November 21, 2015 15:10
THIS IS A BETA VERSION OF THE SOFTWARE. EXPECT BUGS !!!
Backup your data files and entire AmiBroker folder first!
INSTALLATION INSTRUCTIONS
First you need to have full version of AmiBroker 6.00 installed. Then just run the BETA installer and follow the instructions.
Then run AmiBroker. You should see "AmiBroker 6.05.0 BETA" written in the About box.
See CHANGE LOG below for detailed list of changes. Note that only changes
that affect end-user directly are listed here. Internal code changes/refactoring
is usually not mentioned.
CHANGE LOG
CHANGES FOR VERSION 6.05.0 (as compared to 6.04.0)
CHANGES FOR VERSION 6.04.0 (as compared to 6.03.0)
CHANGES FOR VERSION 6.03.0 (as compared to 6.02.0)
Breakpoints currently work with:
a) regular statements (that end with semicolon). For multi-line statements
place breakpoint at the beginning line of the statement
b) for loops
c) while loops
d) do-while loops (you need to place breakpoint where 'while' clause is located,
it won't break at the 'do' line as it essentially is no-op, if you want to
break at the beginning
of do, just place breakpoint on first statement inside { block }
d) if statements
e) return statements
f) switch/case statements
g) break statements
Breakpoints that you place on other lines, won't trigger. The AFL editor won't allow to place breakpoint on empty line, or line that beginnins with // comment or sole brace
x = "test" +
"
second row" +
"
third row";
But if you put two statements in single line like this:
x = 1; y = 2;
Then this line would be treated as two steps (each expression x =1; and y=2; separately).
Keyboard shortcuts:
Step Over F10
Step Into F11
CHANGES FOR VERSION 6.02.0 (as compared to 6.01.0)
m = Matrix( 10, 10, 0 );
// do something with matrix
if( m )
{
// some code
}
is equivalent to
if( NOT IsNull(
m[ 0 ][ 0 ] )
)
{
// some code
}
KeepAll flag when it is set to true emulates the behavior of AddToComposite.
It keeps all values that are already present, so if data holes exists in
current symbol,
the bars that are present in static variable but not present in current symbol
remain untouched.
When KeepAll is set to false then only bars that are present in current symbol
are kept. Any other bars that were present in static variable but not present
in currently
processed symbols are removed. That is what normally happens with StaticVarSet().
In fact when KeepAll is set to False, StaticVarAdd can be seen as the following
pseudo code:
EnterCriticalSection
x = Nz( StaticVarGet( "name" ) ); // read exisiting value (and
convert Nulls to zero)
x += Nz( value ); // add value to existing
StaticVarSet( "name", x ); // store updated value
LeaveCriticalSection
The function can be used to create composites like this:
if( status("stocknum")
== 0 )
{
// remove any earier composite values
StaticVarRemove("~Composite");
}
StaticVarAdd( "~Composite", MACD() > Signal()
);
Buy = 0;
NOTES:
1. StaticVarAdd automatically converts all Nulls to zeros (as AddToComposite
does).
2. If you want to replace AddToComposite with StaticVarAdd, keep in mind
that by default AddToComposite skips symbols in group 253. This is done so
composite symbols
are not added to themselves. If you have composite symbols in your database
and want to skip symbols in group 253 you can use
if( GroupID() != 253 ) StaticVarAdd("~Composite", values );
3. Thanks to extensive code tuning, StaticVarAdd generally offers better
performance than AddToComposite which was already blazing fast. Single threaded
StaticVarAdd may be twice as fast as ATC. With 8 threads running StaticVarAdd
may be 4x as fast (it does not scale as much as naive person may think, because
critical section limits performance due to lock contention). To illustrate
the amount of fine tuning applied it can be said that first 'straightforward'
version of StaticVarAdd was actually 20 times slower than ATC.
4. Be careful when using "quickafl" as StaticVarAdd would not increase
'required bars' (as ATC does), so if you want to actually add all bars and
quick afl is turned on in analysis, it is better to add SetBarsRequired(sbrAll,
sbrAll)
order = Param( "n-th
Order", 10, 1, 16, 1 );
length = 60;
lvb = BarCount - 1;
fvb = lvb - length;
yy = Matrix( length + 1, 1, 0 );
xx = Matrix( length + 1,
order + 1, 1 );
yy = MxSetBlock( yy, 0,
length, 0, 0, Ref( C,
fvb ) );
x = BarIndex() - length/2;
for( j = 1;
j <= order; j++ )
{
xx = MxSetBlock( xx, 0,
length, j, j, x ^ j );
}
xxt = MxTranspose( xx );
aa = MxSolve( xxt @ xx, xxt ) @ yy;
//aa = MxInverse( xxt @ xx ) @ xxt @ yy; // alternative
way
if( aa ) // check
if matrix is not null (so solution exists)
{
rr = Null; // store the fit in rr
for( i = fvb;
i <= lvb; i++ )
{
rr[i] = aa[0][0];
for( j = 1;
j <= order; j++ )
{
rr[i] += aa[j][0] * x[ i - fvb ] ^ j;
}
}
if( IsNan(
rr[ fvb ] ) )
{
// our polynomial yields infinite or not-a-number result
due to overflow/underflow
Title = "Polyfit
failed. The order of polynomial is too High";
}
else
{
SetChartOptions( 0, chartShowDates );
SetBarFillColor( IIf( C > O, ColorRGB( 0, 75, 0 ), IIf( C <= O, ColorRGB( 75, 0, 0 ), colorLightGrey )
) );
Plot( rr, "rr", colorWhite, styleLine | styleThick);
}
}
else
{
Title = "Matrix is singular. The order of polynomial
is too high";
}
Plot( C, "", IIf( C > O, ColorRGB( 0, 255, 0 ), IIf( C <= O, ColorRGB( 255, 0, 0 ), colorLightGrey )
), styleDots | styleNoLine ); CHANGES FOR VERSION 6.01.0 (as compared to 6.00.0)
"slow" method uses Laplace expansion
"
fast" method uses LU decomposition
"
Slow" method for small matrices (1x1, 2x2, 3x3, 4x4) is actually faster
than "fast", equally fast for matrix 5x5 and
slower than "fast" method for matrices larger than 5x5
For this reason "auto" method uses "fast" LU method only for matrices larger than 5x5
LU decomposition is fast but subject to higher numerical errors. "Slow" method
is slower yet produces much more reliable results.
For example Octave/MatLab that use LU decomposition would say that determinant
of singular matrix like this
{ {16, 2, 3, 13}, { 5, 11, 10, 8}, {9, 7, 6, 12}, {4, 14, 15, 1 } }
is -1.4495e-012 due to roundoff errors of LU method.
If you want to calculate determinant using fast (LU decomposition) method, call MxDet with fast parameter set to 2.
CAVEAT: Laplace method has complexity of O(N!) and for this reason, even
if you use method = 1, the maximum dimension for this method is limited
to 10x10.
Matrices larger than that are always calculated using LU method
z = Matrix( 2, 20, 0 );
// first row
z = MxSetBlock( z, 0, 0, 0, 19, Close );
// second row
z = MxSetBlock( z, 1, 1, 0, 19, RSI( 5 ) );
printf("Matrix z\n");
printf( MxToString( z ) );
x = MxGetBlock( z, 0, 1, 0, 19, True );
printf("Items are now in regular array (data series):\n" );
for( i = 0; i < 20; i++ )
printf( NumToStr( x[ i ] ) + "\n" );
z = MxGetBlock( z, 0, 1, 0, 1 ); // retrieve upper 2x2 submatrix
printf("Upper submatrix z\n");
printf( MxToString( z ) );
Note: the function creates new matrix as a result (so source matrix is unaffected unless you do the assignment of the result back to the original variable)
Example 1:
// Create a matrix 6x6
// and fill 4x4 interior (except edges with consecutively increasing numbers)
y = Matrix( 6, 6, 0 );
y = MxSetBlock( y, 1, 4, 1, 4, Cum(1));
printf("Matrix y\n");
printf( MxToString( y ) );
Example 2:
// Create a matrix 2 rows x 20 columns and fill rows 0, 1 with first 20 values
of Close and RSI(5) arrays respectively
z = Matrix( 2, 20, 0 );
// first row
z = MxSetBlock( z, 0, 0, 0, 19, Close );
// second row
z = MxSetBlock( z, 1, 1, 0, 19, RSI( 5 ) );
printf("Matrix z\n");
printf( MxToString( z ) );
B can also be a matrix,with each of its column representing different vector
B. This way single call to MxSolve can solve several systems with same matrix
A but different right hand vectors.
If B is a matrix NxM then MxSolve will produce result also having NxM cells
with each column representing single solution.
Example 1:
A = MxFromString("[ 1, 1, 1, 1; 0, 2, 5, -1; 2, 5, -1, 1; 2, 2, 2,
1 ]");
B = MxFromString("[ 7; -5; 28; 13 ]" ); // single vertical vector
B
printf( "Solving A * X = B\n" );
printf("Matrix A\n");
printf( MxToString( A ) );
printf("\nMatrix B\n");
printf( MxToString( B ) );
X = MxSolve( A, B );
printf("\nSolution X\n");
Example 2:
A = MxFromString("[ 1, 1, 1, 1; 0, 2, 5, -1; 2, 5, -1, 1; 2, 2, 2,
1 ]");
B = MxFromString("[ 7, 14 ; -5, -10; 28, 56; 13, 26 ]" ); // 2
right-hand side vertical vectors
printf( "Solving A * X = B\n" );
printf("Matrix A\n");
printf( MxToString( A ) );
printf("\nMatrix B\n");
printf( MxToString( B ) );
X = MxSolve( A, B );
printf("\nSolutions X\n");
printf( MxToString( X ) ); // two solutions
(Highly) Technical note about numerical precision:
Despite the fact that both MxSolve and MxInverse use double precision arithmetic
solving/inverting matrices is subject to numerical precision of double IEEE
and for example zero result may come up as something like 1.4355e-16 (0.0000000000000001)
due to the fact that double precision is still limited in accuracy (16 digits).
The result of
X = MxInverse( A ) @ B;
although mathematically the same as solving the system of equations, would
yield slightly different result because if you do the inverse the returned
matrix is converted back
to single precision and matrix product is performed with single precision.
When you use MxSolve you are performing all calcs using 64-bit (double) precision
and
only end result is converted back to single precision. So for example polynomial
fit code works better with MxSolve than MxInverse
// Least Squares Polynomial Fit test
order = Param( "n-th Order", 15, 1, 25, 1 );
length = 60;
lvb = BarCount - 1;
fvb = lvb - length;
yy = Matrix( length + 1, 1, 0 );
xx = Matrix( length + 1,
order + 1, 1 );
yy = MxSetBlock( yy, 0,
length, 0, 0, Ref( C,
fvb ) );
x = BarIndex() - length/2;
for( j = 1;
j <= order; j++ )
{
xx = MxSetBlock( xx, 0,
length, j, j, x ^ j );
}
xxt = MxTranspose( xx );
aa = MxSolve( xxt @ xx, xxt ) @ yy;
//aa = MxInverse( xxt @ xx ) @ xxt @ yy; // alternative
way
rr = Null; // store the fit in rr
for( i = fvb; i <= lvb; i++ )
{
rr[i] = aa[0][0];
for( j = 1;
j <= order; j++ )
{
rr[i] += aa[j][0] * x[ i - fvb ] ^ j;
}
}
SetChartOptions( 0, chartShowDates );
SetBarFillColor( IIf( C > O, ColorRGB( 0, 75, 0 ), IIf( C <= O, ColorRGB( 75, 0, 0 ), colorLightGrey )
) );
Plot( rr, "rr", colorWhite, styleLine | styleThick);
Plot( C, "", IIf( C > O, ColorRGB( 0, 255, 0 ), IIf( C <= O, ColorRGB( 255, 0, 0 ), colorLightGrey )
), styleDots | styleNoLine );
// example
m = MxFromString("[ 9, 5, 6; 8, 7, 3 ]");
printf( MxToString( m ) + "\n\n" );
printf("%g, %g\n\n", MxGetSize( m, 0 ), MxGetSize( m, 1 ) );
m2 = MxSort( m, 0 ) ;
printf( MxToString( m2 ) + "\n\n" );
m3 = MxSort( m, 1 ) ;
printf( MxToString( m3 ) + "\n\n" );
Hint: if you want to sort columns instead you can Transpose/Sort rows/Transpose back.
m = MxFromString("[ 9, 1, 6; 40, 30, 20; 8, 7, 3; 3, 5, 1 ]");
printf("Input matrix\n");
printf( MxToString( m ) + "\n\n" );
printf("Rows %g, Cols %g\n\n", MxGetSize( m, 0 ), MxGetSize( m, 1 ) );
printf("Sorting every row separately\n");
m2 = MxSort( m, 0 ) ;
printf( MxToString( m2 ) + "\n\n" );
printf("Sorting every column separately\n");
m3 = MxSort( m, 1 ) ;
printf( MxToString( m3 )+ "\n\n");
printf("Sorting rows by contents of first column\n");
m4 = MxSortRows( m, True, 0 ) ;
printf(MxToString( m4 )+ "\n\n");
printf("Sorting rows by contents of second column\n");
m5 = MxSortRows( m, True, 1 ) ;
printf(MxToString( m5 )+ "\n\n");
HOW TO REPORT BUGS
If you experience any problem with this beta version please send detailed description of the problem (especially the steps needed to reproduce it) to support at amibroker.com