Unable to train with the selected activation function
Unable to use the selected activation function
Unable to use the selected training algorithm
Index is out of bound
The number of input neurons in the ann and data don’t match
No error
The number of output neurons in the ann and data don’t match
Scaling parameters not present
Irreconcilable differences between two struct fann_train_data structures
Trying to take subset which is not within the training set
Wrong version of configuration file
Number of connections not equal to the number expected
Fast (sigmoid like) activation function defined by David Elliott
Fast (symmetric sigmoid like) activation function defined by David Elliott
Used to define error events on struct fann and struct fann_train_data.
Error function used during training.
Error function used during training.
Standard linear error function.
Standard linear error function.
Constant array consisting of the names for the training error functions, so that the name of an error function can be received by:
Tanh error function, usually better but can require a lower learning rate.
Tanh error function, usually better but can require a lower learning rate.
The following example shows a simple program which executes a single input on the ANN.
The FANN namespace groups the C++ wrapper definitions
Cascade training differs from ordinary training in the sense that it starts with an empty neural network and then adds neurons one by one, while it trains the neural network.
The FANN library is designed to be very easy to use.
The two main datatypes used in the fann library is struct fann, which represents an artificial neural network, and struct fann_train_data, which represent training data.
Errors from the fann library are usually reported on stderr.
It is possible to save an entire ann to a file with fann_save for future loading with fann_create_from_file.
There are many different ways of training neural networks and the FANN library supports a number of different approaches.
The Fann Wrapper for C++ provides two classes: neural_net and training_data.
It is possible to run the ANN with fixed point numbers (internally represented as integers).