Functions for Activation Propagation in the Network

krui_err krui_updateSingleUnit( int UnitNo )
evaluates the net input, the activation and the output value of the specified unit; returns an error code if the unit doesn't exist. krui_updateSingleUnit(...) also evaluates 'frozen' units. char *krui_getUpdateFunc( void )
returns the current update function. The default update function is _Order' (see also kr_def.h). krui_err krui_setUpdateFunc( char *update_func )
Changes the current update function; returns an error code if the update function is invalid. krui_err krui_updateNet( float *parameterArray, int NoOfParams )
updates the network according to the update function. The network should be a feed-forward type if one wants to update the network with the topological update function, otherwise the function returns a warning message. To propagate a pattern through the network the use of the following function calls is recommended:

   krui_setPatternNo( pat_no );
   krui_showPattern( OUTPUT_NOTHING );
   krui_updateNet( parameterArray, NoOfParams );

• synchronous firing: the units of the network all change their activation at the same time.

• chronological order: the user defines the order in which the units change their activations.

• random order: the activations are computed in random order. It may happen that some units are updated several times while others are not updated at all.

• random permutation: the activations of all units are computed exactly once, but in a random order.

• topological order: the units change their activations according to their topological order. This mode should be selected only with nets that are free of cycles (feed-forward nets).

The topological order propagation method computes the stable activation pattern of the net in just one cycle. It is therefore the method of choice in cycle free nets. In other modes, depending upon the number of layers in the network, several cycles are required to reach a stable activation pattern (if this is possible at all).