def blockmatrix_irs(self):
scalar_ir = ir.F64(2)
vector_ir = ir.MakeArray([ir.F64(3), ir.F64(2)],
hl.tarray(hl.tfloat64))
read = ir.BlockMatrixRead(
ir.BlockMatrixNativeReader(resource('blockmatrix_example/0')))
add_two_bms = ir.BlockMatrixMap2(
read, read, 'l', 'r',
ir.ApplyBinaryPrimOp('+', ir.Ref('l'), ir.Ref('r')), "Union")
negate_bm = ir.BlockMatrixMap(
read, 'element', ir.ApplyUnaryPrimOp('-', ir.Ref('element')),
False)
sqrt_bm = ir.BlockMatrixMap(
read, 'element',
hl.sqrt(construct_expr(ir.Ref('element'), hl.tfloat64))._ir, False)
persisted = ir.BlockMatrixRead(ir.BlockMatrixPersistReader('x', read))
scalar_to_bm = ir.ValueToBlockMatrix(scalar_ir, [1, 1], 1)
col_vector_to_bm = ir.ValueToBlockMatrix(vector_ir, [2, 1], 1)
row_vector_to_bm = ir.ValueToBlockMatrix(vector_ir, [1, 2], 1)
broadcast_scalar = ir.BlockMatrixBroadcast(scalar_to_bm, [], [2, 2],
256)
broadcast_col = ir.BlockMatrixBroadcast(col_vector_to_bm, [0], [2, 2],
256)
broadcast_row = ir.BlockMatrixBroadcast(row_vector_to_bm, [1], [2, 2],
256)
transpose = ir.BlockMatrixBroadcast(broadcast_scalar, [1, 0], [2, 2],
256)
matmul = ir.BlockMatrixDot(broadcast_scalar, transpose)
rectangle = ir.Literal(hl.tarray(hl.tint64), [0, 1, 5, 6])
band = ir.Literal(hl.ttuple(hl.tint64, hl.tint64), (-1, 1))
intervals = ir.Literal(
hl.ttuple(hl.tarray(hl.tint64), hl.tarray(hl.tint64)),
([0, 1, 5, 6], [5, 6, 8, 9]))
sparsify1 = ir.BlockMatrixSparsify(read, rectangle,
ir.RectangleSparsifier)
sparsify2 = ir.BlockMatrixSparsify(read, band, ir.BandSparsifier(True))
sparsify3 = ir.BlockMatrixSparsify(read, intervals,
ir.RowIntervalSparsifier(True))
densify = ir.BlockMatrixDensify(read)
pow_ir = (construct_expr(ir.Ref('l'), hl.tfloat64)**construct_expr(
ir.Ref('r'), hl.tfloat64))._ir
squared_bm = ir.BlockMatrixMap2(scalar_to_bm, scalar_to_bm, 'l', 'r',
pow_ir, "NeedsDense")
slice_bm = ir.BlockMatrixSlice(
matmul, [slice(0, 2, 1), slice(0, 1, 1)])
return [
read, persisted, add_two_bms, negate_bm, sqrt_bm, scalar_to_bm,
col_vector_to_bm, row_vector_to_bm, broadcast_scalar,
broadcast_col, broadcast_row, squared_bm, transpose, sparsify1,
sparsify2, sparsify3, densify, matmul, slice_bm
]
def test_parses(self):
backend = Env.spark_backend('BlockMatrixIRTests.test_parses')
bmir = hl.linalg.BlockMatrix.fill(1, 1, 0.0)._bmir
backend.execute(ir.BlockMatrixWrite(bmir, ir.BlockMatrixPersistWriter('x', 'MEMORY_ONLY')))
persist = ir.BlockMatrixRead(ir.BlockMatrixPersistReader('x', bmir))
for x in (self.blockmatrix_irs() + [persist]):
backend._parse_blockmatrix_ir(str(x))
backend.unpersist_block_matrix('x')
def blockmatrix_irs(self):
scalar_ir = ir.F64(2)
vector_ir = ir.MakeArray([ir.F64(3), ir.F64(2)],
hl.tarray(hl.tfloat64))
read = ir.BlockMatrixRead(
ir.BlockMatrixNativeReader(resource('blockmatrix_example/0')))
add_two_bms = ir.BlockMatrixMap2(
read, read, 'l', 'r',
ir.ApplyBinaryPrimOp('+', ir.Ref('l'), ir.Ref('r')))
negate_bm = ir.BlockMatrixMap(
read, 'element', ir.ApplyUnaryPrimOp('-', ir.Ref('element')))
sqrt_bm = ir.BlockMatrixMap(
read, 'element',
hl.sqrt(construct_expr(ir.Ref('element'), hl.tfloat64))._ir)
scalar_to_bm = ir.ValueToBlockMatrix(scalar_ir, [1, 1], 1)
col_vector_to_bm = ir.ValueToBlockMatrix(vector_ir, [2, 1], 1)
row_vector_to_bm = ir.ValueToBlockMatrix(vector_ir, [1, 2], 1)
broadcast_scalar = ir.BlockMatrixBroadcast(scalar_to_bm, [], [2, 2],
256)
broadcast_col = ir.BlockMatrixBroadcast(col_vector_to_bm, [0], [2, 2],
256)
broadcast_row = ir.BlockMatrixBroadcast(row_vector_to_bm, [1], [2, 2],
256)
transpose = ir.BlockMatrixBroadcast(broadcast_scalar, [1, 0], [2, 2],
256)
matmul = ir.BlockMatrixDot(broadcast_scalar, transpose)
pow_ir = (construct_expr(ir.Ref('l'), hl.tfloat64)**construct_expr(
ir.Ref('r'), hl.tfloat64))._ir
squared_bm = ir.BlockMatrixMap2(scalar_to_bm, scalar_to_bm, 'l', 'r',
pow_ir)
slice_bm = ir.BlockMatrixSlice(
matmul, [slice(0, 2, 1), slice(0, 1, 1)])
return [
read, add_two_bms, negate_bm, sqrt_bm, scalar_to_bm,
col_vector_to_bm, row_vector_to_bm, broadcast_scalar,
broadcast_col, broadcast_row, squared_bm, transpose, matmul,
slice_bm
]
def block_matrix_irs(self):
scalar_ir = ir.F64(2)
vector_ir = ir.MakeArray([ir.F64(3), ir.F64(2)], hl.tarray(hl.tfloat64))
read = ir.BlockMatrixRead(resource('blockmatrix_example/0'))
add_two_bms = BlockMatrixIRTests._make_element_wise_op_ir(read, read, '+')
scalar_to_bm = ir.ValueToBlockMatrix(scalar_ir, [], 1, [])
vector_to_bm = ir.ValueToBlockMatrix(vector_ir, [2], 1, [False])
broadcast_scalar = ir.BlockMatrixBroadcast(scalar_to_bm, "scalar", [2, 2], 256, [False, False])
broadcast_col = ir.BlockMatrixBroadcast(vector_to_bm, "col", [2, 2], 256, [False, False])
broadcast_row = ir.BlockMatrixBroadcast(vector_to_bm, "row", [2, 2], 256, [False, False])
return [
read,
add_two_bms,
scalar_to_bm,
vector_to_bm,
broadcast_scalar,
broadcast_col,
broadcast_row,
]
def table_irs(self):
b = ir.TrueIR()
table_read = ir.TableRead(
ir.TableNativeReader(
resource('backward_compatability/1.0.0/table/0.ht'), None,
False), False)
table_read_row_type = hl.dtype(
'struct{idx: int32, f32: float32, i64: int64, m: float64, astruct: struct{a: int32, b: float64}, mstruct: struct{x: int32, y: str}, aset: set<str>, mset: set<float64>, d: dict<array<str>, float64>, md: dict<int32, str>, h38: locus<GRCh38>, ml: locus<GRCh37>, i: interval<locus<GRCh37>>, c: call, mc: call, t: tuple(call, str, str), mt: tuple(locus<GRCh37>, bool)}'
)
matrix_read = ir.MatrixRead(
ir.MatrixNativeReader(
resource('backward_compatability/1.0.0/matrix_table/0.hmt'),
None, False), False, False)
block_matrix_read = ir.BlockMatrixRead(
ir.BlockMatrixNativeReader('fake_file_path'))
aa = hl.literal([[0.00], [0.01], [0.02]])._ir
range = ir.TableRange(10, 4)
table_irs = [
ir.TableKeyBy(table_read, ['m', 'd'], False),
ir.TableFilter(table_read, b),
table_read,
ir.MatrixColsTable(matrix_read),
ir.TableAggregateByKey(table_read,
ir.MakeStruct([('a', ir.I32(5))])),
ir.TableKeyByAndAggregate(table_read,
ir.MakeStruct([('a', ir.I32(5))]),
ir.MakeStruct([('b', ir.I32(5))]), 1, 2),
ir.TableJoin(table_read, ir.TableRange(100, 10), 'inner', 1),
ir.MatrixEntriesTable(matrix_read),
ir.MatrixRowsTable(matrix_read),
ir.TableParallelize(
ir.MakeStruct([('rows',
ir.Literal(hl.tarray(hl.tstruct(a=hl.tint32)),
[{
'a': None
}, {
'a': 5
}, {
'a': -3
}])),
('global', ir.MakeStruct([]))]), None),
ir.TableMapRows(
ir.TableKeyBy(table_read, []),
ir.MakeStruct([('a', ir.GetField(ir.Ref('row'), 'f32')),
('b', ir.F64(-2.11))])),
ir.TableMapGlobals(
table_read,
ir.MakeStruct([('foo', ir.NA(hl.tarray(hl.tint32)))])),
ir.TableRange(100, 10),
ir.TableRepartition(table_read, 10,
ir.RepartitionStrategy.COALESCE),
ir.TableUnion([ir.TableRange(100, 10),
ir.TableRange(50, 10)]),
ir.TableExplode(table_read, ['mset']),
ir.TableHead(table_read, 10),
ir.TableOrderBy(ir.TableKeyBy(table_read, []), [('m', 'A'),
('m', 'D')]),
ir.TableDistinct(table_read),
ir.CastMatrixToTable(matrix_read, '__entries', '__cols'),
ir.TableRename(table_read, {'idx': 'idx_foo'},
{'global_f32': 'global_foo'}),
ir.TableMultiWayZipJoin([table_read, table_read], '__data',
'__globals'),
ir.MatrixToTableApply(
matrix_read, {
'name': 'LinearRegressionRowsSingle',
'yFields': ['col_m'],
'xField': 'entry_m',
'covFields': [],
'rowBlockSize': 10,
'passThrough': []
}),
ir.TableToTableApply(table_read, {
'name': 'TableFilterPartitions',
'parts': [0],
'keep': True
}),
ir.BlockMatrixToTableApply(block_matrix_read, aa, {
'name': 'PCRelate',
'maf': 0.01,
'blockSize': 4096
}),
ir.TableFilterIntervals(table_read, [
hl.utils.Interval(hl.utils.Struct(row_idx=0),
hl.utils.Struct(row_idx=10))
],
hl.tstruct(row_idx=hl.tint32),
keep=False),
]
return table_irs
def value_irs(self):
b = ir.TrueIR()
c = ir.Ref('c')
i = ir.I32(5)
j = ir.I32(7)
st = ir.Str('Hail')
a = ir.Ref('a')
aa = ir.Ref('aa')
da = ir.Ref('da')
nd = ir.Ref('nd')
v = ir.Ref('v')
s = ir.Ref('s')
t = ir.Ref('t')
call = ir.Ref('call')
table = ir.TableRange(5, 3)
matrix_read = ir.MatrixRead(
ir.MatrixNativeReader(
resource('backward_compatability/1.0.0/matrix_table/0.hmt'),
None, False), False, False)
block_matrix_read = ir.BlockMatrixRead(
ir.BlockMatrixNativeReader('fake_file_path'))
value_irs = [
i,
ir.I64(5),
ir.F32(3.14),
ir.F64(3.14), s,
ir.TrueIR(),
ir.FalseIR(),
ir.Void(),
ir.Cast(i, hl.tfloat64),
ir.NA(hl.tint32),
ir.IsNA(i),
ir.If(b, i, j),
ir.Coalesce(i, j),
ir.Let('v', i, v),
ir.Ref('x'),
ir.ApplyBinaryPrimOp('+', i, j),
ir.ApplyUnaryPrimOp('-', i),
ir.ApplyComparisonOp('EQ', i, j),
ir.MakeArray([i, ir.NA(hl.tint32), ir.I32(-3)],
hl.tarray(hl.tint32)),
ir.ArrayRef(a, i, ir.Str('foo')),
ir.ArrayLen(a),
ir.ArrayRange(ir.I32(0), ir.I32(5), ir.I32(1)),
ir.ArraySort(a, 'l', 'r',
ir.ApplyComparisonOp("LT", ir.Ref('l'), ir.Ref('r'))),
ir.ToSet(a),
ir.ToDict(da),
ir.ToArray(a),
ir.MakeNDArray(
ir.MakeArray([ir.F64(-1.0), ir.F64(1.0)],
hl.tarray(hl.tfloat64)),
ir.MakeTuple([ir.I64(1), ir.I64(2)]), ir.TrueIR()),
ir.NDArrayShape(nd),
ir.NDArrayReshape(nd, ir.MakeTuple([ir.I64(5)])),
ir.NDArrayRef(nd, [ir.I64(1), ir.I64(2)]),
ir.NDArrayMap(nd, 'v', v),
ir.NDArrayMatMul(nd, nd),
ir.LowerBoundOnOrderedCollection(a, i, True),
ir.GroupByKey(da),
ir.ArrayMap(a, 'v', v),
ir.ArrayZip([a, a], ['a', 'b'], ir.TrueIR(), 'ExtendNA'),
ir.ArrayFilter(a, 'v', v),
ir.ArrayFlatMap(aa, 'v', v),
ir.ArrayFold(a, ir.I32(0), 'x', 'v', v),
ir.ArrayScan(a, ir.I32(0), 'x', 'v', v),
ir.ArrayLeftJoinDistinct(a, a, 'l', 'r', ir.I32(0), ir.I32(1)),
ir.ArrayFor(a, 'v', ir.Void()),
ir.AggFilter(ir.TrueIR(), ir.I32(0), False),
ir.AggExplode(ir.ArrayRange(ir.I32(0), ir.I32(2), ir.I32(1)), 'x',
ir.I32(0), False),
ir.AggGroupBy(ir.TrueIR(), ir.I32(0), False),
ir.AggArrayPerElement(
ir.ArrayRange(ir.I32(0), ir.I32(2), ir.I32(1)), 'x', 'y',
ir.I32(0), False),
ir.ApplyAggOp('Collect', [], [ir.I32(0)]),
ir.ApplyScanOp('Collect', [], [ir.I32(0)]),
ir.ApplyAggOp('CallStats', [ir.I32(2)], [call]),
ir.ApplyAggOp('TakeBy', [ir.I32(10)],
[ir.F64(-2.11), ir.F64(-2.11)]),
ir.Begin([ir.Void()]),
ir.MakeStruct([('x', i)]),
ir.SelectFields(s, ['x', 'z']),
ir.InsertFields(s, [('x', i)], None),
ir.GetField(s, 'x'),
ir.MakeTuple([i, b]),
ir.GetTupleElement(t, 1),
ir.Die(ir.Str('mumblefoo'), hl.tfloat64),
ir.Apply('&&', hl.tbool, b, c),
ir.Apply('toFloat64', hl.tfloat64, i),
ir.Literal(hl.tarray(hl.tint32), [1, 2, None]),
ir.TableCount(table),
ir.TableGetGlobals(table),
ir.TableCollect(ir.TableKeyBy(table, [], False)),
ir.TableToValueApply(table, {'name': 'ForceCountTable'}),
ir.MatrixToValueApply(matrix_read,
{'name': 'ForceCountMatrixTable'}),
ir.TableAggregate(
table,
ir.MakeStruct([('foo', ir.ApplyAggOp('Collect', [],
[ir.I32(0)]))])),
ir.TableWrite(
table,
ir.TableNativeWriter(new_temp_file(), False, True,
"fake_codec_spec$$")),
ir.TableWrite(
table, ir.TableTextWriter(new_temp_file(), None, True, 0,
",")),
ir.MatrixAggregate(
matrix_read,
ir.MakeStruct([('foo', ir.ApplyAggOp('Collect', [],
[ir.I32(0)]))])),
ir.MatrixWrite(
matrix_read,
ir.MatrixNativeWriter(new_temp_file(), False, False, "", None,
None)),
ir.MatrixWrite(
matrix_read,
ir.MatrixNativeWriter(
new_temp_file(), False, False, "",
'[{"start":{"row_idx":0},"end":{"row_idx": 10},"includeStart":true,"includeEnd":false}]',
hl.dtype('array<interval<struct{row_idx:int32}>>'))),
ir.MatrixWrite(
matrix_read,
ir.MatrixVCFWriter(new_temp_file(), None, False, None)),
ir.MatrixWrite(matrix_read, ir.MatrixGENWriter(new_temp_file(),
4)),
ir.MatrixWrite(matrix_read, ir.MatrixPLINKWriter(new_temp_file())),
ir.MatrixMultiWrite([matrix_read, matrix_read],
ir.MatrixNativeMultiWriter(
new_temp_file(), False, False)),
ir.BlockMatrixWrite(
block_matrix_read,
ir.BlockMatrixNativeWriter('fake_file_path', False, False,
False)),
ir.LiftMeOut(ir.I32(1))
]
return value_irs
def value_irs(self):
b = ir.TrueIR()
c = ir.Ref('c')
i = ir.I32(5)
j = ir.I32(7)
st = ir.Str('Hail')
a = ir.Ref('a')
aa = ir.Ref('aa')
da = ir.Ref('da')
nd = ir.Ref('nd')
v = ir.Ref('v')
s = ir.Ref('s')
t = ir.Ref('t')
call = ir.Ref('call')
table = ir.TableRange(5, 3)
matrix_read = ir.MatrixRead(ir.MatrixNativeReader(
resource('backward_compatability/1.0.0/matrix_table/0.hmt')), False, False)
block_matrix_read = ir.BlockMatrixRead('fake_file_path')
value_irs = [
i, ir.I64(5), ir.F32(3.14), ir.F64(3.14), s, ir.TrueIR(), ir.FalseIR(), ir.Void(),
ir.Cast(i, hl.tfloat64),
ir.NA(hl.tint32),
ir.IsNA(i),
ir.If(b, i, j),
ir.Let('v', i, v),
ir.Ref('x'),
ir.ApplyBinaryOp('+', i, j),
ir.ApplyUnaryOp('-', i),
ir.ApplyComparisonOp('EQ', i, j),
ir.MakeArray([i, ir.NA(hl.tint32), ir.I32(-3)], hl.tarray(hl.tint32)),
ir.ArrayRef(a, i),
ir.ArrayLen(a),
ir.ArrayRange(ir.I32(0), ir.I32(5), ir.I32(1)),
ir.ArraySort(a, b, False),
ir.ToSet(a),
ir.ToDict(da),
ir.ToArray(a),
ir.MakeNDArray(ir.MakeArray([ir.F64(-1.0), ir.F64(1.0)], hl.tarray(hl.tfloat64)),
ir.MakeArray([ir.I64(1), ir.I64(2)], hl.tarray(hl.tint64)),
ir.TrueIR()),
ir.NDArrayRef(nd, ir.MakeArray([ir.I64(1), ir.I64(2)], hl.tarray(hl.tint64))),
ir.LowerBoundOnOrderedCollection(a, i, True),
ir.GroupByKey(da),
ir.ArrayMap(a, 'v', v),
ir.ArrayFilter(a, 'v', v),
ir.ArrayFlatMap(aa, 'v', v),
ir.ArrayFold(a, ir.I32(0), 'x', 'v', v),
ir.ArrayScan(a, ir.I32(0), 'x', 'v', v),
ir.ArrayLeftJoinDistinct(a, a, 'l', 'r', ir.I32(0), ir.I32(1)),
ir.ArrayFor(a, 'v', ir.Void()),
ir.AggFilter(ir.TrueIR(), ir.I32(0)),
ir.AggExplode(ir.ArrayRange(ir.I32(0), ir.I32(2), ir.I32(1)), 'x', ir.I32(0)),
ir.AggGroupBy(ir.TrueIR(), ir.I32(0)),
ir.AggArrayPerElement(ir.ArrayRange(ir.I32(0), ir.I32(2), ir.I32(1)), 'x', ir.I32(0)),
ir.ApplyAggOp('Collect', [], None, [ir.I32(0)]),
ir.ApplyScanOp('Collect', [], None, [ir.I32(0)]),
ir.ApplyAggOp('Histogram', [ir.F64(-5.0), ir.F64(5.0), ir.I32(100)], None, [ir.F64(-2.11)]),
ir.ApplyAggOp('CallStats', [], [ir.I32(2)], [call]),
ir.ApplyAggOp('TakeBy', [ir.I32(10)], None, [ir.F64(-2.11), ir.F64(-2.11)]),
ir.Begin([ir.Void()]),
ir.MakeStruct([('x', i)]),
ir.SelectFields(s, ['x', 'z']),
ir.InsertFields(s, [('x', i)], None),
ir.GetField(s, 'x'),
ir.MakeTuple([i, b]),
ir.GetTupleElement(t, 1),
ir.StringSlice(st, ir.I32(1), ir.I32(2)),
ir.StringLength(st),
ir.In(2, hl.tfloat64),
ir.Die(ir.Str('mumblefoo'), hl.tfloat64),
ir.Apply('&&', b, c),
ir.Apply('toFloat64', i),
ir.Uniroot('x', ir.F64(3.14), ir.F64(-5.0), ir.F64(5.0)),
ir.Literal(hl.tarray(hl.tint32), [1, 2, None]),
ir.TableCount(table),
ir.TableGetGlobals(table),
ir.TableCollect(table),
ir.TableToValueApply(table, {'name': 'ForceCountTable'}),
ir.MatrixToValueApply(matrix_read, {'name': 'ForceCountMatrixTable'}),
ir.TableAggregate(table, ir.MakeStruct([('foo', ir.ApplyAggOp('Collect', [], None, [ir.I32(0)]))])),
ir.TableWrite(table, new_temp_file(), False, True, "fake_codec_spec$$"),
ir.TableExport(table, new_temp_file(), None, True, 0, ","),
ir.MatrixAggregate(matrix_read, ir.MakeStruct([('foo', ir.ApplyAggOp('Collect', [], None, [ir.I32(0)]))])),
ir.MatrixWrite(matrix_read, ir.MatrixNativeWriter(new_temp_file(), False, False, "")),
ir.MatrixWrite(matrix_read, ir.MatrixVCFWriter(new_temp_file(), None, False, None)),
ir.MatrixWrite(matrix_read, ir.MatrixGENWriter(new_temp_file(), 4)),
ir.MatrixWrite(matrix_read, ir.MatrixPLINKWriter(new_temp_file())),
ir.MatrixMultiWrite([matrix_read, matrix_read], ir.MatrixNativeMultiWriter(new_temp_file(), False, False)),
ir.BlockMatrixWrite(block_matrix_read, 'fake_file_path', False, False, False)
]
return value_irs
def block_matrix_irs(self):
read = ir.BlockMatrixRead('fake_file_path')
add = ir.BlockMatrixAdd(read, read)
return [read, add]
