"Bi thing" Microsoft neural network algorithm

In the original: "Bi thing" Microsoft neural network algorithm

The Microsoft Neural Network is by far the most powerful and complex algorithm. To find out how complex it is, look at the SQL Server Books Online description of the algorithm: "This algorithm establishes a classification and regression mining model by establishing a multi-layered perceptual neuron network." Similar to the Microsoft Decision tree algorithm, when each state of a predictable attribute is given, the Microsoft Neural Network algorithm calculates the probability of each possible state of the input attribute. These probabilities can then be used to predict the output of the predicted attribute based on the input attribute. ”
When do you use this algorithm? It is recommended to use other algorithms when they are not able to produce meaningful results, such as the output of the lift chart. We often use the Microsoft Neural network as the last resort of the "bottom of the box", which is used when other algorithms do not get meaningful results when dealing with large and complex datasets. This algorithm can accept discrete or continuous data types as input. Before using a Microsoft neural network on a large data source, be sure to test it with a production-level load because the expense of dealing with such models is too high. As with other algorithms, there are several parameters that can be configured in the Algorithm Parameters dialog box. As with some other expensive algorithms, it is only necessary to modify the default values if the business case is very sufficient.
A variant of the Microsoft Neural Network algorithm is the Microsoft Logistic regression algorithm.

Let's move on to the topic, and we'll continue to take advantage of the last solution, followed by the following steps:
Data Source view:

Key: Sequence
Input: command, force, intelligence, politics, glamour
Predictable: Identity

Data content Type:
Continuous (continuous type): command, Force, intelligence, politics, glamour
Discrete (discrete type): identity

Modeling is completed, resulting in a data mining structure interface containing Mining Structure (mining structure), Mining Models (mining model), Mining Model Viewer (Mining Model Viewer), Mining accuracy Chart (mining accuracy diagram) and Mining model prediction (Mining Model Prediction), wherein in mining Structure (mining structure), the main is to render the correlation between data and the analysis of variables.

Mining Model:
In the mining Models (mining model), the main is to list the established mining model, you can also add new mining model, and adjust variables, variable usage includes ignore (ignore), input (variable), Predict (predictor variable, input variable) and Predict Only (predictor variable),.

Right-click on the mining model and select "Set algorithmic parameters ..." to modify the model parameter settings.

These include:
Hidden_node_ratio: Specifies the number of nodes in the hidden layer to be judged. The number of node points in the hidden layer is calculated as: Hidden_node_ratio *sqrt ({# of INPUT nodes} * {Number of OUTPUT nodes}).
Holdout_percentage: Specifies the percentage that is used to calculate the forecast errors for the test group as part of the stop criteria.
Holdout_seed: Specifies the seed data used to randomly generate the test group. If not specified, the algorithm generates random seeds based on the model name to ensure that the test group remains the same when the model is re-processed.
Maximum_input_attributes: Specifies the maximum number of input variables that the algorithm can handle. Setting this value to 0 disables the input variable.
Maximum_output_attributes: Specifies the maximum number of output variables that the algorithm can handle. Setting this value to 0 disables the output variable.
Maximum_states: Specifies the maximum number of variable value states supported by the algorithm. If the number of attribute states is greater than this value, the algorithm intercepts the most commonly used state and treats the remaining states that exceed the maximum value as omissions.
Sample_size: Specifies the number of cases used to train the model. The algorithm takes less than two: Sample_size or total_cases * (1-holdout_percentage/100).

Mining Model Viewer:
The Mining Model Viewer shows the results of the mining model by representing the direction and size of the value state of a variable to the Predictor variable through a histogram.






Lift Chart:

Classification matrix:

Reference documents:
Microsoft Neural Network algorithm
http://msdn.microsoft.com/zh-cn/library/ms174941 (v=sql.105). aspx

"Bi thing" Microsoft neural network algorithm