属性与特征：
attribute: e.g., 'Mileage'
feature: an attribute plus its value, e.g., 'Mileage = 15000'
Note that some regression algorithm can be used for classification as well，and vice versa. For example，Logistic Regression is commonly used for classification，as it can output a value that corresponds to the probability of belonging to a given class.
some of the important supervised learning algorithms:
k-Nearest Neighbors
Linear Regression
Logistic Regression
Support Vector Machines(SVMs)
Decision Trees and Random Forests
Neural networks
unsupervised learning:
Clustering:
k-Means
Hierarchical Cluster Analysis(HCA)
Expectation Maximization
Visualization and dimensionality reduction
Principal Component Analysis(PCA)
Kernel PCA
Locally-Linear Embedding(LLE)
t-distributed Stochastic Neighbor Embedding(t-SNE)
Association rule learning
Apriori
Eclat
online vs batch learning:
whether or not the system can learn incrementally from a stream of incoming data.
batch learning:
it must be trained using all the available data. it will generally takes a lot of time and computing resources.
online learning:
feed data by either individually or mini-batches.
it it great for systems that receive data as a continuous flow and need to adapt to change rapidly or autonomously.
once it has learned about new data instances，it does not need them anymore，so you can discard them. this can save a huge amount of space.
one important parameter of online learning: learning rate，how fast is should adapt to changing data. if you set a high learning rate，then your system will rapidly adapt to new data，but it will also tend to quickly forget the old data. conversely，if you set a low learning rate，the system will learn more slowly，but it will also be less sensitive to noise in the new data or to sequences of nonrepresentative data points.
a big challenge with online learning is that if bad data is fed to the system，the system‘s performance will gradually decline. to reduce this risk，you need to monitor your system closely and promptly switch learning off(and possibly revert to a previously working state) if you detect a drop in performance. you may also want to monitor the input data and react to abnormal data(e.g., using an anomaly detection algorithm).
model-based vs instance-based learning:
the approach to generalization.
model-based:
it tunes some parameters to fit the model to the training set，and then hopefully it will be able to make good predictions on new cases as well.
measure: the cost function
intance-based:
it just learns the examples by heart and uses a similarity measure to generalize to new instances.
e.g., a similarity measure between two emails could be to count the number of words they have in common.
Main Challenge of Machine Learning: bad algorithm and bad data
1). insufficient quantity of training data
data matters more than algorithm for complex problems
however, that small- and medium-sized datasets are still very common，and it is not always easy or cheap to get extra training data，so don't abandon algorithms just yet.
2). nonrepresentatIve training data
it is crucial that your training data be representative of the new cases you want to generalize to.
3). poor-quality data
if some instances are clearly outliers，it may help to simply discard them or try to fix the errors manually.
if some instances are missing a few features，you must decide whether you want to ignore this attribute altogether，ignore theseinstances，fill in the missing values，or train one model with the features and one model without it，and so on.
4). irrelevant features
features selection：selecting the most useful features to train on among existing features.
features extraction：combining existing features to produce a more useful one.
creating new features by gathering new data.
5). overfitting the training data
overfitting happens when the model is too complex relative to the amount and noiseness of the training data.
the solutions:
(1). to simplify the model by selecting one with fewer parameters, by reducing the number of attributes in the training data or by constraining the model(regularization).
(2). to gather more training data
(3). to reduce the noise in the training data(e.g., fix data errors and remove outliers)
6). underfitting the training data
the opposite of overfitting. its predictions are bound to be inaccurate even on the training examples.
the solutions: 对应overfitting的第一种情况
(1). selecting a more powerful model，with more parameters
(2). feeding better features to the learning algorithm
(3). reducing the constraints on the model(e.g., reducing the regularization hyperparameter)
a hyperparameter is a parameter of a learning algorithm(not of the model). it is not affected by the learning algorithm itself；it must be set prior to training and remains constant during training.
testing and validating:
the error rate (of a model) on new cases is called the generalization error(or out-of-samle error).
evaluating a model is simple enough：just use a test set.
suppose you are hesitating between two models(say a linear model and a polynomial model): how can you decide? one option is to train both and compare how well they generalize using the test set.
suppose that the linear model generalize better，but you want to apply some regularization to avoid overfitting. the question is: how do you choose the value of the regularization hyperparameter? one option is to train 100 different models using 100 different values for this hyperparameter.(导致了下面这个问题)
suppose you find the best hyperparameter value that produces a model with the lowest generalization error，say just 5% error. so you launch this model into production，but unfortunately it does not performs as well as expected and produces 15% errors. what just happened? the problem is that you measured the generalization error multiple times on the test，and you adapted the model and hyperparameters to produce the best model for that set. this means that the model is unlikely to perform as well on new data.
(接上)a common solution to this problem is to have a second holdout set called the validation set. you train multiple models with various hyperparameters using the training set，and select the model and hyperparameters that perform best on the validation set，and when you're happy with your model you run a single final test against the test set to get an estimate of the generalization error.
to avoid ‘wasting’ too much training data in validation sets，a common technique is to use cross-validation：the training set is split into complementary subsets，and each model is trained against a different combination of these subsets and validated against the remaining parts. once the model type and hyperparameters have been selected，a final model is trained using these hyperparameters on the full training set，and the generalized error is measured on the test set.
No Free Lunch: assumption
a model is a simplified version of the observation. the simplifications are means to discard the superfluous details that are unlikely to generalize to new instances. to decide what data to discard and what data to keep，you must make assumptions. for example，a Linear model makes the assumption that the data is fundamentally linear and that the distance between the instances and the straight line is just noise，which can safely be ignored. if you make absolutely no assumption about the data， then there is no reason to prefer one model over any other. this is called No Free Lunch(NFL) theorem.