Logo ROOT  
Reference Guide
 
Loading...
Searching...
No Matches
TNeuron.cxx
Go to the documentation of this file.
1// @(#)root/tmva $Id$
2// Author: Matt Jachowski
3
4/**********************************************************************************
5 * Project: TMVA - a Root-integrated toolkit for multivariate data analysis *
6 * Package: TMVA *
7 * Class : TNeuron *
8 * Web : http://tmva.sourceforge.net *
9 * *
10 * Description: *
11 * Implementation (see header for description) *
12 * *
13 * Authors (alphabetical): *
14 * Matt Jachowski <jachowski@stanford.edu> - Stanford University, USA *
15 * *
16 * Copyright (c) 2005: *
17 * CERN, Switzerland *
18 * *
19 * Redistribution and use in source and binary forms, with or without *
20 * modification, are permitted according to the terms listed in LICENSE *
21 * (http://tmva.sourceforge.net/LICENSE) *
22 **********************************************************************************/
23
24/*! \class TMVA::TNeuron
25\ingroup TMVA
26Neuron class used by TMVA artificial neural network methods
27*/
28
29#include "TMVA/TNeuron.h"
30
31#include "TMVA/MsgLogger.h"
32#include "TMVA/TActivation.h"
33#include "TMVA/Tools.h"
34#include "TMVA/TNeuronInput.h"
35#include "TMVA/Types.h"
36
37#include "TH1D.h"
38#include "ThreadLocalStorage.h"
39#include "TObjArray.h"
40
41static const Int_t UNINITIALIZED = -1;
42
43using std::vector;
44
46
47////////////////////////////////////////////////////////////////////////////////
48/// standard constructor
49
51{
52 InitNeuron();
53}
54
55////////////////////////////////////////////////////////////////////////////////
56/// destructor
57
59{
60 if (fLinksIn != NULL) delete fLinksIn;
61 if (fLinksOut != NULL) delete fLinksOut;
62}
63
64////////////////////////////////////////////////////////////////////////////////
65/// initialize the neuron, most variables still need to be set via setters
66
68{
69 fLinksIn = new TObjArray();
70 fLinksOut = new TObjArray();
71 fValue = UNINITIALIZED;
72 fActivationValue = UNINITIALIZED;
73 fDelta = UNINITIALIZED;
74 fDEDw = UNINITIALIZED;
75 fError = UNINITIALIZED;
76 fActivation = NULL;
77 fForcedValue = kFALSE;
78 fInputCalculator = NULL;
79}
80
81////////////////////////////////////////////////////////////////////////////////
82/// force the value, typically for input and bias neurons
83
85{
86 fValue = value;
87 fForcedValue = kTRUE;
88}
89
90////////////////////////////////////////////////////////////////////////////////
91/// calculate neuron input
92
94{
95 if (fForcedValue) return;
96 fValue = fInputCalculator->GetInput(this);
97}
98
99////////////////////////////////////////////////////////////////////////////////
100/// calculate neuron activation/output
101
103{
104 if (fActivation == NULL) {
105 PrintMessage( kWARNING ,"No activation equation specified." );
106 fActivationValue = UNINITIALIZED;
107 return;
108 }
109 fActivationValue = fActivation->Eval(fValue);
110}
111
112////////////////////////////////////////////////////////////////////////////////
113/// calculate error field
114
116{
117 // no need to adjust input neurons
118 if (IsInputNeuron()) {
119 fDelta = 0.0;
120 return;
121 }
122
123 Double_t error;
124
125 // output neuron should have error set all ready
126 if (IsOutputNeuron()) error = fError;
127
128 // need to calculate error for any other neuron
129 else {
130 error = 0.0;
131 TSynapse* synapse = NULL;
132 // Replaced TObjArrayIter pointer by object, as creating it on the stack
133 // is much faster (5-10% improvement seen) than re-allocating the new
134 // memory for the pointer each time. Thanks to Peter Elmer who pointed this out
135 // TObjArrayIter* iter = (TObjArrayIter*)fLinksOut->MakeIterator();
136 TObjArrayIter iter(fLinksOut);
137 while (true) {
138 synapse = (TSynapse*) iter.Next();
139 if (synapse == NULL) break;
140 error += synapse->GetWeightedDelta();
141 }
142
143 }
144
145 fDelta = error * fActivation->EvalDerivative(GetValue());
146}
147
148////////////////////////////////////////////////////////////////////////////////
149/// set input calculator
150
152{
153 if (fInputCalculator != NULL) delete fInputCalculator;
154 fInputCalculator = calculator;
155}
156
157////////////////////////////////////////////////////////////////////////////////
158/// set activation equation
159
161{
162 if (fActivation != NULL) delete fActivation;
163 fActivation = activation;
164}
165
166////////////////////////////////////////////////////////////////////////////////
167/// add synapse as a pre-link to this neuron
168
170{
171 if (IsInputNeuron()) return;
172 fLinksIn->Add(pre);
173}
174
175////////////////////////////////////////////////////////////////////////////////
176/// add synapse as a post-link to this neuron
177
179{
180 if (IsOutputNeuron()) return;
181 fLinksOut->Add(post);
182}
183
184////////////////////////////////////////////////////////////////////////////////
185/// delete all pre-links
186
188{
189 DeleteLinksArray(fLinksIn);
190}
191
192////////////////////////////////////////////////////////////////////////////////
193/// delete an array of TSynapses
194
196{
197 if (links == NULL) return;
198
199 TSynapse* synapse = NULL;
200 Int_t numLinks = links->GetEntriesFast();
201 for (Int_t i=0; i<numLinks; i++) {
202 synapse = (TSynapse*)links->At(i);
203 if (synapse != NULL) delete synapse;
204 }
205 delete links;
206 links = NULL;
207}
208
209////////////////////////////////////////////////////////////////////////////////
210/// set error, this should only be done for an output neuron
211
213{
214 if (!IsOutputNeuron())
215 PrintMessage( kWARNING, "Warning! Setting an error on a non-output neuron is probably not what you want to do." );
216
217 fError = error;
218}
219
220////////////////////////////////////////////////////////////////////////////////
221/// update and adjust the pre-synapses for each neuron (input neuron has no pre-synapse)
222/// this method should only be called in batch mode
223
225{
226 if (IsInputNeuron()) return;
227
228 TSynapse* synapse = NULL;
229 TObjArrayIter iter(fLinksIn);
230 while (true) {
231 synapse = (TSynapse*) iter.Next();
232 if (synapse == NULL) break;
233 synapse->CalculateDelta();
234 }
235
236}
237
238////////////////////////////////////////////////////////////////////////////////
239/// update the pre-synapses for each neuron (input neuron has no pre-synapse)
240/// this method should only be called in sequential mode
241
243{
244 if (IsInputNeuron()) return;
245
246 TSynapse* synapse = NULL;
247 TObjArrayIter iter(fLinksIn);
248
249 while (true) {
250 synapse = (TSynapse*) iter.Next();
251 if (synapse == NULL) break;
252 synapse->InitDelta();
253 synapse->CalculateDelta();
254 synapse->AdjustWeight();
255 }
256
257}
258
259////////////////////////////////////////////////////////////////////////////////
260/// adjust the pre-synapses' weights for each neuron (input neuron has no pre-synapse)
261/// this method should only be called in batch mode
262
264{
265 if (IsInputNeuron()) return;
266
267 TSynapse* synapse = NULL;
268 TObjArrayIter iter(fLinksIn);
269
270
271 while (true) {
272 synapse = (TSynapse*) iter.Next();
273 if (synapse == NULL) break;
274 synapse->AdjustWeight();
275 }
276
277}
278
279////////////////////////////////////////////////////////////////////////////////
280/// initialize the error fields of all pre-neurons
281/// this method should only be called in batch mode
282
284{
285 // an input neuron has no pre-weights to adjust
286 if (IsInputNeuron()) return;
287
288 TSynapse* synapse = NULL;
289 TObjArrayIter iter(fLinksIn);
290
291 while (true) {
292 synapse = (TSynapse*) iter.Next();
293
294 if (synapse == NULL) break;
295 synapse->InitDelta();
296 }
297
298}
299
300////////////////////////////////////////////////////////////////////////////////
301/// print an array of TSynapses, for debugging
302
304{
305 if (links == NULL) {
306 Log() << kDEBUG << "\t\t\t<none>" << Endl;
307 return;
308 }
309
310 TSynapse* synapse;
311
312 Int_t numLinks = links->GetEntriesFast();
313 for (Int_t i = 0; i < numLinks; i++) {
314 synapse = (TSynapse*)links->At(i);
315 Log() << kDEBUG <<
316 "\t\t\tweighta: " << synapse->GetWeight()
317 << "\t\tw-value: " << synapse->GetWeightedValue()
318 << "\t\tw-delta: " << synapse->GetWeightedDelta()
319 << "\t\tl-rate: " << synapse->GetLearningRate()
320 << Endl;
321 }
322}
323
324////////////////////////////////////////////////////////////////////////////////
325/// print activation equation, for debugging
326
328{
329 if (fActivation != NULL) Log() << kDEBUG << fActivation->GetExpression() << Endl;
330 else Log() << kDEBUG << "<none>" << Endl;
331}
332
333////////////////////////////////////////////////////////////////////////////////
334/// print message, for debugging
335
337{
338 Log() << type << message << Endl;
339}
340
341////////////////////////////////////////////////////////////////////////////////
342
344{
345 TTHREAD_TLS_DECL_ARG2(MsgLogger,logger,"TNeuron",kDEBUG); //! message logger, static to save resources
346 return logger;
347}
constexpr Bool_t kFALSE
Definition RtypesCore.h:101
constexpr Bool_t kTRUE
Definition RtypesCore.h:100
#define ClassImp(name)
Definition Rtypes.h:377
Option_t Option_t TPoint TPoint const char GetTextMagnitude GetFillStyle GetLineColor GetLineWidth GetMarkerStyle GetTextAlign GetTextColor GetTextSize void value
Option_t Option_t TPoint TPoint const char GetTextMagnitude GetFillStyle GetLineColor GetLineWidth GetMarkerStyle GetTextAlign GetTextColor GetTextSize void char Point_t Rectangle_t WindowAttributes_t Float_t Float_t Float_t Int_t Int_t UInt_t UInt_t Rectangle_t Int_t Int_t Window_t TString Int_t GCValues_t GetPrimarySelectionOwner GetDisplay GetScreen GetColormap GetNativeEvent const char const char dpyName wid window const char font_name cursor keysym reg const char only_if_exist regb h Point_t winding char text const char depth char const char Int_t count const char ColorStruct_t color const char Pixmap_t Pixmap_t PictureAttributes_t attr const char char ret_data h unsigned char height h Atom_t Int_t ULong_t ULong_t unsigned char prop_list Atom_t Atom_t Atom_t Time_t type
ostringstream derivative to redirect and format output
Definition MsgLogger.h:57
Interface for TNeuron activation function classes.
Definition TActivation.h:42
Interface for TNeuron input calculation classes.
Neuron class used by TMVA artificial neural network methods.
Definition TNeuron.h:49
void AdjustSynapseWeights()
adjust the pre-synapses' weights for each neuron (input neuron has no pre-synapse) this method should...
Definition TNeuron.cxx:263
void ForceValue(Double_t value)
force the value, typically for input and bias neurons
Definition TNeuron.cxx:84
TNeuron()
standard constructor
Definition TNeuron.cxx:50
void UpdateSynapsesSequential()
update the pre-synapses for each neuron (input neuron has no pre-synapse) this method should only be ...
Definition TNeuron.cxx:242
void PrintMessage(EMsgType, TString message)
print message, for debugging
Definition TNeuron.cxx:336
void SetActivationEqn(TActivation *activation)
set activation equation
Definition TNeuron.cxx:160
void InitNeuron()
initialize the neuron, most variables still need to be set via setters
Definition TNeuron.cxx:67
MsgLogger & Log() const
Definition TNeuron.cxx:343
void AddPostLink(TSynapse *post)
add synapse as a post-link to this neuron
Definition TNeuron.cxx:178
void SetInputCalculator(TNeuronInput *calculator)
set input calculator
Definition TNeuron.cxx:151
void PrintActivationEqn()
print activation equation, for debugging
Definition TNeuron.cxx:327
void SetError(Double_t error)
set error, this should only be done for an output neuron
Definition TNeuron.cxx:212
void CalculateValue()
calculate neuron input
Definition TNeuron.cxx:93
void CalculateActivationValue()
calculate neuron activation/output
Definition TNeuron.cxx:102
void UpdateSynapsesBatch()
update and adjust the pre-synapses for each neuron (input neuron has no pre-synapse) this method shou...
Definition TNeuron.cxx:224
void DeleteLinksArray(TObjArray *&links)
delete an array of TSynapses
Definition TNeuron.cxx:195
virtual ~TNeuron()
destructor
Definition TNeuron.cxx:58
void AddPreLink(TSynapse *pre)
add synapse as a pre-link to this neuron
Definition TNeuron.cxx:169
void DeletePreLinks()
delete all pre-links
Definition TNeuron.cxx:187
void InitSynapseDeltas()
initialize the error fields of all pre-neurons this method should only be called in batch mode
Definition TNeuron.cxx:283
void CalculateDelta()
calculate error field
Definition TNeuron.cxx:115
void PrintLinks(TObjArray *links) const
print an array of TSynapses, for debugging
Definition TNeuron.cxx:303
Synapse class used by TMVA artificial neural network methods.
Definition TSynapse.h:42
Double_t GetWeight()
Definition TSynapse.h:53
Double_t GetWeightedValue()
get output of pre-neuron weighted by synapse weight
Definition TSynapse.cxx:76
Double_t GetLearningRate()
Definition TSynapse.h:59
Double_t GetWeightedDelta()
get error field of post-neuron weighted by synapse weight
Definition TSynapse.cxx:87
void InitDelta()
Definition TSynapse.h:83
void AdjustWeight()
adjust the weight based on the error field all ready calculated by CalculateDelta
Definition TSynapse.cxx:98
void CalculateDelta()
calculate/adjust the error field for this synapse
Definition TSynapse.cxx:108
Iterator of object array.
Definition TObjArray.h:117
TObject * Next() override
Return next object in array. Returns 0 when no more objects in array.
An array of TObjects.
Definition TObjArray.h:31
Int_t GetEntriesFast() const
Definition TObjArray.h:58
TObject * At(Int_t idx) const override
Definition TObjArray.h:164
Basic string class.
Definition TString.h:139
MsgLogger & Endl(MsgLogger &ml)
Definition MsgLogger.h:148
static const Int_t UNINITIALIZED
Definition TNeuron.cxx:41