Main Page Namespace List Class Hierarchy Compound List File List Namespace Members Compound Members File Members

LissomMap Class Reference

Inheritance diagram for LissomMap:

[legend]Collaboration diagram for LissomMap:

[legend]List of all members.

Detailed Description

Deprecated class offering old-style interface to routines in kernel.c.

New classes should interface to this class as a generic InternalNeuralRegion instead.

Definition at line 289 of file kernel.c.

Public Methods

LissomMap (const string &name_i, Dimensions dims, bool init_weights_, ParamMap *params_)

Constructor.

virtual void activate (bool learn=false, bool settle=true, bool activatefn=true)

Activate the map given the current inputs.

virtual void backproject ()

yschoe: backprojection from map to input: this code was ripped from response_to_input, and made unoptimized.

virtual void backproject (const string &name, const double gammaaff=1.0)

Backproject local core routine: replaces call to parent class' slower backproject().

virtual const WeightMatrix get_weights (const string &name, int ui=0, int uj=0) const

Provides standard interface for accessing weights.

virtual NeuralRegion * get_input (const string &name)

Returns the specified input region for a unit.

virtual void prune ()

Dead connections have a small negative value on them.

virtual void save_state (const string &)

Appends snapshot to filebase.{t}.

virtual bool restore_state (const int, const string &)

Loads ASCII or binary-format snapshot from filebase.{t} from which to continue.

virtual double size_connection_bytes () const

Memory requirements per PE.

virtual double size_unique_connections () const

Total number of connections.

virtual void add_input (const string &name_, NeuralRegion &upstream_region, WeightFunction &fn, Length size_scale)

Declare that the given region gives input to this region, and define any connections needed.

virtual Dimensions input_dimensions (WeightFunction &fn, Length size_scale=1.0)

Computes input dimensions that would be required to avoid border effects.

virtual WeightBounds * get_weights_bounds (const string &name, int ui=0, int uj=0) const

Allocates and returns a BoundingBox on the map for the weights of the given unit; all active weights must lie within.

virtual int num_aff_inputs () const

Protected Types

typedef std::vector< actlist > ActivityListTable

typedef MatrixType< int
>::rectangular MarkerType

typedef std::vector< double > SumType

typedef NeuralRegion::ActivityMatrix aff_input_type

typedef std::vector< const
aff_input_type * > aff_inputs_type

typedef Boundary::BoundingCircle<
int, int > BoundingCircle

typedef Boundary::AARBoundingBox<
int, int > BoundingRectangle

typedef std::vector< Input * > inputs_type

Type to use for storing inputs.

Protected Methods

void present_inputs (bool learn, bool settle=true, bool activate=true)

Main function called at each iteration to present inputs and modify the weights accordingly.

void init_weights (bool init_weights)

Initialize active weights to specified values.

void collect_activation_data (int dest_pe=Uninitialized)

Some data is ordinarily left distributed on each PE because it is not needed on every PE for the core LISSOM algorithm.

double dist_lat_wt (double value, double amt_of_randomness)

This function returns a value randomly distributed within a range value*amt_of_randomness on either side of value.

double lat_resp (int ui, int uj, int radius, int array_width, l_weight *weights, int weight_idx_bound)

Computes lateral excitation or inhibition response with the given weight vectors and radius.

double lat_resp_rad1 (int ui, int uj, int radius, int array_width, int array_width_2, l_weight *weights, int weight_idx_bound)

Optimized case when radius is just 0 or 1, mainly for excitation resp of organized map.

double lat_resp_named (int ui, int uj, const string &name)

Computes lateral excitation or inhibition response with the given named weights.

double sigmoid (double activity, double sigdelta, double sigbeta)

Transform the activity to a sigmoid response between 0 and 1.0.

double truesigmoid (double activity, double beta)

Transform the activity to a true sigmoid response between -1.0 and 1.0.

double truesigmoidprime (double sigmoidactivity, double beta)

Calculate the derivative of true sigmoid response f'(a) given f(a).

double positivesigmoid (double activity, double beta)

Positive sigmoid shifted to the right f(a) = 1/(1+exp(-(x-0.5)*beta)).

double logsigmoid (double activity, double beta)

Transform the activity to a log response.

double logsigmoidprime (double activity, double beta)

Calculate the derivative of a log response.

int binary_search (const actlist &list, int lower, int upper, int value)

Binary search for value in ordered list of elements constrained to the index range [lower,upper].

double change_lateral_exc_radius (double old_radius, double new_radius)

Change lateral excitatory radius for every row.

void clear_weights (void)

Intialize all weights to known values.

void randomize_lat_wts (void)

void compute_responses (int ts, bool activate, bool continuousdynamics)

Add up input and lateral activations.

void crop_actlist (const actlist &, int lowval, int highval, int *lowidx, int *highidx)

Given an ordered list and a range [lowval,highval], returns the range of list indexes [lowidx,highidx] which contain values in that range.

void initialize_actlists (void)

Initialize the activity list counts to zero.

void initialize_markers (void)

Initialize activity markers for optimization of lat_exc computations.

void modify_all_weights (void)

Updates each set of weights, if the associated learning rate is nonzero.

void modify_lat_wts (double alpha_exc, double alpha_inh)

Modify lateral weights by normalized Hebbian rule.

void modify_latwt_loop (int i, int j, int radius, int array_width, l_weight *weights, int weight_idx_bound, double alpha)

Assumes activity has been collected into the prev_map_activity array.

void modify_latwt_loop_killed (int i, int j, int radius, int array_width, l_weight *weights, int weight_idx_bound, double alpha)

void modify_latwt_loop_rad1 (int ui, int uj, int radius, int array_width, int array_width_2, l_weight *weights, int weight_idx_bound, double alpha)

Optimized case when radius is just 1.

void modify_weights (double alpha, bool normalize)

Change afferent weights by normalized Hebbian rule (sum of weights normalization).

void normalize_afferent_wts (int lowk, int highk, int lowl, int highl, int row, int j, double length)

void normalize_lat_wts (int i, int j, int radius, int array_width, l_weight *weights)

Renormalize a set of lateral weights of the given neuron, such as after changing the radius.

double prune_circular_aff_weights (int ui, int uj, int radius, AfferentWeightsList &weights, bool smooth, double radius_trim, double smooth_trim, double length)

Kill off connections that aren't within the specified radius.

void prune_circular_lat_weights (int ui, int uj, int radius_sq, int array_width, l_weight *weights, bool smooth, double radius_trim, double smooth_trim)

Kill off connections that aren't within the specified radius.

void receptor_normalize (void)

Output normalization for receptors.

void reduce_lateral_radius (int ui, int uj, double old_radius, double new_radius, int array_width, l_weight *weights, bool circular, bool smooth, double smooth_trim)

Redistribute and renormalize lateral excitation.

void response_to_input (bool continuousdynamics)

This code assumes that DEATH_FLAG is either (a) never used, or (b) small relative to the sum of the weights (usually 1.0).

void set_markers (int radius)

Marks whether neurons are active or not.

void settle_responses (bool learn, int tsettle, bool activate, bool continuousdynamics)

Propagate input and let the network settle.

void setup_afferent_weights (Neuron &neuron, AfferentWeightsList &awts, int ui, int uj, int radius, double radius_sq, int preset, double sigma, double amt_of_randomness, bool circular, bool smooth, double radius_trim, double smooth_trim, double force_cx=-1, double force_cy=-1)

void setup_lateral_weights (int ui, int uj, int radius, int array_width, l_weight *weights, int preset, double sigma, double amt_of_randomness, bool circular, bool smooth, double radius_trim, double smooth_trim, int *con_killed_flag)

Sets up lateral weights for one neuron based on where it is in the map.

void verify_saved_lateral_weights (int i, int j, int radius, int array_width, l_weight *old_weights, l_weight *new_weights, double tolerance, const char *description)

void verify_saved_weights (double tolerance)

cmdstat exc_rad_setfn2 (const string &name, Typeless &param, const Typeless &newvalue)

Automatically called by cmdparam.c to set this parameter.

int array_radius_int (const ParamMap &parammap, const string &radius_param) const

Given a radius parameter, returns the actual integer radius to be used, taking smoothing into account.

int array_radius_int (const double radius_float) const

Subscript input_row (const Input &input, Subscript row) const

Subscript input_col (const Input &input, Subscript col) const

Protected Attributes

OwningPointer< ParamMap > params

const int num_aff

int exc_connections_killed

Flag for use in modify_weights.

int inh_connections_killed

Flag for use in modify_weights.

const int exc_array_width_2

Holds 2*exc_array_width (optimization).

const int RN_sq

RN*RN precomputed for index calculations.

ActivityMatrix gather_activity

To collect activity in each row.

ActivityMatrix init_activity

Initial map_activity over network.

ActivityMatrix map_activity

Current map_activity over network.

ActivityMatrix resp_to_inp

Temporary to compute extern response.

ActivityMatrix prev_map_activity

map activity from the last complete iter

ActivityListTable activity_list

Maintains activity lists.

ActivityListTable prev_activity_list

Maintains activity lists.

MarkerType activity_marker

True if neighbors are active.

SumType aff_norm_sum

To do a parallel fetch of the data.

SumType afferent_sum

To accumulate wts of each receptor.

SumType temp_sum

To keep the temporary sum.

aff_inputs_type aff_inputs

std::vector< string > aff_names

const Activity xo

Nominal offset relative to the origin.

const Activity yo

See xo.

inputs_type inputs

Set of inputs to this region.

Friends

class setfnobj2_exc_rad_setfn2

Member Function Documentation

void LissomMap::activate ( bool learn = false,

bool settle = true,

bool activatefn = true

) [virtual]

Activate the map given the current inputs.
Must be implemented by derived classes.
Implements NeuralRegion.
Definition at line 584 of file kernel.c.
References collect_activation_data(), NeuralRegion::output, present_inputs(), and NeuralRegion::Subscript.

void LissomMap::add_input ( const string & name_,

NeuralRegion & upstream_region,

WeightFunction & fn,

Length size_scale

) [virtual]

Declare that the given region gives input to this region, and define any connections needed.
The connections will have the given name. Accepts fn (a unary function of size_scale) to initialize the weights.
Implements InternalNeuralRegion.
Definition at line 607 of file kernel.c.
References assert_equals, NeuralRegion::const_activity(), Generic::insert_named(), and InternalNeuralRegion::WeightMatrix.

int LissomMap::binary_search ( const actlist & list,

int lower,

int upper,

int value

) [protected]

Binary search for value in ordered list of elements constrained to the index range [lower,upper].
If the value is found, its index is returned; otherwise the next higher index is returned.
Definition at line 1574 of file kernel.c.
Referenced by crop_actlist().

void LissomMap::collect_activation_data ( int dest_pe = Uninitialized ) [protected]

Some data is ordinarily left distributed on each PE because it is not needed on every PE for the core LISSOM algorithm.
This procedure collects such data onto the specified PE for simpler graphing and/or analysis. If pe is Uninitialized the data is broadcast to all PEs. This does not need to be called when NPES==1, but it won't take any time to do so.
Definition at line 2904 of file kernel.c.
References ipc_barrier(), and MAPROW.
Referenced by activate().

void LissomMap::compute_responses ( int ts,

bool activatefn,

bool continuousdynamics

) [protected]

Add up input and lateral activations.
Does NOT check for DEATH_FLAG. Make sure that DEATH_FLAG is much less than 1.0 so that dead weights will not significantly affect the results.
Definition at line 1439 of file kernel.c.
References ACTLIST_START, BasicLissomMap::cortex_map, ipc_barrier(), lat_resp(), lat_resp_rad1(), MAPROW, set_markers(), shuffled_rand(), sigmoid(), and Tristate.
Referenced by settle_responses().

void LissomMap::crop_actlist ( const actlist & list,

int lowval,

int highval,

int * lowidx,

int * highidx

) [inline, protected]

Given an ordered list and a range [lowval,highval], returns the range of list indexes [lowidx,highidx] which contain values in that range.
If (lowidx>highidx) on return, there is no item in the list in the given range.
Definition at line 1599 of file kernel.c.
References ACTLIST_START, and binary_search().
Referenced by lat_resp(), and modify_latwt_loop().

const LissomMap::WeightMatrix LissomMap::get_weights ( const string & name,

int ui = 0,

int uj = 0

) const [virtual]

Provides standard interface for accessing weights.
Does NOT check for DEATH_FLAG, so care needs to be taken.
Implements InternalNeuralRegion.
Definition at line 685 of file kernel.c.
References assert_equals, BasicLissomMap::cortex_map, DEATH_FLAG, ISEQ, ROWISLOCAL, and InternalNeuralRegion::WeightMatrix.

LissomMap::WeightBounds * LissomMap::get_weights_bounds ( const string & name,

int ui = 0,

int uj = 0

) const [virtual]

Allocates and returns a BoundingBox on the map for the weights of the given unit; all active weights must lie within.
This box is not guaranteed to be the minimum such box; it generally stays the same even if units are pruned, and some types may only implement a rectangular box even if weights are internally constrained to e.g. a circular region.
Implements InternalNeuralRegion.
Definition at line 798 of file kernel.c.
References array_radius_int(), assert_equals, NeuralRegion::Bounds, NeuralRegion::bounds(), BasicLissomMap::cortex_map, MAPROW, String::non_numeric_basename_matches(), ROWISLOCAL, and InternalNeuralRegion::WeightBounds.

void LissomMap::init_weights ( bool init_weights ) [protected]

Initialize active weights to specified values.
Each neuron is assumed to have a receptive field on the retina. The size of the receptive field of a neuron is (RN/N). Each neuron receives input from a square area of radius rf_radius from around the center of its receptive field projected on the retina.
The total sum of weights is normalized to 1.0.
Definition at line 2416 of file kernel.c.
References clear_weights(), BasicLissomMap::cortex_map, MAPROW, BasicLissomMap::reference_wts, setup_lateral_weights(), shuffled_rand_reset(), and Tristate.
Referenced by LissomMap().

void LissomMap::initialize_markers ( void ) [protected]

Initialize activity markers for optimization of lat_exc computations.
Initially all markers must be True.
Definition at line 1684 of file kernel.c.
References MAPROW.
Referenced by settle_responses().

double LissomMap::lat_resp ( int ui,

int uj,

int radius,

int array_width,

l_weight * weights,

int weight_idx_bound

) [protected]

Computes lateral excitation or inhibition response with the given weight vectors and radius.
The response is calculated only from active neurons. Does NOT check for DEATH_FLAG -- be careful.
Definition at line 1734 of file kernel.c.
References crop_actlist(), and PARTIAL_LAT_INDEX.
Referenced by compute_responses().

double LissomMap::lat_resp_named ( int ui,

int uj,

const string & name

) [protected]

Computes lateral excitation or inhibition response with the given named weights.
Previous activity is taken from prev_map_activity.
Definition at line 1849 of file kernel.c.
References BasicLissomMap::cortex_map.

double LissomMap::logsigmoid ( double activity,

double beta

) [protected]

Transform the activity to a log response.
f(a) = 0.0 if a<=0.0, and log(1+a) if a>0.0
Definition at line 2341 of file kernel.c.

double LissomMap::logsigmoidprime ( double activity,

double beta

) [protected]

Calculate the derivative of a log response.
f'(a) = 1/(1+a)
Definition at line 2350 of file kernel.c.

void LissomMap::modify_lat_wts ( double alpha_exc,

double alpha_inh

) [protected]

Modify lateral weights by normalized Hebbian rule.
The normalization keeps sum of lateral excitatory weights and sum of lateral inhibitory weights constant and equal to 1.0.
Definition at line 2075 of file kernel.c.
References BasicLissomMap::cortex_map, MAPROW, modify_latwt_loop(), and modify_latwt_loop_rad1().
Referenced by modify_all_weights().

void LissomMap::modify_latwt_loop_rad1 ( int ui,

int uj,

int radius,

int array_width,

int array_width_2,

l_weight * weights,

int weight_idx_bound,

double alpha

) [protected]

Optimized case when radius is just 1.
Assumes activity has been collected into the prev_map_activity array.
Definition at line 2156 of file kernel.c.
Referenced by modify_lat_wts().

void LissomMap::modify_weights ( double alpha,

bool normalize

) [protected]

Change afferent weights by normalized Hebbian rule (sum of weights normalization).
Assumption required at present; it can be eliminated except when NUM_EYES_IS_CONSTANT
Definition at line 1909 of file kernel.c.
References BasicLissomMap::cortex_map, DEATH_FLAG, MAPROW, and receptor_normalize().
Referenced by modify_all_weights().

void LissomMap::present_inputs ( bool learn,

bool settle = true,

bool activatefn = true

) [protected]

Main function called at each iteration to present inputs and modify the weights accordingly.
When this routine completes, the following data should be available to plot or analyze (on each PE):

Variable Data Portion Same on each PE? resp_to_inp sum of afferent resp nrows x N No init_activity initial activity nrows x N No prev_map_activity settled activity N x N Yes

To get all of init_activity for graphing, must call collect_activation_data() if NPES>1.
Definition at line 868 of file kernel.c.
References ACTLIST_START, modify_all_weights(), and settle_responses().
Referenced by activate().

void LissomMap::prune ( ) [virtual]

Dead connections have a small negative value on them.
These connections are not modified afterwards. Nor do they make any significant contribution to the response on the map because of their small value.
Reimplemented from NeuralRegion.
Definition at line 2750 of file kernel.c.
References BasicLissomMap::cortex_map, DEATH_FLAG, ipc_pe_msg_delay(), MAPROW, NeuralRegion::name(), and normalize_lat_wts().

void LissomMap::reduce_lateral_radius ( int ui,

int uj,

double old_radius_float,

double new_radius_float,

int array_width,

l_weight * weights,

bool circular,

bool smooth,

double smooth_trim

) [protected]

Redistribute and renormalize lateral excitation.
Lat_exc wts are indexed based on the original exc_array_width, and the current excitation radius. So the weights from the center of the exc_wts array must be copied to this new index range, for the indexing to be still correct.
Works only for reducing, not increasing lateral radius.
Definition at line 2864 of file kernel.c.
References array_radius_int(), normalize_lat_wts(), PARTIAL_LAT_INDEX, and prune_circular_lat_weights().
Referenced by change_lateral_exc_radius().

void LissomMap::response_to_input ( bool continuousdynamics ) [protected]

This code assumes that DEATH_FLAG is either (a) never used, or (b) small relative to the sum of the weights (usually 1.0).
Assumption required at present; it can be eliminated except when NUM_EYES_IS_CONSTANT
Definition at line 924 of file kernel.c.
References NeuralRegion::ActivityMatrix, BasicLissomMap::cortex_map, get_input(), inputs, MAPROW, NeuralRegion::output, and BasicLissomMap::reference_wts.
Referenced by settle_responses().

bool LissomMap::restore_state ( const int,

const string & filename

) [virtual]

Loads ASCII or binary-format snapshot from filebase.{t} from which to continue.
The ASCII format is deprecated, and may eventually be phased out.
Reimplemented from NeuralRegion.
Definition at line 3012 of file kernel.c.
References clear_weights(), ipc_abort(), and StateSaver::read().
Referenced by save_state().

void LissomMap::save_state ( const string & filename ) [virtual]

Appends snapshot to filebase.{t}.
The output is ASCII weight values.
The ASCII format is deprecated, and may eventually be phased out.
Reimplemented from NeuralRegion.
Definition at line 2938 of file kernel.c.
References clear_weights(), BasicLissomMap::cortex_map, ipc_abort(), ipc_barrier(), ipc_exit(), MAPROW, restore_state(), and StateSaver::write().

void LissomMap::set_markers ( int radius ) [protected]

Marks whether neurons are active or not.
If there is any activity in the given neighborhood radius, markers are set to True, otherwise they are set to False for the remainder of the settling iteration. Saves a lot of computation.
Definition at line 1702 of file kernel.c.
References MAPROW.
Referenced by compute_responses().

void LissomMap::setup_lateral_weights ( int ui,

int uj,

int radius,

int array_width,

l_weight * weights,

int preset,

double sigma,

double amt_of_randomness,

bool circular,

bool smooth,

double radius_trim,

double smooth_trim,

int * con_killed_flag

) [protected]

Sets up lateral weights for one neuron based on where it is in the map.
Neurons at the border have fewer connections than at the center. Note that inhibition wts are initialized to positive. The lateral wts are distributed around the mean with the given randomness. The lowk,lowl,highk,highl variables keep track of the spread of connections. Connections can at most spread to both borders of the network.
Definition at line 2586 of file kernel.c.
References dist_lat_wt(), normalize_lat_wts(), and prune_circular_lat_weights().
Referenced by init_weights().

double LissomMap::sigmoid ( double activity,

double sigdelta,

double sigbeta

) [protected]

Transform the activity to a sigmoid response between 0 and 1.0.
Between the thresholds, response is linear.
Definition at line 2297 of file kernel.c.
Referenced by compute_responses().

double LissomMap::truesigmoid ( double activity,

double beta

) [protected]

Transform the activity to a true sigmoid response between -1.0 and 1.0.
f(a) = 1+e^-a / e^a+e^-a
Definition at line 2313 of file kernel.c.
Referenced by backproject().

double LissomMap::truesigmoidprime ( double sigmoidactivity,

double beta

) [protected]

Calculate the derivative of true sigmoid response f'(a) given f(a).
f'(a) = Beta * (1-f(a)^2)
Definition at line 2323 of file kernel.c.

Member Data Documentation

const Activity LissomMap::xo [protected]

Nominal offset relative to the origin.
Only used by input_dimensions.
Definition at line 472 of file kernel.c.

const Activity LissomMap::yo [protected]

See xo.
Only used by input_dimensions.
Definition at line 472 of file kernel.c.

The documentation for this class was generated from the following file:

kernel.c

Generated on Mon Jan 20 02:37:03 2003 for RF-LISSOM by

1.3-rc2


Public Methods
	LissomMap (const string &name_i, Dimensions dims, bool init_weights_, ParamMap *params_)
	Constructor.
virtual void	activate (bool learn=false, bool settle=true, bool activatefn=true)
	Activate the map given the current inputs.
virtual void	backproject ()
	yschoe: backprojection from map to input: this code was ripped from response_to_input, and made unoptimized.
virtual void	backproject (const string &name, const double gammaaff=1.0)
	Backproject local core routine: replaces call to parent class' slower backproject().
virtual const WeightMatrix	get_weights (const string &name, int ui=0, int uj=0) const
	Provides standard interface for accessing weights.
virtual NeuralRegion *	get_input (const string &name)
	Returns the specified input region for a unit.
virtual void	prune ()
	Dead connections have a small negative value on them.
virtual void	save_state (const string &)
	Appends snapshot to filebase.{t}.
virtual bool	restore_state (const int, const string &)
	Loads ASCII or binary-format snapshot from filebase.{t} from which to continue.
virtual double	size_connection_bytes () const
	Memory requirements per PE.
virtual double	size_unique_connections () const
	Total number of connections.
virtual void	add_input (const string &name_, NeuralRegion &upstream_region, WeightFunction &fn, Length size_scale)
	Declare that the given region gives input to this region, and define any connections needed.
virtual Dimensions	input_dimensions (WeightFunction &fn, Length size_scale=1.0)
	Computes input dimensions that would be required to avoid border effects.
virtual WeightBounds *	get_weights_bounds (const string &name, int ui=0, int uj=0) const
	Allocates and returns a BoundingBox on the map for the weights of the given unit; all active weights must lie within.
virtual int	num_aff_inputs () const
Protected Types
typedef std::vector< actlist >	ActivityListTable
typedef MatrixType< int >::rectangular	MarkerType
typedef std::vector< double >	SumType
typedef NeuralRegion::ActivityMatrix	aff_input_type
typedef std::vector< const aff_input_type * >	aff_inputs_type
typedef Boundary::BoundingCircle< int, int >	BoundingCircle
typedef Boundary::AARBoundingBox< int, int >	BoundingRectangle
typedef std::vector< Input * >	inputs_type
	Type to use for storing inputs.
Protected Methods
void	present_inputs (bool learn, bool settle=true, bool activate=true)
	Main function called at each iteration to present inputs and modify the weights accordingly.
void	init_weights (bool init_weights)
	Initialize active weights to specified values.
void	collect_activation_data (int dest_pe=Uninitialized)
	Some data is ordinarily left distributed on each PE because it is not needed on every PE for the core LISSOM algorithm.
double	dist_lat_wt (double value, double amt_of_randomness)
	This function returns a value randomly distributed within a range valueamt_of_randomness on either side of value.*
double	lat_resp (int ui, int uj, int radius, int array_width, l_weight *weights, int weight_idx_bound)
	Computes lateral excitation or inhibition response with the given weight vectors and radius.
double	lat_resp_rad1 (int ui, int uj, int radius, int array_width, int array_width_2, l_weight *weights, int weight_idx_bound)
	Optimized case when radius is just 0 or 1, mainly for excitation resp of organized map.
double	lat_resp_named (int ui, int uj, const string &name)
	Computes lateral excitation or inhibition response with the given named weights.
double	sigmoid (double activity, double sigdelta, double sigbeta)
	Transform the activity to a sigmoid response between 0 and 1.0.
double	truesigmoid (double activity, double beta)
	Transform the activity to a true sigmoid response between -1.0 and 1.0.
double	truesigmoidprime (double sigmoidactivity, double beta)
	Calculate the derivative of true sigmoid response f'(a) given f(a).
double	positivesigmoid (double activity, double beta)
	Positive sigmoid shifted to the right f(a) = 1/(1+exp(-(x-0.5)beta)).*
double	logsigmoid (double activity, double beta)
	Transform the activity to a log response.
double	logsigmoidprime (double activity, double beta)
	Calculate the derivative of a log response.
int	binary_search (const actlist &list, int lower, int upper, int value)
	Binary search for value in ordered list of elements constrained to the index range [lower,upper].
double	change_lateral_exc_radius (double old_radius, double new_radius)
	Change lateral excitatory radius for every row.
void	clear_weights (void)
	Intialize all weights to known values.
void	randomize_lat_wts (void)
void	compute_responses (int ts, bool activate, bool continuousdynamics)
	Add up input and lateral activations.
void	crop_actlist (const actlist &, int lowval, int highval, int lowidx, int highidx)
	Given an ordered list and a range [lowval,highval], returns the range of list indexes [lowidx,highidx] which contain values in that range.
void	initialize_actlists (void)
	Initialize the activity list counts to zero.
void	initialize_markers (void)
	Initialize activity markers for optimization of lat_exc computations.
void	modify_all_weights (void)
	Updates each set of weights, if the associated learning rate is nonzero.
void	modify_lat_wts (double alpha_exc, double alpha_inh)
	Modify lateral weights by normalized Hebbian rule.
void	modify_latwt_loop (int i, int j, int radius, int array_width, l_weight *weights, int weight_idx_bound, double alpha)
	Assumes activity has been collected into the prev_map_activity array.
void	modify_latwt_loop_killed (int i, int j, int radius, int array_width, l_weight *weights, int weight_idx_bound, double alpha)
void	modify_latwt_loop_rad1 (int ui, int uj, int radius, int array_width, int array_width_2, l_weight *weights, int weight_idx_bound, double alpha)
	Optimized case when radius is just 1.
void	modify_weights (double alpha, bool normalize)
	Change afferent weights by normalized Hebbian rule (sum of weights normalization).
void	normalize_afferent_wts (int lowk, int highk, int lowl, int highl, int row, int j, double length)
void	normalize_lat_wts (int i, int j, int radius, int array_width, l_weight *weights)
	Renormalize a set of lateral weights of the given neuron, such as after changing the radius.
double	prune_circular_aff_weights (int ui, int uj, int radius, AfferentWeightsList &weights, bool smooth, double radius_trim, double smooth_trim, double length)
	Kill off connections that aren't within the specified radius.
void	prune_circular_lat_weights (int ui, int uj, int radius_sq, int array_width, l_weight *weights, bool smooth, double radius_trim, double smooth_trim)
	Kill off connections that aren't within the specified radius.
void	receptor_normalize (void)
	Output normalization for receptors.
void	reduce_lateral_radius (int ui, int uj, double old_radius, double new_radius, int array_width, l_weight *weights, bool circular, bool smooth, double smooth_trim)
	Redistribute and renormalize lateral excitation.
void	response_to_input (bool continuousdynamics)
	This code assumes that DEATH_FLAG is either (a) never used, or (b) small relative to the sum of the weights (usually 1.0).
void	set_markers (int radius)
	Marks whether neurons are active or not.
void	settle_responses (bool learn, int tsettle, bool activate, bool continuousdynamics)
	Propagate input and let the network settle.
void	setup_afferent_weights (Neuron &neuron, AfferentWeightsList &awts, int ui, int uj, int radius, double radius_sq, int preset, double sigma, double amt_of_randomness, bool circular, bool smooth, double radius_trim, double smooth_trim, double force_cx=-1, double force_cy=-1)
void	setup_lateral_weights (int ui, int uj, int radius, int array_width, l_weight weights, int preset, double sigma, double amt_of_randomness, bool circular, bool smooth, double radius_trim, double smooth_trim, int con_killed_flag)
	Sets up lateral weights for one neuron based on where it is in the map.
void	verify_saved_lateral_weights (int i, int j, int radius, int array_width, l_weight old_weights, l_weight new_weights, double tolerance, const char *description)
void	verify_saved_weights (double tolerance)
cmdstat	exc_rad_setfn2 (const string &name, Typeless &param, const Typeless &newvalue)
	Automatically called by cmdparam.c to set this parameter.
int	array_radius_int (const ParamMap &parammap, const string &radius_param) const
	Given a radius parameter, returns the actual integer radius to be used, taking smoothing into account.
int	array_radius_int (const double radius_float) const
Subscript	input_row (const Input &input, Subscript row) const
Subscript	input_col (const Input &input, Subscript col) const
Protected Attributes
OwningPointer< ParamMap >	params
const int	num_aff
int	exc_connections_killed
	Flag for use in modify_weights.
int	inh_connections_killed
	Flag for use in modify_weights.
const int	exc_array_width_2
	Holds 2exc_array_width (optimization).*
const int	RN_sq
	RNRN precomputed for index calculations.*
ActivityMatrix	gather_activity
	To collect activity in each row.
ActivityMatrix	init_activity
	Initial map_activity over network.
ActivityMatrix	map_activity
	Current map_activity over network.
ActivityMatrix	resp_to_inp
	Temporary to compute extern response.
ActivityMatrix	prev_map_activity
	map activity from the last complete iter
ActivityListTable	activity_list
	Maintains activity lists.
ActivityListTable	prev_activity_list
	Maintains activity lists.
MarkerType	activity_marker
	True if neighbors are active.
SumType	aff_norm_sum
	To do a parallel fetch of the data.
SumType	afferent_sum
	To accumulate wts of each receptor.
SumType	temp_sum
	To keep the temporary sum.
aff_inputs_type	aff_inputs
std::vector< string >	aff_names
const Activity	xo
	Nominal offset relative to the origin.
const Activity	yo
	See xo.
inputs_type	inputs
	Set of inputs to this region.
Friends
class	setfnobj2_exc_rad_setfn2