How to use the calculation power of kaggle

1. Just change to a higher configuration
lack of computing power can not be solved by software
only upgrade the configuration
or use the server instead.

2. I think that kaggle is a good place to train data mining, but it is not a big improvement for machine learning. You can see winner's solutions for the games that have ended. Most of them are ensemble learning

3. Each computer memory at least 8 g, install virtual machine, install Linux system (I use centos6.6 xx86)_ 64 Series), all minimal installation, each installation of five virtual machines, do a good job, the rest of all clones, and then with their own cluster it. Virtualization technology will be widely used in operation and maintenance in the future.

4. I think that kaggle is a good place to train data mining, but it is not a big improvement for machine learning. You can go to see winner's solutions for the games that have ended. Most of them are ensemble learning, and there are few mathematically elegant solutions

we can start from the following aspects:
1. Feature Engineering
- continue variable
- categorical variable
2. Classic machine learning algorithm
- LR, KNN, SVM, random forest, gbrt, FM, NN
3. Cross validation, model selection
- grid search, random search, Hyper opt
4. Assemble learning
recommend this kaggle assembling guide
5. Take a look at the winner solutions of previous competitions

5. Matlab is a commercial software, according to the rule of kaggle, it is not allowed to use
numpy, MATLAB in Python, and based on this, panda, sklearn are extremely convenient tools. Python package, only unexpected, nothing can not be done
Python stands out as an easy-to-use, powerful, fast and powerful tool.

6. Most column values are integer and floating-point, so we can't do too much feature engineering. For example, user_ location_ Country is not the name of a country, but an integer value. This makes it difficult to construct new features because we don't know what that number really means.

7. The kaggle competition is a very good way to learn data science and invest time. I learned a lot of concepts and ideas of data science through kaggle. I started the kaggle competition a few months after I learned programming, and I won several competitions recently
to achieve good results in the kaggle competition, we need not only to know some machine learning algorithms, but also to have an accurate mode of thinking, be easy to learn, and spend a lot of time exploring data.

8. Method 1: select the layer, press Ctrl + C to , press Ctrl + V to paste. If you want to multiple times, press Ctrl + D
method 2: drag the left mouse button away from the layer, and then right-click to . Hold down the CTRL key while dragging, so that the copied layer will remain at the same level as the original layer. If you want to multiple times, press Ctrl + D
method 3: select the layer and press the shortcut key & quot+& quot;， The is complete. If you want to multiple times, press & quot+& quot;
OK. Note that the copied layer is overlapped with the original layer. Drag the layer out after ing.