Build neural network
Description
This app is used to create, train and test a neural network. The input data can be made available to the neural network from the system via a list or via an external Excel file.
Functionality
Editing window
Within this window is all the functionality for loading, training and testing a new neural network to be created.
In the upper area there are the two tabs "Object list view" and "Excel file" to define the source of the training data. After the training data has been defined or read in, the input and output data of the neural network are defined using the columns of the list with the training data.
The last step is to specify the parameters of the neural network (see "Initialise NN" and "Fine-tune NN" buttons).
Now you can train the network with the input and output data using the button "Train NN". This can be executed several times in succession, whereby the result of each training cycle is output in a list.
By clicking on a training cycle, the quality of the neural network can be displayed again on the basis of all input data.
| Field | Description | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Object list view | |||||||||
| : Dialogue window | Title of the window from which the list was sent to this app as a data source for the neural network. | ||||||||
| : list | Name of the list view (if available). | ||||||||
| Excel file | |||||||||
| : Excel file | - | ||||||||
| : : file | Name of the Excel file that is to be the data source for the neural network. A file can be selected with the button to the right of the field. | ||||||||
| : data | - | ||||||||
| : : table | Number of the table sheet in the Excel file. | ||||||||
| : : row heading | Specification of the line number with the heading of the data. The designation of the data fields is generated from the heading. | ||||||||
| :: Read from line | Specifies the line number from which the data is to be read in from the Excel file. | ||||||||
| :: Read | Button for reading the data from the Excel file. | ||||||||
| :: Read training data directly from Excel | In the mode to read data from Excel (notebook tab is set to "Excel file"), you can choose whether the data should be loaded directly from Excel during training or whether the data should come from the Excel file already loaded into the list. If this identifier is set, only a few rows are loaded into the list window in order to determine the columns and their data type. This option is useful for very large Excel files (several hundred thousand rows). Otherwise, the entire Excel file is loaded into the list window and you can then also change the columns and the list entries (see also Report button in the Training data subwindow). | ||||||||
| Training data | |||||||||
| : data | If a list was sent to this app, the entries of this list are listed here as training data. If an Excel file was read in, these are listed here as training data. | ||||||||
:  | Call up the report app to change the column format of the list. | ||||||||
| : Remove | If some entries in the list are not to be included in the training data, you can mark these unwanted lines in the list and remove them from the list with this button. | ||||||||
| Training columns | |||||||||
| : data | Output of the columns of the training data. Using drag & drop, you can define entries from this list either as input or output training data. | ||||||||
| : input | Listing of the columns that you have previously inserted as input columns/data for training the neural network by dragging and dropping them from the above list with the column details. If you want to remove an entry, you can drag it back to the list of available columns or move it to the list of output columns. | ||||||||
| : output | List of the columns that you have previously inserted as output columns/data for training the neural network by dragging and dropping from the above list with the column specifications. If you want to remove an entry, you can drag it back to the list of available columns or move it to the list of input columns. | ||||||||
| : Training cycles | Output of the data of each training cycle. If you double-click on a training cycle, the Training values window opens with the input values, the original output values and the calculated output values. (This is not possible if the Read training data directly from Excel flag is set).
| ||||||||
| Apply | |||||||||
| : Input data | Dynamic display of input fields (depending on the selected input columns) for the collection of test data for the neural network. If the neural network is of the type "regression" and you have only defined one input and one output column, you can have test data generated automatically by means of start, step size and end value. The result of such a test series can then also be output as a graph. | ||||||||
| : Apply | With this button, the input data is transferred to the neural network, the result is output in the output data. | ||||||||
| : Output data | Dynamic fade-in of output fields (depending on the selected output columns) with the result of the test data from the neural network. | ||||||||
| : Save NN as XML | If the testing of the neural network has been successful, this button can be used to save the neural network as an XML file and thus make it available to other users/applications. Additional description data can be added to the trained neural network in the editing app, and the neural network can also be tested directly there. | ||||||||
| Button | Description |
|---|---|
| Action | - |
| : Back | The app is closed. |
Training values
All input and output columns of the original training data are listed in this window. In addition, the value determined by the neural network is also displayed for each original value of an output column. The additional first column indicates whether and which row was used for validation: If a row number is available, this row was used for validation.
If you double-click on a line, the corresponding line is selected in the training data.
The table columns are structured as follows: first the validation column, followed by the input columns. This is followed by the actual output and the output calculated by the neural network, alternating for each output column. In the case of numerical output values, the deviations are also specified.
| button | Description |
|---|---|
| Action | - |
| : Back | The window is closed. |
| : Draw | The specified output data and the output data determined with the neural network are output as a graph. This allows the quality of the neural network to be shown by comparing the data. The creation of the graphs can be controlled by switching columns on and off and changing their order. If columns are hidden, they are not drawn in the graph. The first column is always interpreted as the X-axis. If rows are selected, only these rows are included in the graph. |
| : Restore list | Deleted values from the list can be restored by clicking this button. |
| Menu item | Description |
|---|---|
| Invert selection | Selected rows are inverted, i.e. previously unselected rows are selected and vice versa. |
| Remove selected lines | Selected rows are removed. |
Related topics
Technical documentation
Loading module
Module name
neuralNetworkLoad.app
Classes
Security
In addition to restricting access rights via the class and its data fields, the module can be restricted in its use via some of the received messages.
| Message | Parameter | Function | Security |
|---|---|---|---|
LOAD_NEURAL_NETWORK | |||
| Message | Parameters | Function | Receive module |
|---|---|---|---|
| PLOT_CHART |
;){kind=link}
;){kind=link}
;){kind=link}