The first chapter ---- Summary of "Hands-On Machine Learning with Scikit-Learn and TensorFlow" study notes

First, the machine learning the basics (tools sklearn)

• Machine learning definitions, some of the main concepts, the main challenge
• A complete machine-learning project contains what the title ##
  data + models are indispensable

Data:

Data preprocessing, feature selection, data visualization

Model areas:

Selection algorithm adaptation data, evaluation model (cross-validation, overfitting / underfitting learning / verification curve, a variety of precision index), the model parameter selection

Some additional concepts

Learning outside the core concept : can not cope with large amounts of data in the processing computer, the data is divided into small quantities and use online learning techniques to learn from these small quantities of data.
Batch learning techniques: the use of MapReduce technology
model parameters : These parameters determine the predictive value of the model (such as a slope)
hyper-parameters : trying to find the best values for these parameters themselves (such as study of regularization and learning rate)

The main challenge faced by machine learning (both from data and models concerned)

• Lack of data, poor data quality
• Data are not representative, non-feature information
• Model is too simple lack of training data fitting, and the model is too complex to over-fitting the training data

Solution overfitting:

• Get more data
• Simplified model (selected simpler algorithm which reduces the number of parameters or features used, regularization model)
• Reduce noise in the training data

Cross-validation does not require separate validation set of model parameters to achieve ultra-select and adjust parameters, saving valuable training data

Common unsupervised learning methods:

Clustering, dimension reduction, association rule learning, visualization

Build machine learning programs in eight steps

See the Chinese version of the book on page 611.

Second, the depth of the neural network learning (the use of tools framework tensorflow)

Neural networks and deep learning methods for natural language processing, image recognition, voice recognition

• Some common neural network architecture

CNN / RNN / LSTM / EU / DNN

• Neural network training techniques and parameter adjustment
• Derivation of neural network back-propagation
• How to fine-tune the neural network model adapted to different scenarios