深度学习算法和机器学习算法
“A Breakthrough in Machine Learning would be Worth Ten Microsofts”
“机器学习方面的突破将值得十个微软”
-Bill Gates-
-比尔盖茨-
Machine learning is the main buzzword In the modern world. So everyone is interested to learn Machine Learning. In the Information Technology Field, Good knowledge about the Machine Learning is a Key Factor to have a Good Career in the Field. We need to have a deep understanding of the main concepts in Machine Learning. Machine Learning is normally used to Resolve Business Problems. But with the capability of the Machine Learning can be used for Many Fields. Sometimes people use machine learning everywhere without having a clear understanding. We need to identify the problem before applying Machine Learning. Because simply hand-coded instructions based system can work better than machine learning Algorithm. So Identifying the Problem is a key factor through the process of applying machine learning. Because Adding machine learning to the systems is a huge time and cost consuming task. After we identify the problem correctly, then we can move towards selecting the suitable Machine Learning algorithm to solve the problem. Supervised Learning, Unsupervised Learning, Transfer Learning and Reinforcement Learning are the four main types of Machine Learning Algorithms. Let’s talk about each of them with their Specifications.
机器学习是现代世界中的主要流行词。 因此，每个人都有兴趣学习机器学习。 在信息技术领域，对机器学习的良好了解是在该领域拥有良好职业的关键因素。 我们需要对机器学习的主要概念有深刻的了解。 机器学习通常用于解决业务问题。 但是凭借机器学习的能力，可以用于许多领域。 有时人们在没有清晰理解的情况下到处使用机器学习。 在应用机器学习之前，我们需要确定问题。 因为简单的基于手工编码指令的系统可以比机器学习算法更好地工作。 因此，确定问题是应用机器学习过程中的关键因素。 因为向系统中添加机器学习是一项巨大的时间和成本消耗的任务。 在正确识别问题之后，我们可以选择合适的机器学习算法来解决问题。 监督学习，无监督学习，转移学习和强化学习是机器学习算法的四种主要类型。 让我们来讨论它们各自的规范。
This is the easiest Algorithm. Developer directly gives supervise to the Algorithm. Supervised learning means you have the data and the labels. Then the Machine Learning algorithm tries to predict the correct label. If the algorithm guesses the label wrong, then algorithm corrects itself and tries it again. Due to this behaviour, this type is called Supervised Learning Algorithms. Simply meaning of this is trying to get improved from the previous faults.
这是最简单的算法。 开发人员直接监督该算法。 监督学习意味着您拥有数据和标签。 然后，机器学习算法会尝试预测正确的标签。 如果算法猜测标签错误，则算法会自行纠正并再次尝试。 由于这种行为，这种类型称为监督学习算法。 这的简单含义是试图从先前的故障中得到改进。
There are two main types in the supervised Learning called as Classification and Regression. Classification means this can predict the correct decision from the Options. Binary Classification means which have two Options. If more than two options to predict, then it called as multi-class
在监督学习中，有两种主要类型，称为分类和回归。 分类意味着可以从选项中预测正确的决策。 二进制分类意味着具有两个选项。 如果要预测两个以上的选项，则称为多类
Classification. Predicting the Heart disease of the Patient can be considered as a binary Classification. Predicting the breed of the dog from the image can be considered as the Multiclass classification.
分类。 可以将预测患者的心脏病视为二元分类。 从图像预测狗的品种可以视为多类分类。
Regression means to try to predict the Number According to the Available data. Considering the price of the House according to the number of rooms, the area of the house and other resources in the house can be given as an example of regression.
回归表示尝试根据可用数据预测数量。 根据房间数量考虑房屋价格，可以将房屋面积和房屋中的其他资源作为回归示例。
The developer does not involve direct control to the algorithm as supervised learning. Unsupervised learning algorithm has the data. But the labels are not available. For example, we can think about a scenario about selecting customer according to their buying behaviour. So then the sales teams can send emails about telling their promotions and sales to the customers. So without this, the sales team should send every customer in their database. This helps to reduce sending thousands of unwanted emails. The main thing in the unsupervised learning algorithm is that it makes cluster without labels. So as a developer, you should identify the commonalities of the cluster and put labels by yourself to the cluster. Recommendation Problems are considered as unsupervised learning problems at the start.
开发人员不会将对算法的直接控制作为监督学习。 无监督学习算法具有数据。 但是标签不可用。 例如，我们可以考虑根据客户的购买行为选择客户的方案。 因此，销售团队可以向客户发送有关其促销和销售情况的电子邮件。 因此，如果没有此功能，销售团队应该在数据库中发送每个客户。 这有助于减少发送数千封不需要的电子邮件。 无监督学习算法的主要内容是它使聚类没有标签。 因此，作为开发人员，您应该确定集群的共性，并将标签自己贴在集群上。 推荐问题从一开始就被视为无人监督的学习问题。
Transfer Learning leverages what is machine learning model learning has learned in another Machine Learning Model. This means you can use the pre-trained model to solve a similar problem as your problem. Then you can do the further improvements according to your need. This a very good approach than others.Because training the machine learning algorithm which means letting it to find all of the patterns in the data. This can be a very expensive task with a huge dataset. So instead of learning from sketch, using a pre-trained model is a good choice.
转移学习利用了另一种机器学习模型中学到的机器学习模型。 这意味着您可以使用预先训练的模型来解决与您的问题类似的问题。 然后，您可以根据需要进行进一步的改进。 这是一个比其他方法更好的方法。因为训练了机器学习算法，这意味着让它可以找到数据中的所有模式。 庞大的数据集可能是一项非常昂贵的任务。 因此，与其从素描中学习，不如使用预先训练的模型。
Reinforcement learning involves having a computer programme that performs some actions with a defined space and rewarding it for doing it well or punishing it to perform poorly. Good Example is to teach machine learning model to play chess.
强化学习包括拥有一个计算机程序，该程序可以在定义的空间内执行某些动作，并奖励表现出色的人或惩罚表现不佳的人。 一个很好的例子是教机器学习模型下棋。
So these are the four main types of machine learning algorithms. Selecting a suitable type is coming with working on more projects and experience. So having a good understanding of these machine learning types is helpful in the work.
因此，这是机器学习算法的四种主要类型。 选择合适的类型会带来更多的项目和经验。 因此，对这些机器学习类型有很好的了解将对工作有所帮助。
I like to thank you for reading my article. I hope that you will improve your knowledge by reading my article. I hope to write more articles on machine learning.
感谢您阅读我的文章。 我希望您能通过阅读我的文章来提高自己的知识水平。 我希望写更多有关机器学习的文章。
My Previous Articles on Machine Learning :
我以前关于机器学习的文章：
翻译自: https://medium.com/swlh/machine-learning-algorithm-types-selecting-suitable-algorithm-to-resolve-your-problem-87d37cf1f020
深度学习算法和机器学习算法
