Neuromat's Model Manager Tutorial

Creating a new model

The first screen you should see when you start the Model Manager is as follows:

Installing the Model Manager
Creating a new model
Training your model
Editing your models
Checking your models
Making predictions with your models

In this page, you can create a new model.

Advice and requirements for the name of the model

The interest of choosing a meaningful name is obvious. If you choose to call your models "model1", "model2", etc, it is quite likely that you will soon have forgotten what they are. Suppose that you want to build a model for the Yield Strength of a family of steel, and your database includes a large number of possible variables. You might want to start using only the composition as input variables, but later try to include the grain size as an input variable. You could then use the two names "YS_austenitics_basic" and "YS_austenitic_with_grain_size"

Requirement: The name of the model should not contain any space or tabulation. To avoid problems, use characters, numbers and underscore only.

Setting the path for your model

There are three categories of path.

Absolute: you give the path from the root directory (for example: /mphil/users/bob/neuralnet/model1).
Relative: you give the path without the initial "/". It is then appended to the current directory. It is not recommended to enter a relative path unless you are sure what you are doing.
Absolute, abbreviated: as in the example, the "~" can be used to replace the path to your home directory (e.g. "/mphil/users/bob/neuralnet/model1" can be written "~/neuralnet/model1")

Once you have entered the path, the model manager will check that you are not using an already existing model and create the necessary directories for your model.

Advices: you will seldom have to access the directory yourself, and the model manager will always be able to remind you which directory corresponds to which model. It is therefore not essential to have meaningful names.

Back to main

Top