You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
398 lines
10 KiB
C
398 lines
10 KiB
C
/* ----------------------------------------------------------------------
|
|
* Project: CMSIS DSP Library
|
|
* Title: arm_naive_gaussian_bayes_predict_f32
|
|
* Description: Naive Gaussian Bayesian Estimator
|
|
*
|
|
*
|
|
* Target Processor: Cortex-M and Cortex-A cores
|
|
* -------------------------------------------------------------------- */
|
|
/*
|
|
* Copyright (C) 2010-2019 ARM Limited or its affiliates. All rights reserved.
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*
|
|
* Licensed under the Apache License, Version 2.0 (the License); you may
|
|
* not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an AS IS BASIS, WITHOUT
|
|
* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
*/
|
|
|
|
#include "arm_math.h"
|
|
#include <limits.h>
|
|
#include <math.h>
|
|
|
|
#define PI_F 3.1415926535897932384626433832795f
|
|
#define DPI_F (2.0f*3.1415926535897932384626433832795f)
|
|
|
|
/**
|
|
* @addtogroup groupBayes
|
|
* @{
|
|
*/
|
|
|
|
/**
|
|
* @brief Naive Gaussian Bayesian Estimator
|
|
*
|
|
* @param[in] *S points to a naive bayes instance structure
|
|
* @param[in] *in points to the elements of the input vector.
|
|
* @param[in] *pBuffer points to a buffer of length numberOfClasses
|
|
* @return The predicted class
|
|
*
|
|
* @par If the number of classes is big, MVE version will consume lot of
|
|
* stack since the log prior are computed on the stack.
|
|
*
|
|
*/
|
|
|
|
#if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
|
|
|
|
#include "arm_helium_utils.h"
|
|
#include "arm_vec_math.h"
|
|
|
|
uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_instance_f32 *S,
|
|
const float32_t * in,
|
|
float32_t *pBuffer)
|
|
{
|
|
uint32_t nbClass;
|
|
const float32_t *pTheta = S->theta;
|
|
const float32_t *pSigma = S->sigma;
|
|
float32_t *buffer = pBuffer;
|
|
const float32_t *pIn = in;
|
|
float32_t result;
|
|
f32x4_t vsigma;
|
|
float32_t tmp;
|
|
f32x4_t vacc1, vacc2;
|
|
uint32_t index;
|
|
float32_t logclassPriors[S->numberOfClasses];
|
|
float32_t *pLogPrior = logclassPriors;
|
|
|
|
arm_vlog_f32((float32_t *) S->classPriors, logclassPriors, S->numberOfClasses);
|
|
|
|
pTheta = S->theta;
|
|
pSigma = S->sigma;
|
|
|
|
for (nbClass = 0; nbClass < S->numberOfClasses; nbClass++) {
|
|
pIn = in;
|
|
|
|
vacc1 = vdupq_n_f32(0);
|
|
vacc2 = vdupq_n_f32(0);
|
|
|
|
uint32_t blkCnt =S->vectorDimension >> 2;
|
|
while (blkCnt > 0U) {
|
|
f32x4_t vinvSigma, vtmp;
|
|
|
|
vsigma = vaddq_n_f32(vld1q(pSigma), S->epsilon);
|
|
vacc1 = vaddq(vacc1, vlogq_f32(vmulq_n_f32(vsigma, 2.0f * PI)));
|
|
|
|
vinvSigma = vrecip_medprec_f32(vsigma);
|
|
|
|
vtmp = vsubq(vld1q(pIn), vld1q(pTheta));
|
|
/* squaring */
|
|
vtmp = vmulq(vtmp, vtmp);
|
|
|
|
vacc2 = vfmaq(vacc2, vtmp, vinvSigma);
|
|
|
|
pIn += 4;
|
|
pTheta += 4;
|
|
pSigma += 4;
|
|
blkCnt--;
|
|
}
|
|
|
|
blkCnt = S->vectorDimension & 3;
|
|
if (blkCnt > 0U) {
|
|
mve_pred16_t p0 = vctp32q(blkCnt);
|
|
f32x4_t vinvSigma, vtmp;
|
|
|
|
vsigma = vaddq_n_f32(vld1q(pSigma), S->epsilon);
|
|
vacc1 =
|
|
vaddq_m_f32(vacc1, vacc1, vlogq_f32(vmulq_n_f32(vsigma, 2.0f * PI)), p0);
|
|
|
|
vinvSigma = vrecip_medprec_f32(vsigma);
|
|
|
|
vtmp = vsubq(vld1q(pIn), vld1q(pTheta));
|
|
/* squaring */
|
|
vtmp = vmulq(vtmp, vtmp);
|
|
|
|
vacc2 = vfmaq_m_f32(vacc2, vtmp, vinvSigma, p0);
|
|
|
|
pTheta += blkCnt;
|
|
pSigma += blkCnt;
|
|
}
|
|
|
|
tmp = -0.5f * vecAddAcrossF32Mve(vacc1);
|
|
tmp -= 0.5f * vecAddAcrossF32Mve(vacc2);
|
|
|
|
*buffer = tmp + *pLogPrior++;
|
|
buffer++;
|
|
}
|
|
|
|
arm_max_f32(pBuffer, S->numberOfClasses, &result, &index);
|
|
|
|
return (index);
|
|
}
|
|
|
|
#else
|
|
|
|
#if defined(ARM_MATH_NEON)
|
|
|
|
#include "NEMath.h"
|
|
|
|
|
|
|
|
uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_instance_f32 *S,
|
|
const float32_t * in,
|
|
float32_t *pBuffer)
|
|
{
|
|
|
|
const float32_t *pPrior = S->classPriors;
|
|
|
|
const float32_t *pTheta = S->theta;
|
|
const float32_t *pSigma = S->sigma;
|
|
|
|
const float32_t *pTheta1 = S->theta + S->vectorDimension;
|
|
const float32_t *pSigma1 = S->sigma + S->vectorDimension;
|
|
|
|
float32_t *buffer = pBuffer;
|
|
const float32_t *pIn=in;
|
|
|
|
float32_t result;
|
|
float32_t sigma,sigma1;
|
|
float32_t tmp,tmp1;
|
|
uint32_t index;
|
|
uint32_t vecBlkCnt;
|
|
uint32_t classBlkCnt;
|
|
float32x4_t epsilonV;
|
|
float32x4_t sigmaV,sigmaV1;
|
|
float32x4_t tmpV,tmpVb,tmpV1;
|
|
float32x2_t tmpV2;
|
|
float32x4_t thetaV,thetaV1;
|
|
float32x4_t inV;
|
|
|
|
epsilonV = vdupq_n_f32(S->epsilon);
|
|
|
|
classBlkCnt = S->numberOfClasses >> 1;
|
|
while(classBlkCnt > 0)
|
|
{
|
|
|
|
|
|
pIn = in;
|
|
|
|
tmp = logf(*pPrior++);
|
|
tmp1 = logf(*pPrior++);
|
|
tmpV = vdupq_n_f32(0.0f);
|
|
tmpV1 = vdupq_n_f32(0.0f);
|
|
|
|
vecBlkCnt = S->vectorDimension >> 2;
|
|
while(vecBlkCnt > 0)
|
|
{
|
|
sigmaV = vld1q_f32(pSigma);
|
|
thetaV = vld1q_f32(pTheta);
|
|
|
|
sigmaV1 = vld1q_f32(pSigma1);
|
|
thetaV1 = vld1q_f32(pTheta1);
|
|
|
|
inV = vld1q_f32(pIn);
|
|
|
|
sigmaV = vaddq_f32(sigmaV, epsilonV);
|
|
sigmaV1 = vaddq_f32(sigmaV1, epsilonV);
|
|
|
|
tmpVb = vmulq_n_f32(sigmaV,DPI_F);
|
|
tmpVb = vlogq_f32(tmpVb);
|
|
tmpV = vmlsq_n_f32(tmpV,tmpVb,0.5f);
|
|
|
|
tmpVb = vmulq_n_f32(sigmaV1,DPI_F);
|
|
tmpVb = vlogq_f32(tmpVb);
|
|
tmpV1 = vmlsq_n_f32(tmpV1,tmpVb,0.5f);
|
|
|
|
tmpVb = vsubq_f32(inV,thetaV);
|
|
tmpVb = vmulq_f32(tmpVb,tmpVb);
|
|
tmpVb = vmulq_f32(tmpVb, vinvq_f32(sigmaV));
|
|
tmpV = vmlsq_n_f32(tmpV,tmpVb,0.5f);
|
|
|
|
tmpVb = vsubq_f32(inV,thetaV1);
|
|
tmpVb = vmulq_f32(tmpVb,tmpVb);
|
|
tmpVb = vmulq_f32(tmpVb, vinvq_f32(sigmaV1));
|
|
tmpV1 = vmlsq_n_f32(tmpV1,tmpVb,0.5f);
|
|
|
|
pIn += 4;
|
|
pTheta += 4;
|
|
pSigma += 4;
|
|
pTheta1 += 4;
|
|
pSigma1 += 4;
|
|
|
|
vecBlkCnt--;
|
|
}
|
|
tmpV2 = vpadd_f32(vget_low_f32(tmpV),vget_high_f32(tmpV));
|
|
tmp += tmpV2[0] + tmpV2[1];
|
|
|
|
tmpV2 = vpadd_f32(vget_low_f32(tmpV1),vget_high_f32(tmpV1));
|
|
tmp1 += tmpV2[0] + tmpV2[1];
|
|
|
|
vecBlkCnt = S->vectorDimension & 3;
|
|
while(vecBlkCnt > 0)
|
|
{
|
|
sigma = *pSigma + S->epsilon;
|
|
sigma1 = *pSigma1 + S->epsilon;
|
|
|
|
tmp -= 0.5f*logf(2.0f * PI_F * sigma);
|
|
tmp -= 0.5f*(*pIn - *pTheta) * (*pIn - *pTheta) / sigma;
|
|
|
|
tmp1 -= 0.5f*logf(2.0f * PI_F * sigma1);
|
|
tmp1 -= 0.5f*(*pIn - *pTheta1) * (*pIn - *pTheta1) / sigma1;
|
|
|
|
pIn++;
|
|
pTheta++;
|
|
pSigma++;
|
|
pTheta1++;
|
|
pSigma1++;
|
|
vecBlkCnt--;
|
|
}
|
|
|
|
*buffer++ = tmp;
|
|
*buffer++ = tmp1;
|
|
|
|
pSigma += S->vectorDimension;
|
|
pTheta += S->vectorDimension;
|
|
pSigma1 += S->vectorDimension;
|
|
pTheta1 += S->vectorDimension;
|
|
|
|
classBlkCnt--;
|
|
}
|
|
|
|
classBlkCnt = S->numberOfClasses & 1;
|
|
|
|
while(classBlkCnt > 0)
|
|
{
|
|
|
|
|
|
pIn = in;
|
|
|
|
tmp = logf(*pPrior++);
|
|
tmpV = vdupq_n_f32(0.0f);
|
|
|
|
vecBlkCnt = S->vectorDimension >> 2;
|
|
while(vecBlkCnt > 0)
|
|
{
|
|
sigmaV = vld1q_f32(pSigma);
|
|
thetaV = vld1q_f32(pTheta);
|
|
inV = vld1q_f32(pIn);
|
|
|
|
sigmaV = vaddq_f32(sigmaV, epsilonV);
|
|
|
|
tmpVb = vmulq_n_f32(sigmaV,DPI_F);
|
|
tmpVb = vlogq_f32(tmpVb);
|
|
tmpV = vmlsq_n_f32(tmpV,tmpVb,0.5f);
|
|
|
|
tmpVb = vsubq_f32(inV,thetaV);
|
|
tmpVb = vmulq_f32(tmpVb,tmpVb);
|
|
tmpVb = vmulq_f32(tmpVb, vinvq_f32(sigmaV));
|
|
tmpV = vmlsq_n_f32(tmpV,tmpVb,0.5f);
|
|
|
|
pIn += 4;
|
|
pTheta += 4;
|
|
pSigma += 4;
|
|
|
|
vecBlkCnt--;
|
|
}
|
|
tmpV2 = vpadd_f32(vget_low_f32(tmpV),vget_high_f32(tmpV));
|
|
tmp += tmpV2[0] + tmpV2[1];
|
|
|
|
vecBlkCnt = S->vectorDimension & 3;
|
|
while(vecBlkCnt > 0)
|
|
{
|
|
sigma = *pSigma + S->epsilon;
|
|
tmp -= 0.5f*logf(2.0f * PI_F * sigma);
|
|
tmp -= 0.5f*(*pIn - *pTheta) * (*pIn - *pTheta) / sigma;
|
|
|
|
pIn++;
|
|
pTheta++;
|
|
pSigma++;
|
|
vecBlkCnt--;
|
|
}
|
|
|
|
*buffer++ = tmp;
|
|
|
|
classBlkCnt--;
|
|
}
|
|
|
|
arm_max_f32(pBuffer,S->numberOfClasses,&result,&index);
|
|
|
|
return(index);
|
|
}
|
|
|
|
#else
|
|
|
|
/**
|
|
* @brief Naive Gaussian Bayesian Estimator
|
|
*
|
|
* @param[in] *S points to a naive bayes instance structure
|
|
* @param[in] *in points to the elements of the input vector.
|
|
* @param[in] *pBuffer points to a buffer of length numberOfClasses
|
|
* @return The predicted class
|
|
*
|
|
*/
|
|
uint32_t arm_gaussian_naive_bayes_predict_f32(const arm_gaussian_naive_bayes_instance_f32 *S,
|
|
const float32_t * in,
|
|
float32_t *pBuffer)
|
|
{
|
|
uint32_t nbClass;
|
|
uint32_t nbDim;
|
|
const float32_t *pPrior = S->classPriors;
|
|
const float32_t *pTheta = S->theta;
|
|
const float32_t *pSigma = S->sigma;
|
|
float32_t *buffer = pBuffer;
|
|
const float32_t *pIn=in;
|
|
float32_t result;
|
|
float32_t sigma;
|
|
float32_t tmp;
|
|
float32_t acc1,acc2;
|
|
uint32_t index;
|
|
|
|
pTheta=S->theta;
|
|
pSigma=S->sigma;
|
|
|
|
for(nbClass = 0; nbClass < S->numberOfClasses; nbClass++)
|
|
{
|
|
|
|
|
|
pIn = in;
|
|
|
|
tmp = 0.0;
|
|
acc1 = 0.0f;
|
|
acc2 = 0.0f;
|
|
for(nbDim = 0; nbDim < S->vectorDimension; nbDim++)
|
|
{
|
|
sigma = *pSigma + S->epsilon;
|
|
acc1 += logf(2.0f * PI_F * sigma);
|
|
acc2 += (*pIn - *pTheta) * (*pIn - *pTheta) / sigma;
|
|
|
|
pIn++;
|
|
pTheta++;
|
|
pSigma++;
|
|
}
|
|
|
|
tmp = -0.5f * acc1;
|
|
tmp -= 0.5f * acc2;
|
|
|
|
|
|
*buffer = tmp + logf(*pPrior++);
|
|
buffer++;
|
|
}
|
|
|
|
arm_max_f32(pBuffer,S->numberOfClasses,&result,&index);
|
|
|
|
return(index);
|
|
}
|
|
|
|
#endif
|
|
#endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
|
|
|
|
/**
|
|
* @} end of groupBayes group
|
|
*/
|