Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: The accuracy of the float data type


From   Nick Cox <[email protected]>
To   "[email protected]" <[email protected]>
Subject   Re: st: The accuracy of the float data type
Date   Sat, 25 Jan 2014 00:41:39 +0000

OK then;

bysort firmID year : egen double maxsale=max(sales)

is, we bet, your solution.

Nick
[email protected]


On 24 January 2014 20:13, R Zhang <[email protected]> wrote:

> sales was double %12.0g, maxsale was float %9.0g. My apology.

On Fri, Jan 24, 2014 at 1:09 PM, Nick Cox <[email protected]> wrote:

>> I wondered that too, but Rochelle said that both variables were
>> -float-. But if that is not so, then it's likely to be the
>> explanation.
>>
>> Note by the way that Stata does not use terminology such as "storage
>> format". Display format and variable type are, as Nick Winter implies,
>> quite different notions.

On 24 January 2014 17:32, Nick Winter <[email protected]> wrote:

>>> Perhaps the problem comes because the *storage* format of sales and maxsale
>>> are different.  (This is not the same as the *display* format).
>>>
>>> Consider:
>>>
>>> clear
>>> set seed 1234567
>>> set obs 10
>>> gen double sales = round(uniform()*100,.001)
>>> gen year = _n
>>> egen float maxsale = max(sales), by(year)
>>> gen equal = sales == maxsale
>>>
>>> egen double maxsale2 = max(sales), by(year)
>>> gen equal2 = sales == maxsale2
>>>
>>> gen equal3 = float(sales) == maxsale
>>>
>>> list
>>>
>>>
>>>      +--------------------------------------------------------------+
>>>      |  sales   year   maxsale   equal   maxsale2   equal2   equal3 |
>>>      |--------------------------------------------------------------|
>>>   1. |   2.65      1      2.65       0       2.65        1        1 |
>>>   2. | 17.274      2    17.274       0     17.274        1        1 |
>>>   3. |  2.923      3     2.923       0      2.923        1        1 |
>>>   4. | 75.377      4    75.377       0     75.377        1        1 |
>>>   5. | 65.559      5    65.559       0     65.559        1        1 |
>>>      |--------------------------------------------------------------|
>>>   6. | 81.163      6    81.163       0     81.163        1        1 |
>>>   7. | 17.459      7    17.459       0     17.459        1        1 |
>>>   8. | 24.531      8    24.531       0     24.531        1        1 |
>>>   9. | 11.195      9    11.195       0     11.195        1        1 |
>>>  10. | 75.953     10    75.953       0     75.953        1        1 |
>>>      +--------------------------------------------------------------+
>>>
>>>
>>> If that's the case, then you need to assure that your sales and maxsale
>>> variables are in the same storage precision (float, double); OR you need to
>>> explicitly round the one that is double-precision to float precision when
>>> you make the comparison, using the float() function.
>>>
>>> See -help precision- for more on what's going on here.

On 1/24/2014 11:55 AM, R Zhang wrote:

>>>> Thanks to you both, Sergiy and Nick .
>>>>
>>>> Nick,
>>>>
>>>> 1.are you saying that I should follow Sergiy's advice to change
>>>> format? If so, given the large number of observations I have , how do
>>>> I automate the process?
>>>>
>>>> 2. if I do not change the format, I listed some observations below to
>>>> show you that sales and maxsale look the same, however, when I use" l
>>>> if sales == maxsale" it does not list all of the observations that
>>>> appear equal.
>>>>
>>>>
>>>> *****************
>>>>     +--------------------+
>>>>       |   sales   maxsale1 |
>>>>       |--------------------|
>>>>    1. |  25.395     25.395 |
>>>>    2. |  32.007     32.007 |
>>>>    3. |  53.798     53.798 |
>>>>    4. |  12.748     12.748 |
>>>>    5. |  13.793     13.793 |
>>>>   ..... omitted to save space
>>>>
>>>>   31. | 166.181    166.181 |
>>>>   32. |  21.927    166.181 |
>>>>   33. |  26.328    189.897 |
>>>>   34. |  31.787    189.897 |
>>>>   35. | 189.897    189.897 |
>>>>       |--------------------|
>>>>   36. | 264.582    264.582 |
>>>>   37. |   33.61    264.582 |
>>>>   38. | 312.227    312.227 |
>>>>   39. |  35.413    312.227 |
>>>>   40. |  406.36     406.36 |
>>>>       |--------------------|
>>>>   41. | 444.875    444.875 |
>>>>
>>>>
>>>>   egen maxsale=max(sales), by (gvkey year)
>>>>
>>>>   l if sales == maxsale,
>>>>
>>>> the first observation that is listed is  444.875    444.875 ,
>>>>
>>>> why is that?
>>>>
>>>> thanks!
>>>>
>>>> On Fri, Jan 24, 2014 at 11:34 AM, Nick Cox <[email protected]> wrote:
>>>>>
>>>>> This is very good advice in general, but in this case the maxima are
>>>>> selected from the original values, so that equality is to be expected
>>>>> for some observations.
>>>>> Nick
>>>>> [email protected]
>>>>>
>>>>>
>>>>> On 24 January 2014 16:31, Sergiy Radyakin <[email protected]> wrote:
>>>>>>
>>>>>> Zhang, avoid comparing floating point numbers for equality. Instead
>>>>>> there is a system variable c(epsfloat) , which you can refer to when
>>>>>> you need to deal with precision:
>>>>>>
>>>>>> clear
>>>>>> input float sales
>>>>>> 25.395
>>>>>> 32.007
>>>>>> end
>>>>>>
>>>>>> list
>>>>>>
>>>>>> display c(epsfloat)
>>>>>>
>>>>>> list if sales==25.395
>>>>>> list if abs(sales-25.395)<=10*c(epsfloat)
>>>>>>
>>>>>> list if sales==32.007
>>>>>> list if abs(sales-32.007)<=10*c(epsfloat)
>>>>>>
>>>>>>
>>>>>> Best, Sergiy Radyakin
>>>>>>
>>>>>> On Fri, Jan 24, 2014 at 11:23 AM, Maarten Buis <[email protected]>
>>>>>> wrote:
>>>>>>>
>>>>>>> I would do this differently:
>>>>>>>
>>>>>>> *------------------ begin example ------------------
>>>>>>> // get some example data
>>>>>>> sysuse auto
>>>>>>>
>>>>>>> // create a variable denoting missing values
>>>>>>> gen byte miss = missing(rep78, price)
>>>>>>>
>>>>>>> // create our indicator variable
>>>>>>> bys rep78 miss (price) : gen max = _n == _N if !miss
>>>>>>>
>>>>>>> // admire the result
>>>>>>> list rep78 miss price max in 1/12, sepby(rep78)
>>>>>>> *------------------- end example -------------------
>>>>>>> * (For more on examples I sent to the Statalist see:
>>>>>>> * http://www.maartenbuis.nl/example_faq )
>>>>>>>
>>>>>>> Hope this helps,
>>>>>>> Maarten
>>>>>>>
>>>>>>>
>>>>>>> On Fri, Jan 24, 2014 at 4:53 PM, R Zhang <[email protected]> wrote:
>>>>>>>>
>>>>>>>> Dear Statalist,
>>>>>>>>
>>>>>>>> my data structure is as follows
>>>>>>>>
>>>>>>>> firmID    segmentID   sales year
>>>>>>>> 1001       1               25.395     1990
>>>>>>>> 1001       1                32.007     1991
>>>>>>>>
>>>>>>>> ............
>>>>>>>>
>>>>>>>> a firm can operate in multiple segments as identified by  segmentID .
>>>>>>>> I wanted to identify the largest segment by sales,so I used
>>>>>>>>
>>>>>>>> bysort firmID year : egen maxsale=max(sales)
>>>>>>>>
>>>>>>>> then I did
>>>>>>>> gen PriSIC=0
>>>>>>>> replace PriSIC=1 if sales=maxsale
>>>>>>>>
>>>>>>>> I got
>>>>>>>> firmID    segmentID   sales year                  maxsale    prisic
>>>>>>>> 1001       1               25.395     1990            25.395         0
>>>>>>>> 1001       1                32.007     1991            32.007       0
>>>>>>>>
>>>>>>>> I could not figure out why prisic is 0, so I compute the diffderence
>>>>>>>> (sales-maxsale), it shows a very small negative number , and the data
>>>>>>>> dictionary shows sales format float %12.0g, and maxsale format float
>>>>>>>> %9.0g
>>>>>>>>
>>>>>>>> what should I do to correct this?
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index