Machine Learning

10 must read books for Machine Learning and Data Science

ゼロから作るDeep Learning https://www.oreilly.co.jp/books/9784873117584/

https://gigazine.net/news/20170127-learning-tensorflow-3hours/

https://cloud.google.com/blog/big-data/2017/01/learn-tensorflow-and-deep-learning-without-a-phd

tensorflow predict stock

https://www.google.co.jp/search?q=tensorflow+predict+stock&oq=tensorflow+predict+stock&aqs=chrome..69i57j69i64.17108j0j4&sourceid=chrome&ie=UTF-8

RNN (Recursive Neural Network), CNN, LSTM

In feed-forward network, training data can be fit very good with a deep network (many hidden layers). However, this may cause overfitting such that the test data result is very bad.

In feed-forward network (non-RNN), overfitting can be reduced with Dropout and Batch Normalzation.

Prepare "Mac + Anaconda + TensorFlow" Environment

1. Install Anaconda (see https://sites.google.com/site/ivansetiawantky/programming-misc#TOC-Python)
1. source activate py36
   1. (py36) conda install pip
      1. Check with pip -V
   2. See http://no-title.com/programming/python3-tensorflow:
   3. conda install -c conda-forge tensorflow
      1. Check with:
         • >>> import tensorflow as tf
         • >>> hello = tf.constant("Hello, TensorFlow")
         • >>> sess = tf.Session()
         • >>> print(sess.run(hello))
         • b'Hello, TensorFlow'
         • >>> a = tf.constant(10)
         • >>> b = tf.constant(32)
         • >>> print(sess.run(a+b))
         • 42
   4. See https://codelabs.developers.google.com/codelabs/cloud-tensorflow-mnist/#1

Visualizing network

Use TensorBoard: https://www.tensorflow.org/programmers_guide/summaries_and_tensorboard

Run the service:

python -m tensorboard.main --logdir=/tmp/tblog

Hyperparameter search with TensorBoard to see get tthe optimum parameter.

Activation function.