Artificial intelligence technology is getting more and more popular. It is also very popular with deep learning technology and machine learning technology. However, due to lack of experience or insufficient technology, the development of new technology is often chaotic in the process of learning. Unclear direction, today we will talk about the machine to learn the novice engineers! At the beginning of the entry, the mean square error as a loss function is a good default choice. But when it comes to dealing with real-world problems, this undesigned loss function rarely gives an optimal solution. Take fraud detection as an example. In order to be consistent with your true business goals, you need to punish the false negatives in proportion to the amount of dollar damage lost by fraud. Using the mean square error can give you a good result, but it won't be the best result. Key points: Customize the loss function each time to match your target. Once many people have completed the introductory tutorial, they start using the same algorithm in all cases. This is very common, they feel that the effect of this algorithm is the same as other algorithms. This assumption is very bad and will eventually lead to poor results. The solution is to let the data choose the model for you. Once you've preprocessed the data, feed it into a number of different models to see what the results are. You will know which models are most suitable and which ones are not. Important: If you have been using the same algorithm all the time, this may mean that your results are not the best. Outliers are sometimes important and sometimes negligible, depending on the situation. Take the income forecast as an example. Sometimes the income will suddenly change a lot. It is helpful to observe this phenomenon and understand the cause. Sometimes outliers are caused by some kind of error, then you can safely ignore them and remove them from your data. From a model perspective, some models are more sensitive to outliers. Taking Adaboost as an example, it treats outliers as an important example and gives the exception values ​​a great weight, while the decision tree might simply treat the outliers as a false classifier (false classificaTIon). Key points: Before each start of work, carefully observe the data to determine if the outliers should be ignored. If you can't decide, look more closely. 24 hours a day, seven days a week, 12 months a year, and the wind direction are periodic characteristics. Many machine learning novice engineers don't know how to convert these features into representations that can hold information, such as 23 o'clock and 0. In the case of hours, the best way to do this is to calculate its sin and cos so that you can represent the period feature as a (x,y) coordinate of a circle. In the time represented in this way, 23 o'clock and 0 o'clock are the two numbers immediately next to each other, and nothing more. Important: If you encounter periodic features in your research but don't convert them into representations, then you are adding garbage data to the model. The L1 and L2 regularization penalizes larger coefficients and is a common way to regularize linear or logisTIc regression. However, many machine learning engineers are unaware of the importance of standardizing features before using regularization. Suppose you have a linear regression model, one of which is the "transaction amount." If the unit of the transaction amount is US dollars, then its coefficient should be 100 times the coefficient of the unit of cents. This can cause bias, causing the model to punish the actual smaller features. In order to avoid this problem, these features need to be standardized so that regularization is equal for all features. Key points: Regularization works well, but if you don't standardize features, it can be a headache. Many off-the-shelf linear regressions return p-values ​​for each coefficient, and some machine learning novice engineers believe that for linear models, the larger the value of the coefficient, the more important the feature is. This is not accurate because the size of the variable changes the absolute value of the coefficient. If the features are collinear, the coefficients can be converted from one feature to another. The more features of the data set, the greater the likelihood that the feature is collinear, and the less reliable the simple interpretation of the importance of the feature. Key points: It is important to understand which features have the greatest impact on the results, but they cannot be determined by the coefficients alone. Doing some projects and getting good results feels like winning a million! You work hard and the results prove that you are doing well, but like any industry, the devil is always hidden in the details, and the precise chart may hide deviations and errors. The errors listed in this article are not all, just to motivate the reader to think about the subtle issues that may be hidden in your solution. To get good results, it's important to follow the process and double check to make sure you don't make common mistakes. Oil coolers are also called oil coolers. According to the principle of the refrigeration system, the liquid refrigerant with low temperature and low pressure exchanges heat with the surrounding water in the evaporator. The evaporator absorbs the heat of the oil and evaporates into a gaseous state with low temperature and low pressure. , The low-temperature and low-pressure gaseous refrigerant enters the compressor, is compressed by the compressor, is compressed into a high-temperature and high-pressure gaseous state, and then enters the condenser, where it exchanges heat with the indoor medium, and part of the heat in the high-temperature and high-pressure gaseous state is removed. The medium absorbs, the temperature of the medium rises, the refrigerant releases heat and turns into a high temperature and high pressure liquid state, and the condenser process temperature remains unchanged, and then enters the expansion valve for throttling. Throttling is a process of rapid cooling, and the refrigerant becomes a low temperature and low pressure liquid state. The latter refrigerant enters the evaporator for heat exchange and evaporation, thereby realizing the entire process of the refrigeration system. This cycle is carried out continuously, so that the oil can be continuously refrigerated. Marine oil cooling unit,oil cooling unit,Oil coolers,Marine oil cooling unit price Taizhou Jiabo Instrument Technology Co., Ltd. , https://www.taizhoujbcbyq.com