Databases
StackOverflow2013
Postsdbo
POFast Exp calculation: possible to improve accuracy without losing too much performance?
 

Note that there are some explanatory texts on larger screens.

plurals
  1. POFast Exp calculation: possible to improve accuracy without losing too much performance?
    primarykey
    data
    text
    <p>I am trying out the fast Exp(x) function that previously was described in <a href="https://stackoverflow.com/a/412988/650012">this</a> answer to an SO question on improving calculation speed in C#:</p> <pre><code>public static double Exp(double x) { var tmp = (long)(1512775 * x + 1072632447); return BitConverter.Int64BitsToDouble(tmp &lt;&lt; 32); } </code></pre> <p>The expression is using some IEEE floating point "tricks" and is primarily intended for use in neural sets. The function is approximately 5 times faster than the regular <code>Math.Exp(x)</code> function. </p> <p>Unfortunately, the numeric accuracy is only -4% -- +2% relative to the regular <code>Math.Exp(x)</code> function, ideally I would like to have accuracy within at least the sub-percent range. </p> <p>I have plotted the quotient between the approximate and the regular Exp functions, and as can be seen in the graph the relative difference appears to be repeated with practically constant frequency.</p> <p><img src="https://i.stack.imgur.com/gUgSI.png" alt="Quotient between fast and regular exp function"></p> <p>Is it possible to take advantage of this regularity to improve the accuracy of the "fast exp" function further without substantially reducing the calculation speed, or would the computational overhead of an accuracy improvement outweigh the computational gain of the original expression?</p> <p>(As a side note, I have also tried <a href="https://stackoverflow.com/a/412102/650012">one of the alternative</a> approaches proposed in the same SO question, but this approach does not seem to be computationally efficient in C#, at least not for the general case.)</p> <p><strong>UPDATE MAY 14</strong></p> <p>Upon request from @Adriano, I have now performed a very simple benchmark. I have performed 10 million computations using each of the alternative <em>exp</em> functions for floating point values in the range [-100, 100]. Since the range of values I am interested in spans from -20 to 0 I have also explicitly listed the function value at x = -5. Here are the results:</p> <pre><code> Math.Exp: 62.525 ms, exp(-5) = 0.00673794699908547 Empty function: 13.769 ms ExpNeural: 14.867 ms, exp(-5) = 0.00675211846828461 ExpSeries8: 15.121 ms, exp(-5) = 0.00641270968867667 ExpSeries16: 32.046 ms, exp(-5) = 0.00673666189488182 exp1: 15.062 ms, exp(-5) = -12.3333325982094 exp2: 15.090 ms, exp(-5) = 13.708332516253 exp3: 16.251 ms, exp(-5) = -12.3333325982094 exp4: 17.924 ms, exp(-5) = 728.368055056781 exp5: 20.972 ms, exp(-5) = -6.13293614238501 exp6: 24.212 ms, exp(-5) = 3.55518353166184 exp7: 29.092 ms, exp(-5) = -1.8271053775984 exp7 +/-: 38.482 ms, exp(-5) = 0.00695945286970704 </code></pre> <p><em>ExpNeural</em> is equivalent to the <em>Exp</em> function specified in the beginning of this text. <em>ExpSeries8</em> is the <a href="https://stackoverflow.com/a/412102/650012">formulation</a> that I originally claimed was not very efficient on .NET; when implementing it exactly like Neil it was actually very fast. <em>ExpSeries16</em> is the analogous formula but with 16 multiplications instead of 8. <em>exp1</em> through <em>exp7</em> are the different functions from Adriano's answer below. The final variant of <em>exp7</em> is a variant where the sign of <em>x</em> is checked; if negative the function returns <code>1/exp(-x)</code> instead.</p> <p>Unfortunately, neither of the <em>expN</em> functions listed by <a href="https://stackoverflow.com/a/10552567/650012">Adriano</a> are sufficient in the broader negative value range I am considering. The series expansion approach by <a href="https://stackoverflow.com/a/412102/650012">Neil Coffey</a> seems to be more suitable in "my" value range, although it is too rapidly diverging with larger negative <em>x</em>, especially when using "only" 8 multiplications.</p>
    singulars
    1. PO
      singulars
      1. PTAnswer
    1. This table or related slice is empty.
    1. PTQuestion
    1. USCommunity
    1. USAnders Gustafsson
    plurals
    1. PL
      singulars
      1. LTLinked
    1. PL
      singulars
      1. LTLinked
    2. PL
      singulars
      1. LTLinked
    1. This table or related slice is empty.
    1. PO
      singulars
      1. PTAnswer
    2. PO
      singulars
      1. PTAnswer
    3. PO
      singulars
      1. PTAnswer
    1. VO
      singulars
      1. POFast Exp calculation: possible to improve accuracy without losing too much performance?
      1. This table or related slice is empty.
      1. VTUpMod
    2. VO
      singulars
      1. POFast Exp calculation: possible to improve accuracy without losing too much performance?
      1. This table or related slice is empty.
      1. VTUpMod
    3. VO
      singulars
      1. POFast Exp calculation: possible to improve accuracy without losing too much performance?
      1. USdasblinkenlight
      1. VTFavorite
    1. This table or related slice is empty.
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload