編寫高效率的IDL程序

1、程序的效率問題，在大數(shù)據(jù)或復雜運算的時候是不能忽略的。但在IDL程序的編寫方式上，不能按照常規(guī)的循環(huán) for依次處理方式寫，簡單歸納下，提高效率的運行的寫法注意下 面兩種方式。1、盡量避免或少用循環(huán)2、多用 where 和 Histogram；說起來很容易，但實際寫的時候一定要多斟酌斟酌。 舉例1:對2000*2000的數(shù)組中大于100的值進行累加。 PRO TEST_TIME ;a = DIST ( 2000 , 2000 ) sum =0.sum1 =0.start = SYSTIME ( 1) FOR i= 0L , N_ELEMENTS (a)- 1L DO BEGIN IF (ai

2、GT 100.0 ) THEN BEGIN sum = sum +aiENDIF ENDFOR fortime = SYSTIME ( 1 )-start PRINT , for time: fortime i= 0L start = SYSTIME ( 1) WHILE i LT N_ELEMENTS (a)- 1L DO BEGINIF (ai GT 100.0 ) THEN sum = sum +ai i+ ENDWHILE whiletime = SYSTIME ( 1 )-start PRINT , while time: ,whiletime start = SYSTIME ( 1

3、) sum = TOTAL (a * (a GT 100.0 ) funtime = SYSTIME ( 1 )-start PRINT , function time:,funtime;倍數(shù) print , result for VS function:, fortime/funtimeprint , result while VS function:,whiletime/funtimeEND 運行后的輸出 IDL test_time % Compiled module: TEST_TIME. for time: 1.4510000 while time: 2.0430000 functio

4、n time: 0.0 result for VS function: 26.870301 result while VS function: 37.833235看到差別了吧，循環(huán)比函數(shù)直接運算慢至少一個數(shù)量級！舉例2：對一個圖像中的特定值，若存在，則以該像素為中心，特定半徑內的元素統(tǒng)一修改 為某值。以IDL自帶的一個圖像為例，將數(shù)據(jù)值等于142的賦為0.源碼如下：:；;計算兩個點的距離；function CalDistance , pointl, point2compile_optidl2Return , SQRT(point1end；0-point20)A 2+(point1 1 -poi

5、nt21)A 2)搜索當前坐標周圍 Distance 內的下標，注意輸入 x和y方向的坐標范圍xRange和yRangefunctioncalldxInDistance,curLoc, distance;,xRange,yRange;初始化臨時下標suitLoc = 0, 0；循環(huán)一次，計算矩形范圍內的符合要求下標for xLoc = curLoc 0-distance, curLoc0+distance do beginfor yLoc = curLoc1 -distance, curLoc1 +distancedo beginif calDistance (curLoc, xLoc,yLo

6、c)LE distance then suitLoc =suitLoc,xLoc,yLoc endfor end ； return , suitLoc*, 1:( N_Elements (suitLoc)/ 2-1) end ; ； ;測試即調用主函數(shù) pro test_process ;原數(shù)據(jù) file = FILEPATH ( rbcells.jpg , $ SUBDIRECTORY = examples , data) READ_JPEG , file, data ； if size (data,/n_dimensions)ne 2 then returndims =size (data

7、,/dimension)startTime =systime(1)；當值等于142時，半徑5內的元素賦值為0eqValue =142repValue = i;符合要求的坐標suitIdx =calldxInDistance(0, 0, 5)；新數(shù)據(jù)nData = dataidxs =where (data eq eqValue,count)for curldx =0 ,count- 1 do beginnSuit = suitIdxsuitLoc =ARRAY_INDICES (data, idxscurIdx)nSuit0,*= nSuit0 ,*+suitLoc0nSuit1 ,*= nS

8、uit;下標要在數(shù)組自身范圍內1 ,*+suitLoc1nSuit0 ,* = dims0 nSuit1 ,* = dims;符合要求的位置賦值1 000,*,nSuit1 ,* = repValue;轉換為二維坐標nDatanSuit endfor；輸出花費時間 print , systime;常規(guī)的循環(huán)寫法(1) - startTimens = dims nl = dims01new = datanew1 = datastartTime =for i= 0 ,ns-systime ( 1)1 do beginfor j= 0 ,nl- if newi,j1 doeqfor m= 0,ns- 1begineqValue then begin do beginfor n= 0,nl-if (m-i*endfor1 do begin1L2+(n-j*1L)A 2 le25then new1m,n= 0endforendifendforendfor1print , common time: window , 1 ,xSize = dims tvscl ,Data, 0,systime (1) - startTime 0* 2,ySize = dims0* 2,ySize = dims1tvscl ,ndata, 1window , 2 ,xSi