It is time to set our goals and introduce the framework for creating a machine learning algorithm; it means building a model that outputs correct information.

现在是设定目标并介绍用于创建机器学习算法的框架的时候了。 这意味着建立一个输出正确信息的模型。

Generally, a machine learning algorithm can be thought of as a black box. It takes inputs and gives outputs.

通常，机器学习算法可以被视为黑匣子。 它接受输入并给出输出。

For instance, we may want to create a model that predicts the weather tomorrow given meteorological information for the past few days.

例如，我们可能想要创建一个模型，该模型根据过去几天的气象信息来预测明天的天气。

The input will feed to the model could be metrics such as temperature humidity and precipitation.

输入到模型的输入可能是诸如温度湿度和降水量的度量。

The output we will obtain would be the weather forecast for tomorrow.

我们将获得的输出将是明天的天气预报。

Before we get comfortable and confident about the model’s output we must train the model.

在对模型的输出感到满意和自信之前，我们必须训练模型。

Training is a fundamental concept in machine learning as this is the process through which the model learns; how to make sense of the input data? Once we have trained our model we can simply feed it with data and obtain an output.

训练是机器学习的基本概念，因为这是模型学习的过程。 如何理解输入数据？ 一旦我们训练了模型，我们就可以简单地将其输入数据并获得输出。

训练和算法背后的逻辑涉及四个要素。 (The logic behind training and algorithm involves four ingredients.)

1. Data数据
2. Model模型
3. Objective function目标功能
4. An optimization algorithm.优化算法。

Let’s explore each of them.

让我们探索每个。

First, we must prepare a certain amount of data to train with. Normally, this is historical data which is easily available. Second, we need a model. The simplest model we can train is a linear model in the weather forecast example.

首先，我们必须准备一定数量的数据进行训练。 通常，这是历史数据，很容易获得。 其次，我们需要一个模型。 在天气预报示例中，我们可以训练的最简单模型是线性模型。

That would mean to find some coefficients multiply each variable with them and sum everything to get the output.

这意味着要找到一些系数，将每个变量与它们相乘，并对所有变量求和以得到输出。

Moving on the linear model deep, machine learning allows us to create complicated non-linear models. They usually fit the data much better than a simple linear relationship. The third ingredient is the objective function. We have data, we fed it to the model and obtained an output, we want this output to be as close to reality as possible. Objective function checks, how correct the model’s outputs are?.

深入学习线性模型，机器学习使我们能够创建复杂的非线性模型 。 它们通常比简单的线性关系更好地拟合数据。 第三个要素是目标函数。 我们有数据，将其提供给模型并获得了输出，我们希望该输出尽可能接近实际情况。 目标函数检查，模型输出的正确性如何？

On average the entire machine learning framework boils down to optimizing this function. For example, if our function is measuring the prediction error of the model we would want to minimize the error or minimize the objective function.

平均而言，整个机器学习框架归结为优化此功能。 例如，如果我们的函数正在测量模型的预测误差，则我们希望将误差最小化或将目标函数最小化。

The last ingredient is the optimization algorithm. It consists of the mechanics through which we can vary the parameters of the model to optimize the objective function. For example, if the weather forecast model is, the weather tomorrow equals w1 times temperature plus W2 times humidity the optimization algorithm may go through values like 1.05 * temperature + 1.2 * humidity or 1.05 * temperature — 1.2 * humidity or 1.04 * temperature — 1.1 * humidity and so on.

最后一个因素是优化算法。 它由机制组成，通过这些机制我们可以更改模型的参数以优化目标函数。 例如，如果天气预报模型是，明天的天气等于w1乘以温度加W2乘以湿度，则优化算法可能会经过1.05 *温度+ 1.2 *湿度或1.05 *温度-1.2 *湿度或1.04 *温度-1.1的值*湿度等。

W 1 and w 2 are the parameters that will change for each set of parameters.

W 1和w 2是随每组参数而变化的参数。

We would calculate the objective function then choose the model with the highest predictive power. How do we know which one is the best? It would be the one with an optimal objective function

我们将计算目标函数，然后选择具有最高预测能力的模型。 我们怎么知道哪个最好？ 这将是具有最佳目标函数的那个

The machine learning process is iterative. We feed data into the model and compare the accuracy through the objective function. Then we vary the model’s parameters and repeat the operation.

机器学习过程是迭代的。 我们将数据输入模型，并通过目标函数比较准确性。 然后，我们更改模型的参数并重复操作。

模型 (MODEL)

线性模型 (Linear Model)

The linear model display appearing oversimplified, it is extremely important as it is the basis for more complicated models including non-linear ones. In the linear model world f(x) = x *w + b

线性模型的显示显得过于简单，这是非常重要的，因为它是包括非线性模型在内的更复杂模型的基础。 在线性模型世界中f(x)= x * w + b

x is the input

x是输入

w the coefficient of x

w x的系数

B would be the intercept

B将是截距

In machine learning W is called Weight or weights when we have more than one parameter b is called the bias or biases.

在机器学习中，当我们有多个参数b时，W被称为权重或权重。

There are many ways to define the linear model

有很多定义线性模型的方法

The goal of the machine learning algorithm would be to find such values for W and B. So the output of X*w + B is as close to the observed values as possible.

机器学习算法的目标是找到W和B的此类值。因此X * w + B的输出尽可能接近观测值。

For example, our goal is to predict the price of an apartment based on its size.

例如，我们的目标是根据公寓的大小预测公寓的价格。

So the input X is the size

所以输入X是大小

X * w + B is the model

X * w + B是模型

we are using the calculation of this expression gives us the price or the output y.

我们正在使用此表达式的计算得出价格或输出y。

The size of an apartment is 743 square feet. A possible model for predicting its price is X * 336.1–3237.51. If we calculate the output following this model for 743 square feet apartment we would obtain a price of $246,484.79.

公寓的面积为743平方英尺。 预测其价格的可能模型是X * 336.1–3237.51。 如果我们按照此模型为743平方英尺的公寓计算输出，我们将获得246,484.79美元的价格。

Let’s extend our real estate valuation model to multiple inputs if we had additional information say the apartments we are located by the sea. The closer an apartment is to the beach the higher the price should be.

如果我们还有其他信息说我们位于海边的公寓，那么让我们将房地产评估模型扩展到多个输入 。 公寓距离海滩越近，价格应该越高。

The linear model would predict a price based on both size and proximity to the beach. If we want to state this expression in linear algebraic terms we can use the same framework as before x * w + B. X and W are both vectors.

线性模型将根据海滩的大小和邻近程度来预测价格。 如果我们想用线性代数来表示该表达式，则可以使用与x * w + B之前相同的框架。X和W都是向量。

Let’s continue our housing example with the two input variables the output y is equal to x times w plus B. X has two elements a size of 743 and a distance from the beach of 1.2 1 miles. W has two elements to one for each input.

让我们继续我们的住房示例，使用两个输入变量，输出y等于x乘以W加上B。X有两个元素，大小为743，距海滩的距离为1.2 1英里。 对于每个输入，W有两个元素。

The greater the distance from the beach the lower the price.

离海滩的距离越远，价格越低。

The new situation we may be interested in predicting not only the price of the apartment when buying it but also the price for which we can rent it out.

在这种新情况下，我们可能不仅希望预测购买公寓时的价格，而且还要预测我们可以出租的价格。

Our inputs are unchanged. Therefore we can create two linear models.

我们的输入没有改变。 因此，我们可以创建两个线性模型。

The price as a function of the size and proximity to the beach and the rent as a function of the size and proximity to the beach.

价格与海滩的大小和距离有关，而租金与海滩的大小和距离有关。

Y1 = x1 * w11 + x2 * w21 + B1

Y1 = x1 * w11 + x2 * w21 + B1

Y2 =x 1*w12 + x2 * w22 + B2

Y2 = x 1 * w12 + x2 * w22 + B2

Notice the indices of the weights the first number refers to the respective input while the second to the output.

注意权重的索引，第一个数字表示相应的输入，第二个数字表示输出。

We have two outputs two inputs for weights and two bias’s the number of weights depends on the inputs and outputs.

我们有两个输出，两个输入用于权重，两个偏置用于权重的数量取决于输入和输出。

There is a different weight for each input in each equation.

每个方程式中每个输入的权重都不同。

In general, if we have k inputs and outputs the number of weights would be k times M. The number of biases is equal to the number of outputs.

通常，如果我们有k个输入和输出，则权重的数量将是k乘以M。偏差的数量等于输出的数量。

Finally, an example of many inputs outputs and observations. The output matrix will be entered by N where N is the number of observations an M is the number of output variables.

最后，是许多输入输出和观察结果的示例。 输出矩阵将由N输入，其中N是观察数，M是输出变量数。

The input matrix will be n by k where k is the number of input variables.

输入矩阵将为n×k，其中k为输入变量的数量。

The weights matrix remains the same as the weights don’t change depending on the number of observations. The same applies to the biases.

权重矩阵保持不变，权重不随观察次数而变化。 偏见也是如此。

Let's discuss two situations in which machine learning and neural networks come in handy.

让我们讨论一下机器学习和神经网络派上用场的两种情况。

How a linear model can solve our cats and dogs photo problem?

线性模型如何解决我们的猫和狗的拍照问题？

Here is a scatterplot of a dataset consisting of animal photos.

这是由动物照片组成的数据集的散点图。

The blue dots be the dogs is photos from one of our previous examples and the orange ones the cats a linear model or a straight line can solve this problem fairly easily.

蓝点是狗，是我们先前示例之一的照片，橙点是猫的线性模型或直线可以很容易地解决此问题。

Everything below the line is one category and above it is the other in this situation we have a classification problem.

在这种情况下，该行下方的所有内容都是一个类别，而在其上方的其他类别则是分类问题。

We are trying to classify the photos into dogs and cats.

我们正在尝试将照片分类为猫和猫。

Such a model is called a linear classifier. The data represented in the graph is linearly separable.

这种模型称为线性分类器。 图中表示的数据是线性可分离的。

But what about this case we have only several categories but we can’t fit a straight line through them. This data is not linearly separable. Therefore we must use a non-linear model.

但是，在这种情况下，我们只有几个类别，但是我们无法通过它们来拟合一条直线。 该数据不是线性可分离的。 因此，我们必须使用非线性模型。

We will learn how to handle such problems in the section where we will learn about deep neural networks.

我们将在学习深度神经网络的部分中学习如何处理此类问题。

Here is the graph of our apartment price example. Each point represents an apartment a linear model explains the data. This is one of the well known linear relationships

这是我们的公寓价格示例的图表。 每个点代表一个公寓，线性模型解释了数据。 这是众所周知的线性关系之一

however different regression problems may not necessarily be solved by linear model.

但是，线性模型不一定可以解决不同的回归问题。

Totally not linear.

完全不是线性的。

In the next blog, we will cover the objective function and optimization algorithm. Stay tuned!

在下一个博客中，我们将介绍目标函数和优化算法。 敬请关注！