topic Re: Condition error in Discussions

Condition error

AdamChoen — Fri, 09 Jun 2023 18:14:20 GMT

hello

I'm trying to do a delta check between neighboring cells value on the same column.

For( i = 1, i <= N Row( dtB ), i++,
    If(
        i == 1,
            If( Abs( :Delta[i] - :Delta[i + 1] ) <= 0.1,
                dtB:Check[i] = "Pass",
                dtB:Check[i] = "Fail"
            ),
        i == N Row( dtB ),
            If( Abs( :Delta[i] - :Delta[i - 1] ) <= 0.1,
                dtB:Check[i] = "Pass",
                dtB:Check[i] = "Fail"
            ),
        If( Abs( :Delta[i] - :Delta[i + 1] ) <= 0.1 | Abs( :Delta[i] - :Delta[i - 1] ) <= 0.1,
            dtB:Check[i] = "Pass",
            dtB:Check[i] = "Fail"
        )
    )
);

i get this result

but line 5 should not fail
when I check it thru the log it seems like a bug or I'm missing something
appreciate your inputs

i
/*:

5
//:*/
:Delta[i]
/*:

0.3
//:*/
:Delta[i+1]
/*:

0.4
//:*/
abs(:Delta[i] - :Delta[i+1])
/*:

0.1
//:*/
abs(:Delta[i] - :Delta[i+1]) <= 0.1
/*:

0

Re: Condition error

Mark_Bailey — Fri, 25 Mar 2022 13:50:15 GMT

If I understand what you are trying to do, then this simplified version might work:

For Each Row(
	dtB:Check = If( Abs( :Delta - Lag(:Delta) ) <= 0.1, "Pass", "Fail" );
);

Re: Condition error

txnelson — Fri, 25 Mar 2022 14:21:10 GMT

abs(-0.4 - 0.3 ) = 0.7

abs(-0.4) - abs(0.3) = 0.1

Re: Condition error

AdamChoen — Sat, 26 Mar 2022 20:25:20 GMT

I like to check the cell value matching (up to 0.1) from both "sides" n+1 and n-1 in this column. 1st and last rows on one side.
thank you

Re: Condition error

AdamChoen — Sat, 26 Mar 2022 20:37:49 GMT

sorry, didn't get it.

i = 5;
:Delta[i] = 0.3;
:Delta[i+1] = 0.4;
abs(:Delta[i] - :Delta[i+1]) = 0.1;
// but 
abs(:Delta[i] - :Delta[i+1]) <= 0.1 // give me 0

why is that? also == give me 0, >= give 1.

Re: Condition error

Mark_Bailey — Mon, 28 Mar 2022 12:48:46 GMT

I confirm this error with the follow code (view results in Log):

/* Abs( 0.4 - 0.3) == 0.1; returns 0 or false */

y = 0.1; x1 = 0.4; x2 = 0.3; diff = x1 - x2;
Show( y, x1, x2, diff, diff == 0.1 );

Re: Condition error

jthi — Mon, 28 Mar 2022 13:10:32 GMT

One more thing to check would be the difference between diff and 0.1

diff - 0.1 = 2.77555756156289e-17;

so floating number accuracy "problem"?

Round(diff, 1) == Round(0.1, 1) = 1;

Re: Condition error

pmroz — Mon, 28 Mar 2022 13:12:24 GMT

That's your basic floating point roundoff error. Add an epsilon and it works:

Deltai = 0.3;
Deltai1 = 0.4;
epsilon = .000000001;
print(abs(Deltai - Deltai1));	// = 0.1;
print(abs(Deltai - Deltai1) <= 0.1);
print(abs(Deltai - Deltai1) <= 0.1 + epsilon);

From the log:

0.1
0
1

Re: Condition error

Mark_Bailey — Mon, 28 Mar 2022 13:58:38 GMT

JMP Development confirmed that this is simply a finite precision numerical problem. Users are disturbed by floating point numbers that do not follow the math (e.g., subtraction). JMP tries to present the result as it is expected, but not the actual result in cases like yours (i.e., 0.4 - 0.3). They suggested also using this approach if you want a better view of the actual result:

Format( 0.1, "Fixed Dec", 20, 22 );
Format( 0.4-0.3, "Fixed Dec", 20, 22 );

Re: Condition error

txnelson — Mon, 28 Mar 2022 14:35:04 GMT

The issue is that the difference from minus .4 to a positive .3 is .7. The Absolute of the values are taken after the calculation!

Re: Condition error

AdamChoen — Tue, 29 Mar 2022 08:23:55 GMT

it's not an issue for me, i like to do that

Re: Condition error

Craige_Hales — Tue, 29 Mar 2022 13:56:17 GMT

0.1 is a repeating fraction in binary. In IEEE floating point, it is rounded at 52 or 53 bits, so it must be either a bit too big or a bit too small.

Good binary fractions are sums of 1/2, 1/4, 1/8, 1/16, 1/32, ... 1/2^53 which are the individual bits of the fraction.

Sometimes the "bit too big" or "bit too small" works out the way you want, and sometimes not.

The second example shows an epsilon chosen for numbers near 1, which have about 14 places after the decimal (thus 1e-12).

If( .3 - .2 == .1, // "no"
    "yes"
, // else
    "no"
);


If( Abs( (.3 - .2) - .1 ) < 1e-12, // "yes"
    "yes"
, // else
    "no"
);

Old post