def __init__( self, fn, mode ):
Frontend.__init__( self, fn, mode );
self._kdbfn = None;
self._kdb = None;
self._ldbdn = None;
self._ldb = None;
self._len_c = None;
self._len_b = None;
self._len_x = None;
self._ic = None;
self._icbp = None;
self._needs_initialization = True;
self._core_dims = set();
self._satellite_dims = set();
self._removed_dims = set();
self._remove_c = set();
self._remove_b = set();
self._remove_x = set();
self.bypass_c = False;
self.bypass_b = False;
self.bypass_x = False;
def __init__( self, fn, mode ):
Frontend.__init__( self, fn, mode );
self._max_rows = 50000;
self._dbs = {};
self._dbdirs = [];
def _finalize( self ):
assert Frontend._finalize( self ) is None;
rowids_for_ntile_bounds = [];
for i in range( 1, 32 ):
rowids_for_ntile_bounds.append(
int( float(self._rowcount) * float(i)/32.0 )
);
ntile_bound_by_dim = {};
for i in self._dbs:
ntile_bound_by_dim[ i ] = [];
for ( i, db ) in self._dbs.items():
with db.iterator() as it:
rowid = 0;
for ( valid, ( xval, xcnt ) ) in enumerate( it ):
xval = unpack( '>I', xval )[ 0 ] - (1<<31);
xcnt = unpack( '>I', xcnt )[ 0 ];
rowid_old = rowid;
rowid = rowid_old + xcnt;
for bound in rowids_for_ntile_bounds:
if rowid_old < bound <= rowid:
ntile_bound_by_dim[ i ].append( xval );
self._state = ntile_bound_by_dim;
def train( self, row ):
if Frontend.train( self, row ):
return True;
for ( i, xval ) in enumerate( row ):
if not i in self._dbs:
dbdn = None;
with TemporaryDirectory() as tmpdirname:
dbdn = tmpdirname;
self._dbdirs.append( dbdn );
self._dbs[ i ] = plyvel.DB( dbdn, create_if_missing=True );
xval \
= int( float(xval) * 1000.0 );
assert xval < (1<<30);
xval \
= pack( ">I", (1<<31) + xval );
xcnt \
= unpack(
">I",
self._dbs[ i ].get( xval, default=pack( ">I", 0 ) )
)[ 0 ];
xcnt += 1;
self._dbs[ i ].put( xval, pack( ">I", xcnt ) );
return False;
def train( self, row ):
if Frontend.train( self, row ):
return True;
self._data.append( row );
return False;
def __exit__( self, exc_type, exc_value, traceback ):
assert Frontend.__exit__( self, exc_type, exc_value, traceback ) == False;
sleep( 3.0 );
for db in self._dbs.values():
db.close();
for dn in self._dbdirs:
rmtree( dn );
return False;
def train( self, row ):
if Frontend.train( self, row ):
return True;
b = 0;
for i in range( 0, len(row) ):
if row[i] == 1:
b |= (1<<i);
self._stats[ b ] = self._stats.get( b, 0 ) + 1;
return False;
def _finalize( self ):
assert Frontend._finalize( self ) is None;
data = np.array( self._data ).T;
rest = [];
for i in range( 1, len(data)+1 ):
rest.append( {i} );
while True:
len_rest_before = len(rest);
rest = self._merge( data, rest );
# print( rest );
if len( rest ) == len_rest_before:
break;
self._state = rest;
if False:
print( "-- CLUSTERS --" );
print( len(rest), rest );
if False:
print( "-- INSIDE CORRELATIONS --" );
for dc in rest:
print( dc );
for di in dc:
for dj in dc:
if abs(di) >= abs(dj):
continue;
corr = np.corrcoef( data[abs(di)-1], data[abs(dj)-1] )[ 0 ][ 1 ];
print( di, dj, corr );
if False:
print( "-- OUTSIDE CORRELATIONS --" );
for ( i, dci ) in enumerate( rest ):
ivals = self._cluster_val( data, dci );
for ( j, dcj ) in enumerate( rest ):
if i >= j:
continue;
jvals = self._cluster_val( data, dcj );
corr = np.corrcoef( ivals, jvals )[ 0 ][ 1 ];
print( i, j, corr );
def __call__( self, row ):
assert Frontend.__call__( self, row ) is row;
val = 0;
for (i,row_i) in enumerate( row ):
if row_i == 1:
val |= (1<<i);
row_ = [];
for cluster in self._state:
mask = 0;
for dim in cluster:
mask |= (1<<dim);
row_.append( val & mask );
return row_;
def __exit__( self, exc_type, exc_value, traceback ):
assert Frontend.__exit__( self, exc_type, exc_value, traceback ) == False;
if self._ldb is not None:
sleep( 3.0 );
self._ldb.close()
if self._ldbdn is not None:
rmtree( self._ldbdn );
if self._kdb is not None:
try:
assert self._kdb.close();
except:
print( str( self._kdb.error() ) );
raise;
if self._kdbfn is not None:
remove( self._kdbfn );
def __call__( self, row ):
assert Frontend.__call__( self, row ) is row;
row_ = [];
for cluster in self._state:
val = 0.0;
for dim in cluster:
if dim > 0:
val += row[dim];
else:
assert dim < 0;
val -= row[dim];
val /= len(cluster);
row_.append( val );
return row_;
def train( self, row ):
( y, c, b, x ) = row;
if self._len_c is None:
self._len_c = len(c);
assert self._len_c == len(c);
if self._len_b is None:
self._len_b = len(b);
assert self._len_b == len(b);
if self._len_x is None:
self._len_x = len(x);
assert self._len_x == len(x);
row = c + b + x;
if Frontend.train( self, row ):
return True;
keyfmt = '>IIIII';
for i in range( 0, self._lenrow ):
for j in range( 0, self._lenrow ):
if ( i >= j ) and ( not ( i == self._lenrow-1 ) ):
continue;
key = pack( keyfmt, i, j, y, row[i], row[j] );
try:
assert self._kdb.increment( key, 1, 0 );
except:
print( str(self._kdb.error()) );
raise;
def _finalize( self ):
assert Frontend._finalize( self ) is None;
self._state = [];
rest = set( range( 0, self._lenrow ) );
while rest:
if len( rest ) >= 2:
( left, right, rest ) = self._split( rest );
self._state.append( left );
self._state.append( right );
else:
self._state.append( rest );
rest = set();
if False:
for ( cluster_id, cluster ) in enumerate( self._state ):
print( "-->", cluster_id, cluster );
for i in cluster:
a = { i };
b = cluster - a;
print( " {:d} {:1.4f}".format( i, self._i_corr( a, b ) ) );
def _finalize( self ):
assert Frontend._finalize( self ) is None;
if False:
print( "unique combinations = ", self._kdb.count() );
keyfmt = '>IIIII';
valfmt = '>Q';
c = self._kdb.cursor();
c.jump();
gt2 = 0;
gt4 = 0;
gt8 = 0;
gt16 = 0;
gt32 = 0;
while True:
r = c.get( True );
if not r:
break;
self._ldb.put( r[0], r[1] );
key = unpack( keyfmt, r[0] );
val = unpack( valfmt, r[1] )[ 0 ];
if val > 2:
gt2 += 1;
if val > 4:
gt4 += 1;
if val > 8:
gt8 += 1;
if val > 16:
gt16 += 1;
if val > 32:
gt32 += 1;
if False:
print( gt2, gt4, gt8, gt16, gt32 );
self._ic = {};
for i in range( 0, self._lenrow ):
self._ic[ i ] = self._get_info_content_by_dimension( i );
self._icbp = {};
for i in range( 0, self._lenrow ):
for j in range( 0, self._lenrow ):
if i >= j:
continue;
self._icbp[ (i,j) ] = self._get_info_content_by_pair( i, j );
self._state \
= { "ic": self._ic,
"icbp": self._icbp,
"c": self._len_c,
"b": self._len_b,
"x": self._len_x };
def __init__( self, fn, mode ):
Frontend.__init__( self, fn, mode );
self._max_rows = 100000;
self._stats = {};
def __init__( self, fn, mode ):
Frontend.__init__( self, fn, mode );
self._max_rows = 100000;
self._data = [];