํ”„๋กœ์ ํŠธ ์†Œ๊ฐœ
  • Image Segmentation
  • Colorization
  • Style Transfer for Anime Sketches with Enhanced Residual U-net and Auxiliary Classifier GAN

- ์ธ๊ณต์ง€๋Šฅ์œผ๋กœ ํ…์ŠคํŠธ ๋‹ค๋ฃจ๊ธฐ

๊ธฐ์ดˆ ๊ฐœ๋…
  • Language Model
  • TF-IDF
  • Naive Bayes Classifier
  • Embedding
    • One-hot Encoding
    • Word Embedding model
    • Word2Vec
    • Doc2Vec
ํ…์ŠคํŠธ ๋ถ„์„ ๋ชจ๋ธ
  • TextCNN
  • Seq2CNN
  • SyntaxNet
์‘์šฉ
  • ๋ฌธ์žฅ ๋ถ„๋ฅ˜
  • ๊ธฐ๊ณ„ ๋ฒˆ์—ญ
ํ•œ๊ตญ์–ด ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ
  • ์ฃผ์˜์ 
  • ํ•œ๊ตญ์–ด ๋ง๋ญ‰์น˜

- Word2Vec ๋งŒ๋“ค์–ด ๋ณด๊ธฐ

  • ํ•œ๊ตญ์–ด๋กœ ๋œ ๊ธ€ ๋ชจ์œผ๊ธฐ
  • ์ „์ฒ˜๋ฆฌ
  • word2vec ๋งŒ๋“ค๊ธฐ
  • word2vec ์‚ฌ์šฉํ•˜๊ธฐ

- CNN ๋งŒ๋“ค๊ธฐ

- ํด๋Ÿฌ์Šคํ„ฐ๋ง ํŒŒํ—ค์น˜๊ธฐ

(์ด์‚ฌ ์˜ˆ์ •)

- ๊ฒŒ์ž„์—์„œ์˜ ๊ฐ•ํ™”ํ•™์Šต

Reinforcement learning
  • Reinforcement Learning์˜ ๊ธฐ๋ณธ ๊ตฌ์„ฑ ์š”์†Œ
  • Q-Learning :๋น„์–ด์žˆ๋‹ค
  • ๊ฐ•ํ™”ํ•™์Šต์„ ์ ์šฉํ•œ ๊ฒŒ์ž„
  • ์ฐธ๊ณ ์ž๋ฃŒ

๋‚ด์šฉ ์ˆœ์„œ ์ •๋ฆฌ ์ข€ ํ•ด์•ผํ•  ๊ฒƒ ๊ฐ™์€๋ฐ.. ์–ธ์ œํ•˜์ง€

source ~/tensorflow/bin/activate

๊ณต๋ถ€ : ๋ชจ๋‘์˜ ๋จธ์‹ ๋Ÿฌ๋‹

์†Œ ๋ถ€ ํ•ด ๊ณต ์˜ ์•„

๋ถ€ ํ•ด ๊ณต ์˜ ์•„ ์†Œ

ํ•ด ๊ณต ์˜ ์•„ ์†Œ ๋ถ€

๊ณต ์˜ ์•„ ์†Œ ๋ถ€ ํ•ด

์˜ ์•„ ์†Œ ๋ถ€ ํ•ด ๊ณต

์•„ ์†Œ ๋ถ€ ํ•ด ๊ณต ์˜