`lib-ml` 库参考

原文: https://finproducts.atlassian.net/wiki/spaces/JIDD3/pages/34793881835/FIN+5.3.1+lib-ml 中文整理基于 FIN 5.3.1

导读： 机器学习（ML）函数库

以下为 API 参考，保留英文以便对照源码；目录已列出（英文标题便于检索）；概述已尽量中译。

目录（导读）

lib
meta
docs
funcs
Overview
- System Requirements
Supervised Learning
- Kernel Functions
Training and Predicting
- Saving a Trained Model
Feature Scaling
- def
mlAddMeta(model, meta: null)
- supertypes
mlGridSearchCv(estimator, paramGrid, opts: {})
mlLinearKernel()
mlLoadModel(handle)
mlMeta(model)
mlOls(qr: false)
mlParamSweep(func, paramGrid)
mlPolyKernel(degree, gamma: null, offset: null)
mlPredict(model, x)
mlRbfKernel(gamma: null)
mlSaveModel(model, handle)
mlScaleMinMax(features, min: 0, max: Number.one)
mlScaleZ(features)
mlStats(model)
mlSvm(kernel, k, opts: {})
mlSvr(kernel, opts: {})
mlToGrid(val, opts: {})
mlToMatrix(val)
mlTrain(model, x, y)

概述（中译）

机器学习（ML）函数库

API 正文（英文）

docs

Overview • Supervised Learning • Training and Predicting • Feature Scaling

funcs


mlAddMeta	Add metadata to the given trained model
mlGridSearchCv	Configure a cross-validation grid-search
mlLinearKernel	Create a Linear kernel
mlLoadModel	Load a `trained model` from the given I/O handle
mlMeta	Get the model metadata
mlOls	Create a training model for doing ordinary least-squares regression (OLS)
mlParamSweep	Do a "parameter sweep" by invoking the given func for all combinations of parameters in the parameter grid
mlPolyKernel	Create a Polynomial kernel
mlPredict	Make predictions against a `trained model`
mlRbfKernel	Create a RBF (RBF) kernel
mlSaveModel	Save a `trained model` to the given I/O handle
mlScaleMinMax	Perform Min-Max feature scaling
mlScaleZ	Perform Z-score normalization on the features
mlStats	Get the model statistics
mlSvm	Create a Support Vector Machine (SVM) for classification
mlSvr	Create a Support Vector Machine for Regression (SVR)
mlToGrid	Convert a matrix to a Grid
mlToMatrix	Convert the given obj to an optimized matrix type for machine learning
mlTrain	Train a machine learning model with the given training data

lib ml

Overview

The mlExt provides a library of functions to perform machine learning (ML) within SkySpark. There are three broad categories of machine learning algorithms. Currently, the ext provides support for the first category (supervised learning)

Supervised Learning:
The algorithm builds a model from input data for which the desired output values are known.
Unsupervised Learning:
The algorithm builds a model from input data for which the desired output values are not known. That is, the algorithm attempts to find the structure in the data.
Reinforcement Learning:
The algorithm interacts with a dynamic environment and learns from the consequences of its actions. It is not "trained" with data.

The general set of steps for running a machine learning algorithm are:

Prepare your input data/features.
Configure a particular algorithm.
Train your model.
Make predictions against the model.

The rest of this guide provides an overview of the capabilities of the ext and what functions are available for performing these steps.

System Requirements

This ext requires Java 1.8. If you get an exception similar to the following, then you are not running Java 1.8

sys::Err: java.lang.UnsupportedClassVersionError: .... details ....

Supervised Learning

The following algorithms are available for performing supervised learning. See the function documentation for details on configuring the various algorithms for use.

Regression
- Ordinary Least Squares (OLS). See mlOls()
- Support Vector Regression (SVR). See mlSvr()
Classification
- Support Vector Machine (SVM), with support for binary and multiclass classification. See mlSvm()

Kernel Functions

Support vector machines (including SVR) require a kernel function to be configured. The following kernel functions are supported:

Linear Kernel. See mlLinearKernel()
RBF Kernel (Gaussian). See mlRbfKernel()
Polynomial Kernel. See mlPolyKernel()

Example:

svr: mlSvr(mlRbfKernel())

Training and Predicting

Once you have configured an algorithm, you can use it to train your model. See the mlTrain() function for details. The result of running that function is called a trained model. You can use a trained model to make predictions using the mlPredict() function.

svr:   mlSvr(mlLinearKernel())
model: mlTrain(svr, trainingX, trainingY)
p:     mlPredict(model, testX)

Saving a Trained Model

Training a model can be an expensive operation. You can save your trained model to disk using mlSaveModel(). You can load the model again using mlLoadModel().

Sometimes you need to associate some state or metadata with your trained model so that you can properly configure your test data for making predictions. Use the mlAddMeta() function to add metadata to a trained model. Any metadata associated with the model will also be saved/loaded using the the above functions You can retreive the metadata from your trained model using mlMeta().

// train a model using OLS and add some metadata
model: mlTrain(mlOls(), trainingX, trainingY)
model = model.mlAddMeta(["dates":dates, "site":site])

// save the model
model.mlSaveModel(`io/models/ols.model`)

// reload the model and get the site from the metadata
model = mlLoadModel(`io/models/ols.model`)
site  = model.mlMeta->site

Feature Scaling

When preparing your data for training a model, it is often necessary to perform feature scaling on the inputs/features. This process is sometimes called data normalization. Feature scaling is a method to standardize the range of values for a particular feature. Normalzing your features is often a critical step in preparing your data for modeling. For example, certain algorithms will perform very poorly if features are not scaled (e.g. SVM). Also, feature scaling will prevent a feature with a large range of values from dominating the model.

The ext has the following scaling algorithms available:

Min-Max Scaling: Scales the range of features to a given range. Common ranges are [0, 1] or [-1, 1]. See mlScaleMinMax()
Z-score Normalization: Scales the features so that they have the property that μ = 0 and σ = 1. See mlScaleZ()

Example:

trainingX = trainingX.mlScaleMinMax(0, 1)


feature	Feature namespace of definitions formatted as `feature:name`
func	Axon function


feature	Feature namespace of definitions formatted as `feature:name`
func	Axon function


feature	Feature namespace of definitions formatted as `feature:name`
func	Axon function


feature	Feature namespace of definitions formatted as `feature:name`
func	Axon function

lib-ml 库参考 ​

目录（导读） ​

概述（中译） ​

API 正文（英文） ​

docs ​

funcs ​

lib ml ​

Overview ​

System Requirements ​

Supervised Learning ​

Kernel Functions ​

Training and Predicting ​

Saving a Trained Model ​

Feature Scaling ​

func:mlAddMeta ​

mlAddMeta(model, meta: null) ​

func:mlGridSearchCv ​

mlGridSearchCv(estimator, paramGrid, opts: {}) ​

Examples: ​

func:mlLinearKernel ​

mlLinearKernel() ​

func:mlLoadModel ​

mlLoadModel(handle) ​

func:mlMeta ​

mlMeta(model) ​

func:mlOls ​

mlOls(qr: false) ​

func:mlParamSweep ​

mlParamSweep(func, paramGrid) ​

func:mlPolyKernel ​

mlPolyKernel(degree, gamma: null, offset: null) ​

func:mlPredict ​

mlPredict(model, x) ​

func:mlRbfKernel ​

mlRbfKernel(gamma: null) ​

func:mlSaveModel ​

mlSaveModel(model, handle) ​

func:mlScaleMinMax ​

mlScaleMinMax(features, min: 0, max: Number.one) ​

func:mlScaleZ ​

mlScaleZ(features) ​

func:mlStats ​

mlStats(model) ​

func:mlSvm ​

mlSvm(kernel, k, opts: {}) ​

func:mlSvr ​

mlSvr(kernel, opts: {}) ​

func:mlToGrid ​

mlToGrid(val, opts: {}) ​

func:mlToMatrix ​

mlToMatrix(val) ​

func:mlTrain ​

mlTrain(model, x, y) ​

`lib-ml` 库参考

目录（导读）

概述（中译）

API 正文（英文）

docs

funcs

lib ml

Overview

System Requirements

Supervised Learning

Kernel Functions

Training and Predicting

Saving a Trained Model

Feature Scaling

func:mlAddMeta

mlAddMeta(model, meta: null)

func:mlGridSearchCv

mlGridSearchCv(estimator, paramGrid, opts: {})

Examples:

func:mlLinearKernel

mlLinearKernel()

func:mlLoadModel

mlLoadModel(handle)

func:mlMeta

mlMeta(model)

func:mlOls

mlOls(qr: false)

func:mlParamSweep

mlParamSweep(func, paramGrid)

func:mlPolyKernel

mlPolyKernel(degree, gamma: null, offset: null)

func:mlPredict

mlPredict(model, x)

func:mlRbfKernel

mlRbfKernel(gamma: null)

func:mlSaveModel

mlSaveModel(model, handle)

func:mlScaleMinMax

mlScaleMinMax(features, min: 0, max: Number.one)

func:mlScaleZ

mlScaleZ(features)

func:mlStats

mlStats(model)

func:mlSvm

mlSvm(kernel, k, opts: {})

func:mlSvr

mlSvr(kernel, opts: {})

func:mlToGrid

mlToGrid(val, opts: {})

func:mlToMatrix

mlToMatrix(val)

func:mlTrain

mlTrain(model, x, y)