Skip to content

Commit 4e0ef33

Browse files
Neel-Shah-29Neel-Shah-29
authored
Some error messages and spelling mistakes fixed related to operators (root-project#10435)
* Some error messages and spelling mistakes fixed related to operaters * Update the indentation of BatchNormalization.hxx Co-authored-by: Neel Shah <neelshah29042002.com>
1 parent 8baea4c commit 4e0ef33

File tree

6 files changed

+183
-183
lines changed

6 files changed

+183
-183
lines changed

‎tmva/sofie/inc/TMVA/ROperator_Add.hxx‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -45,14 +45,14 @@ public:
4545
throw std::runtime_error(std::string("TMVA SOFIE Add Op Input Tensor ") + fNX1 + "is not found in model");
4646
}
4747
if (model.CheckIfTensorAlreadyExist(fNX2) == false) {
48-
throw std::runtime_error(std::string("TMVA SOFIE Add Op Input Tensor ") + fNX1 + "is not found in model");
48+
throw std::runtime_error(std::string("TMVA SOFIE Add Op Input Tensor ") + fNX2 + "is not found in model");
4949
}
5050
auto shapeX1 = model.GetTensorShape(fNX1);
5151
auto shapeX2 = model.GetTensorShape(fNX2);
5252
// assume same shape X1 and X2
5353
if (shapeX1 != shapeX2) {
5454
std::string msg = "TMVA SOFIE Add Op: Support only inputs with same shape, shape 1 is " +
55-
ConvertShapeToString(shapeX1) + "shape 2 is " + ConvertShapeToString(shapeX2);
55+
ConvertShapeToString(shapeX1) + "and shape 2 is " + ConvertShapeToString(shapeX2);
5656
throw std::runtime_error(msg);
5757
}
5858
fShape = shapeX1;

‎tmva/sofie/inc/TMVA/ROperator_BatchNormalization.hxx‎

Lines changed: 167 additions & 167 deletions
Original file line numberDiff line numberDiff line change
@@ -19,203 +19,203 @@ class ROperator_BatchNormalization final : public ROperator
1919

2020
private:
2121

22-
/* Attributes */
23-
float fepsilon = 1e-05;
24-
float fmomentum = 0.9;
25-
std::size_t ftraining_mode = 0;
26-
27-
std::string fNX;
28-
std::string fNScale;
29-
std::string fNB;
30-
std::string fNMean;
31-
std::string fNVar;
32-
std::string fNY;
33-
34-
std::vector<size_t> fShapeX;
35-
std::vector<size_t> fShapeScale;
36-
std::vector<size_t> fShapeB;
37-
std::vector<size_t> fShapeMean;
38-
std::vector<size_t> fShapeVar;
39-
std::vector<size_t> fShapeY;
40-
41-
std::string fType;
22+
/* Attributes */
23+
float fepsilon = 1e-05;
24+
float fmomentum = 0.9;
25+
std::size_t ftraining_mode = 0;
26+
27+
std::string fNX;
28+
std::string fNScale;
29+
std::string fNB;
30+
std::string fNMean;
31+
std::string fNVar;
32+
std::string fNY;
33+
34+
std::vector<size_t> fShapeX;
35+
std::vector<size_t> fShapeScale;
36+
std::vector<size_t> fShapeB;
37+
std::vector<size_t> fShapeMean;
38+
std::vector<size_t> fShapeVar;
39+
std::vector<size_t> fShapeY;
40+
41+
std::string fType;
4242

4343
public:
44-
ROperator_BatchNormalization() = delete;
45-
46-
/* Constructor */
47-
ROperator_BatchNormalization( float epsilon, float momentum, std::size_t training_mode,
48-
std::string nameX, std::string nameScale, std::string nameB,
49-
std::string nameMean, std::string nameVar, std::string nameY):
50-
fepsilon(epsilon), fmomentum(momentum), ftraining_mode(training_mode),
51-
fNX(UTILITY::Clean_name(nameX)), fNScale(UTILITY::Clean_name(nameScale)),
52-
fNB(UTILITY::Clean_name(nameB)), fNMean(UTILITY::Clean_name(nameMean)),
53-
fNVar(UTILITY::Clean_name(nameVar)), fNY(UTILITY::Clean_name(nameY))
54-
{
55-
if(std::is_same<T, float>::value){
56-
fType = "float";
57-
}
58-
else{
59-
throw
60-
std::runtime_error("TMVA SOFIE Encountered unsupported type parsing a BatchNormalization operator");
61-
}
62-
}
44+
ROperator_BatchNormalization() = delete;
45+
46+
/* Constructor */
47+
ROperator_BatchNormalization( float epsilon, float momentum, std::size_t training_mode,
48+
std::string nameX, std::string nameScale, std::string nameB,
49+
std::string nameMean, std::string nameVar, std::string nameY):
50+
fepsilon(epsilon), fmomentum(momentum), ftraining_mode(training_mode),
51+
fNX(UTILITY::Clean_name(nameX)), fNScale(UTILITY::Clean_name(nameScale)),
52+
fNB(UTILITY::Clean_name(nameB)), fNMean(UTILITY::Clean_name(nameMean)),
53+
fNVar(UTILITY::Clean_name(nameVar)), fNY(UTILITY::Clean_name(nameY))
54+
{
55+
if(std::is_same<T, float>::value){
56+
fType = "float";
57+
}
58+
else{
59+
throw
60+
std::runtime_error("TMVA SOFIE Encountered unsupported type parsing a BatchNormalization operator");
61+
}
62+
}
6363

6464

65-
std::vector<ETensorType> TypeInference(std::vector<ETensorType> input) {
66-
ETensorType out = input[0];
67-
return {out};
68-
}
69-
70-
std::vector<std::vector<size_t>> ShapeInference(std::vector<std::vector<size_t>> input) {
71-
if (input.size() != 5 ) {
72-
throw
73-
std::runtime_error("TMVA SOFIE BatchNormalization Op Shape inference need 5 input tensors");
74-
}
75-
for(size_t i = 0; i < input.size(); i++) {
76-
if (input[i].size() != 4) {
77-
throw
78-
std::runtime_error("TMVA SOFIE BatchNormalization Op Shape inference only accept tensor with 4 dimensions");
79-
}
80-
}
81-
82-
auto ret = input;
83-
return ret;
84-
}
85-
86-
void Initialize(RModel& model){
87-
if (!model.CheckIfTensorAlreadyExist(fNX)) {
88-
throw
89-
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNX + " fnx is not found in model");
90-
}
91-
if (!model.CheckIfTensorAlreadyExist(fNScale)) {
92-
throw
93-
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNScale + " fns is not found in model");
94-
}
95-
if (!model.CheckIfTensorAlreadyExist(fNB)) {
96-
throw
97-
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNB + " fnb is not found in model");
98-
}
99-
if (!model.CheckIfTensorAlreadyExist(fNMean)) {
100-
throw
101-
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNMean + " fnm is not found in model");
102-
}
103-
if (!model.CheckIfTensorAlreadyExist(fNVar)) {
104-
throw
105-
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNVar + " fnv is not found in model");
106-
}
107-
108-
fShapeX = model.GetTensorShape(fNX);
109-
65+
std::vector<ETensorType> TypeInference(std::vector<ETensorType> input) {
66+
ETensorType out = input[0];
67+
return {out};
68+
}
69+
70+
std::vector<std::vector<size_t>> ShapeInference(std::vector<std::vector<size_t>> input) {
71+
if (input.size() != 5 ) {
72+
throw
73+
std::runtime_error("TMVA SOFIE BatchNormalization Op Shape inference need 5 input tensors");
74+
}
75+
for(size_t i = 0; i < input.size(); i++) {
76+
if (input[i].size() != 4) {
77+
throw
78+
std::runtime_error("TMVA SOFIE BatchNormalization Op Shape inference only accept tensor with 4 dimensions");
79+
}
80+
}
81+
82+
auto ret = input;
83+
return ret;
84+
}
85+
86+
void Initialize(RModel& model){
87+
if (!model.CheckIfTensorAlreadyExist(fNX)) {
88+
throw
89+
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNX + " fnx is not found in model");
90+
}
91+
if (!model.CheckIfTensorAlreadyExist(fNScale)) {
92+
throw
93+
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNScale + " fns is not found in model");
94+
}
95+
if (!model.CheckIfTensorAlreadyExist(fNB)) {
96+
throw
97+
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNB + " fnb is not found in model");
98+
}
99+
if (!model.CheckIfTensorAlreadyExist(fNMean)) {
100+
throw
101+
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNMean + " fnm is not found in model");
102+
}
103+
if (!model.CheckIfTensorAlreadyExist(fNVar)) {
104+
throw
105+
std::runtime_error("TMVA SOFIE BatchNormalization op Input Tensor " + fNVar + " fnv is not found in model");
106+
}
107+
108+
fShapeX = model.GetTensorShape(fNX);
109+
110110
if (fShapeX.size() < 2 || fShapeX.size() > 4) {
111111
throw
112-
std::runtime_error("TMVA SOFIE BatchNormalization Op input tensor " + fNX + " fnx has wrong shape : " + ConvertShapeToString(fShapeX));
112+
std::runtime_error("TMVA SOFIE BatchNormalization Op input tensor " + fNX + " fnx has wrong shape : " + ConvertShapeToString(fShapeX));
113113
}
114114

115115
fShapeScale = model.GetTensorShape(fNScale);
116-
fShapeB = model.GetTensorShape(fNB);
117-
fShapeMean = model.GetTensorShape(fNMean);
118-
fShapeVar = model.GetTensorShape(fNVar);
119-
fShapeY = fShapeX;
120-
model.AddIntermediateTensor(fNY, model.GetTensorType(fNX), fShapeY);
121-
122-
if (fShapeB.size() == 1) {
123-
// Broadcast scale, bias, input_mean and input_var to shape_X
124-
auto original_B = model.GetInitializedTensorData(fNB);
125-
auto original_S = model.GetInitializedTensorData(fNScale);
126-
auto original_M = model.GetInitializedTensorData(fNMean);
127-
auto original_V = model.GetInitializedTensorData(fNVar);
128-
size_t batchSize = fShapeX[0];
129-
size_t channels = fShapeX[1];
130-
size_t height = (fShapeX.size() > 2) ? fShapeX[2] : 1;
131-
size_t width = (fShapeX.size() > 3) ? fShapeX[3] : 1;
132-
size_t n = batchSize * channels * height * width;
116+
fShapeB = model.GetTensorShape(fNB);
117+
fShapeMean = model.GetTensorShape(fNMean);
118+
fShapeVar = model.GetTensorShape(fNVar);
119+
fShapeY = fShapeX;
120+
model.AddIntermediateTensor(fNY, model.GetTensorType(fNX), fShapeY);
121+
122+
if (fShapeB.size() == 1) {
123+
// Broadcast scale, bias, input_mean and input_var to shape_X
124+
auto original_B = model.GetInitializedTensorData(fNB);
125+
auto original_S = model.GetInitializedTensorData(fNScale);
126+
auto original_M = model.GetInitializedTensorData(fNMean);
127+
auto original_V = model.GetInitializedTensorData(fNVar);
128+
size_t batchSize = fShapeX[0];
129+
size_t channels = fShapeX[1];
130+
size_t height = (fShapeX.size() > 2) ? fShapeX[2] : 1;
131+
size_t width = (fShapeX.size() > 3) ? fShapeX[3] : 1;
132+
size_t n = batchSize * channels * height * width;
133133
if (fType == "float") {
134-
float *original_bias = static_cast<float*>(original_B.get());
135-
float *original_scale = static_cast<float*>(original_S.get());
136-
float *original_mean = static_cast<float*>(original_M.get());
137-
float *original_var = static_cast<float*>(original_V.get());
138-
float *new_bias = new float[n];
139-
float *new_scale = new float[n];
140-
float *new_mean = new float[n];
141-
float *new_var = new float[n];
142-
size_t bs = 0, ch = 0, h = 0, w = 0;
143-
for(ch=0; ch<channels; ch++){
144-
for(h=0; h<height; h++){
145-
for(w=0; w<width; w++){
146-
new_bias[bs*channels*height*width + ch*height*width + h*width + w] = original_bias[ch];
147-
new_scale[bs*channels*height*width + ch*height*width + h*width + w] = original_scale[ch];
148-
new_mean[bs*channels*height*width + ch*height*width + h*width + w] = original_mean[ch];
149-
new_var[bs*channels*height*width + ch*height*width + h*width + w] = original_var[ch];
150-
}
151-
}
152-
}
153-
size_t Batchoffset = channels*height*width;
154-
for(bs = 1; bs<batchSize; bs++){
155-
std::copy(new_bias, new_bias+Batchoffset, new_bias+(bs*Batchoffset));
156-
std::copy(new_scale, new_scale+Batchoffset, new_scale+(bs*Batchoffset));
157-
std::copy(new_mean, new_mean+Batchoffset, new_mean+(bs*Batchoffset));
158-
std::copy(new_var, new_var+Batchoffset, new_var+(bs*Batchoffset));
159-
}
160-
//// new_var =1. / sqrt(input_var + fepsilon)
161-
for(size_t i=0; i<n; i++){
162-
new_var[i] = 1./sqrt(new_var[i] + fepsilon);
163-
}
164-
std::vector<size_t> new_bias_shape = {batchSize,channels,height,width};
165-
std::shared_ptr<void> new_bias_ptr(new_bias, std::default_delete<float[]>());
166-
std::shared_ptr<void> new_scale_ptr(new_scale, std::default_delete<float[]>());
167-
std::shared_ptr<void> new_mean_ptr(new_mean, std::default_delete<float[]>());
168-
std::shared_ptr<void> new_var_ptr(new_var, std::default_delete<float[]>());
169-
model.UpdateInitializedTensor(fNB, model.GetTensorType(fNB), new_bias_shape, new_bias_ptr);
170-
model.UpdateInitializedTensor(fNScale, model.GetTensorType(fNScale), new_bias_shape, new_scale_ptr);
171-
model.UpdateInitializedTensor(fNMean, model.GetTensorType(fNMean), new_bias_shape, new_mean_ptr);
172-
model.UpdateInitializedTensor(fNVar, model.GetTensorType(fNVar), new_bias_shape, new_var_ptr);
173-
fShapeB = model.GetTensorShape(fNB);
174-
fShapeScale = model.GetTensorShape(fNScale);
175-
fShapeMean = model.GetTensorShape(fNMean);
176-
fShapeVar = model.GetTensorShape(fNVar);
134+
float *original_bias = static_cast<float*>(original_B.get());
135+
float *original_scale = static_cast<float*>(original_S.get());
136+
float *original_mean = static_cast<float*>(original_M.get());
137+
float *original_var = static_cast<float*>(original_V.get());
138+
float *new_bias = new float[n];
139+
float *new_scale = new float[n];
140+
float *new_mean = new float[n];
141+
float *new_var = new float[n];
142+
size_t bs = 0, ch = 0, h = 0, w = 0;
143+
for(ch=0; ch<channels; ch++){
144+
for(h=0; h<height; h++){
145+
for(w=0; w<width; w++){
146+
new_bias[bs*channels*height*width + ch*height*width + h*width + w] = original_bias[ch];
147+
new_scale[bs*channels*height*width + ch*height*width + h*width + w] = original_scale[ch];
148+
new_mean[bs*channels*height*width + ch*height*width + h*width + w] = original_mean[ch];
149+
new_var[bs*channels*height*width + ch*height*width + h*width + w] = original_var[ch];
150+
}
151+
}
152+
}
153+
size_t Batchoffset = channels*height*width;
154+
for(bs = 1; bs<batchSize; bs++){
155+
std::copy(new_bias, new_bias+Batchoffset, new_bias+(bs*Batchoffset));
156+
std::copy(new_scale, new_scale+Batchoffset, new_scale+(bs*Batchoffset));
157+
std::copy(new_mean, new_mean+Batchoffset, new_mean+(bs*Batchoffset));
158+
std::copy(new_var, new_var+Batchoffset, new_var+(bs*Batchoffset));
159+
}
160+
//// new_var =1. / sqrt(input_var + fepsilon)
161+
for(size_t i=0; i<n; i++){
162+
new_var[i] = 1./sqrt(new_var[i] + fepsilon);
163+
}
164+
std::vector<size_t> new_bias_shape = {batchSize,channels,height,width};
165+
std::shared_ptr<void> new_bias_ptr(new_bias, std::default_delete<float[]>());
166+
std::shared_ptr<void> new_scale_ptr(new_scale, std::default_delete<float[]>());
167+
std::shared_ptr<void> new_mean_ptr(new_mean, std::default_delete<float[]>());
168+
std::shared_ptr<void> new_var_ptr(new_var, std::default_delete<float[]>());
169+
model.UpdateInitializedTensor(fNB, model.GetTensorType(fNB), new_bias_shape, new_bias_ptr);
170+
model.UpdateInitializedTensor(fNScale, model.GetTensorType(fNScale), new_bias_shape, new_scale_ptr);
171+
model.UpdateInitializedTensor(fNMean, model.GetTensorType(fNMean), new_bias_shape, new_mean_ptr);
172+
model.UpdateInitializedTensor(fNVar, model.GetTensorType(fNVar), new_bias_shape, new_var_ptr);
173+
fShapeB = model.GetTensorShape(fNB);
174+
fShapeScale = model.GetTensorShape(fNScale);
175+
fShapeMean = model.GetTensorShape(fNMean);
176+
fShapeVar = model.GetTensorShape(fNVar);
177177
}
178-
}
179-
}
178+
}
179+
}
180180

181181

182-
std::string Generate(std::string OpName){
183-
OpName = "op_" + OpName;
184-
if (fShapeX.empty()){
185-
throw std::runtime_error("TMVA SOFIE Batch Normalization called to Generate without being initialized first");
186-
}
182+
std::string Generate(std::string OpName){
183+
OpName = "op_" + OpName;
184+
if (fShapeX.empty()){
185+
throw std::runtime_error("TMVA SOFIE Batch Normalization called to Generate without being initialized first");
186+
}
187187

188-
std::stringstream out;
189-
//// Batch Norm op
188+
std::stringstream out;
189+
//// Batch Norm op
190190
size_t batchSize = fShapeX[0];
191191
size_t channels = fShapeX[1];
192192
size_t height = (fShapeX.size() > 2) ? fShapeX[2] : 1;
193193
size_t width = (fShapeX.size() > 3) ? fShapeX[3] : 1;
194194
size_t n = batchSize * channels * height * width;
195195

196-
//// copy X into Y
196+
//// copy X into Y
197197
out << SP << "constexpr int " << OpName << "_N =" << batchSize * channels * height * width << ";\n";
198198
out << SP << "constexpr int "<<OpName<< "_incx = 1;\n";
199-
out << SP << "constexpr int "<<OpName<< "_incy = 1;\n";
199+
out << SP << "constexpr int "<<OpName<< "_incy = 1;\n";
200200
out << SP << "BLAS::scopy_(&" << OpName << "_N, " << "tensor_" << fNX << ", &" << OpName << "_incx," << "tensor_" << fNY << ", &" << OpName << "_incy);\n\n";
201201

202202
//// blas saxpy (Y = -Bmean + Y)
203-
out << SP << "float "<<OpName<< "_alpha = -1;\n";
204-
out << SP << "BLAS::saxpy_(&" << OpName << "_N, &" << OpName << "_alpha, " << "tensor_" << fNMean << ", &" << OpName << "_incx,"
205-
<< "tensor_" << fNY <<", &" << OpName << "_incy);\n\n ";
203+
out << SP << "float "<<OpName<< "_alpha = -1;\n";
204+
out << SP << "BLAS::saxpy_(&" << OpName << "_N, &" << OpName << "_alpha, " << "tensor_" << fNMean << ", &" << OpName << "_incx,"
205+
<< "tensor_" << fNY <<", &" << OpName << "_incy);\n\n ";
206206

207-
//// Y *= scale*var
207+
//// Y *= scale*var
208208
out << SP << "for (size_t i = 0; i < " << n << "; i++) {\n";
209209
out << SP << SP << "tensor_" << fNY << "[i] *= tensor_" << fNScale << "[i] * tensor_" << fNVar << "[i]; \n";
210-
out << SP << "}\n";
211-
212-
//// blas saxpy (Y = Bbias + Y)
213-
out << SP <<OpName<< "_alpha = 1;\n";
210+
out << SP << "}\n";
211+
212+
//// blas saxpy (Y = Bbias + Y)
213+
out << SP <<OpName<< "_alpha = 1;\n";
214214
out << SP << "BLAS::saxpy_(&" << OpName << "_N, &" << OpName << "_alpha, " << "tensor_" << fNB << ", &" << OpName << "_incx, "
215-
<< "tensor_" << fNY << ", &" << OpName << "_incy);\n\n";
215+
<< "tensor_" << fNY << ", &" << OpName << "_incy);\n\n";
216216

217-
return out.str();
218-
}
217+
return out.str();
218+
}
219219

220220
};
221221

0 commit comments

Comments
 (0)