ClickHouse/src/Functions/vectorFunctions.cpp

1002 lines
40 KiB
C++
Raw Normal View History

#include <Columns/ColumnTuple.h>
#include <DataTypes/DataTypeTuple.h>
#include <DataTypes/DataTypesNumber.h>
#include <Functions/FunctionFactory.h>
#include <Functions/FunctionHelpers.h>
#include <Functions/TupleIFunction.h>
#include <Functions/castTypeToEither.h>
namespace DB
{
namespace ErrorCodes
{
2021-08-20 21:14:26 +00:00
extern const int ILLEGAL_TYPE_OF_ARGUMENT;
}
2021-08-30 12:52:00 +00:00
static const char PLUS_NAME[] = "plus";
static const char MINUS_NAME[] = "minus";
static const char MULTIPLY_NAME[] = "multiply";
static const char DIVIDE_NAME[] = "divide";
static constexpr char L1_LABEL[] = "1";
static constexpr char L2_LABEL[] = "2";
static constexpr char Linf_LABEL[] = "inf";
static constexpr char Lp_LABEL[] = "p";
2021-08-27 15:02:37 +00:00
/// str starts from the lowercase letter; not constexpr due to the compiler version
2021-08-27 15:44:04 +00:00
/*constexpr*/ std::string makeFirstLetterUppercase(std::string && str)
2021-08-27 15:02:37 +00:00
{
std::string res(str);
res[0] += 'A' - 'a';
return res;
}
2021-08-25 14:56:31 +00:00
template <const char * func_name>
2021-08-27 15:44:04 +00:00
class FunctionTupleOperator : public TupleIFunction
2021-08-26 18:17:44 +00:00
{
2021-08-25 14:56:31 +00:00
public:
2021-08-27 15:02:37 +00:00
/// constexpr cannot be used due to std::string has not constexpr constructor in this compiler version
2021-08-27 15:44:04 +00:00
static inline auto name = "tuple" + makeFirstLetterUppercase(func_name);
2021-08-27 15:02:37 +00:00
2021-08-27 15:44:04 +00:00
explicit FunctionTupleOperator(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionTupleOperator>(context_); }
2021-08-27 15:02:37 +00:00
String getName() const override { return name; }
2021-08-25 14:56:31 +00:00
size_t getNumberOfArguments() const override { return 2; }
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
const auto * left_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto * right_tuple = checkAndGetDataType<DataTypeTuple>(arguments[1].type.get());
if (!left_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 0 of function {} should be tuples, got {}",
getName(), arguments[0].type->getName());
if (!right_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 1 of function {} should be tuples, got {}",
getName(), arguments[1].type->getName());
const auto & left_types = left_tuple->getElements();
const auto & right_types = right_tuple->getElements();
Columns left_elements;
Columns right_elements;
if (arguments[0].column)
left_elements = getTupleElements(*arguments[0].column);
if (arguments[1].column)
right_elements = getTupleElements(*arguments[1].column);
if (left_types.size() != right_types.size())
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT,
"Expected tuples of the same size as arguments of function {}. Got {} and {}",
getName(), arguments[0].type->getName(), arguments[1].type->getName());
size_t tuple_size = left_types.size();
if (tuple_size == 0)
return std::make_shared<DataTypeUInt8>();
auto func = FunctionFactory::instance().get(func_name, context);
DataTypes types(tuple_size);
for (size_t i = 0; i < tuple_size; ++i)
{
try
{
ColumnWithTypeAndName left{left_elements.empty() ? nullptr : left_elements[i], left_types[i], {}};
ColumnWithTypeAndName right{right_elements.empty() ? nullptr : right_elements[i], right_types[i], {}};
auto elem_func = func->build(ColumnsWithTypeAndName{left, right});
types[i] = elem_func->getResultType();
}
catch (DB::Exception & e)
{
e.addMessage("While executing function {} for tuple element {}", getName(), i);
throw;
}
}
return std::make_shared<DataTypeTuple>(types);
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
const auto * left_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto * right_tuple = checkAndGetDataType<DataTypeTuple>(arguments[1].type.get());
const auto & left_types = left_tuple->getElements();
const auto & right_types = right_tuple->getElements();
auto left_elements = getTupleElements(*arguments[0].column);
auto right_elements = getTupleElements(*arguments[1].column);
size_t tuple_size = left_elements.size();
if (tuple_size == 0)
return DataTypeUInt8().createColumnConstWithDefaultValue(input_rows_count);
auto func = FunctionFactory::instance().get(func_name, context);
Columns columns(tuple_size);
for (size_t i = 0; i < tuple_size; ++i)
{
ColumnWithTypeAndName left{left_elements[i], left_types[i], {}};
ColumnWithTypeAndName right{right_elements[i], right_types[i], {}};
auto elem_func = func->build(ColumnsWithTypeAndName{left, right});
2021-08-27 13:14:32 +00:00
columns[i] = elem_func->execute({left, right}, elem_func->getResultType(), input_rows_count)
->convertToFullColumnIfConst();
2021-08-25 14:56:31 +00:00
}
return ColumnTuple::create(columns);
}
};
2021-08-27 15:44:04 +00:00
using FunctionTuplePlus = FunctionTupleOperator<PLUS_NAME>;
2021-08-27 15:44:04 +00:00
using FunctionTupleMinus = FunctionTupleOperator<MINUS_NAME>;
2021-08-27 15:44:04 +00:00
using FunctionTupleMultiply = FunctionTupleOperator<MULTIPLY_NAME>;
2021-08-27 15:44:04 +00:00
using FunctionTupleDivide = FunctionTupleOperator<DIVIDE_NAME>;
2021-08-25 14:02:04 +00:00
class FunctionTupleNegate : public TupleIFunction
{
public:
2021-08-25 14:02:04 +00:00
static constexpr auto name = "tupleNegate";
2021-08-25 14:02:04 +00:00
explicit FunctionTupleNegate(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionTupleNegate>(context_); }
String getName() const override { return name; }
2021-08-25 14:02:04 +00:00
size_t getNumberOfArguments() const override { return 1; }
2021-08-20 16:23:51 +00:00
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
2021-08-25 14:02:04 +00:00
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
2021-08-25 14:02:04 +00:00
if (!cur_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 0 of function {} should be tuples, got {}",
getName(), arguments[0].type->getName());
2021-08-25 14:02:04 +00:00
const auto & cur_types = cur_tuple->getElements();
2021-08-25 14:02:04 +00:00
Columns cur_elements;
if (arguments[0].column)
2021-08-25 14:02:04 +00:00
cur_elements = getTupleElements(*arguments[0].column);
2021-08-25 14:02:04 +00:00
size_t tuple_size = cur_types.size();
if (tuple_size == 0)
return std::make_shared<DataTypeUInt8>();
2021-08-25 14:02:04 +00:00
auto negate = FunctionFactory::instance().get("negate", context);
DataTypes types(tuple_size);
for (size_t i = 0; i < tuple_size; ++i)
{
try
{
2021-08-25 14:02:04 +00:00
ColumnWithTypeAndName cur{cur_elements.empty() ? nullptr : cur_elements[i], cur_types[i], {}};
auto elem_negate = negate->build(ColumnsWithTypeAndName{cur});
types[i] = elem_negate->getResultType();
}
catch (DB::Exception & e)
{
e.addMessage("While executing function {} for tuple element {}", getName(), i);
throw;
}
}
return std::make_shared<DataTypeTuple>(types);
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
2021-08-25 14:02:04 +00:00
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto & cur_types = cur_tuple->getElements();
auto cur_elements = getTupleElements(*arguments[0].column);
2021-08-25 14:02:04 +00:00
size_t tuple_size = cur_elements.size();
if (tuple_size == 0)
return DataTypeUInt8().createColumnConstWithDefaultValue(input_rows_count);
2021-08-25 14:02:04 +00:00
auto negate = FunctionFactory::instance().get("negate", context);
Columns columns(tuple_size);
for (size_t i = 0; i < tuple_size; ++i)
{
2021-08-25 14:02:04 +00:00
ColumnWithTypeAndName cur{cur_elements[i], cur_types[i], {}};
2021-08-27 15:44:04 +00:00
auto elem_negate = negate->build(ColumnsWithTypeAndName{cur});
2021-08-25 14:02:04 +00:00
columns[i] = elem_negate->execute({cur}, elem_negate->getResultType(), input_rows_count);
}
return ColumnTuple::create(columns);
}
};
2021-08-27 15:44:04 +00:00
template <const char * func_name>
class FunctionTupleOperatorByNumber : public TupleIFunction
{
public:
/// constexpr cannot be used due to std::string has not constexpr constructor in this compiler version
static inline auto name = "tuple" + makeFirstLetterUppercase(func_name) + "ByNumber";
explicit FunctionTupleOperatorByNumber(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionTupleOperatorByNumber>(context_); }
String getName() const override { return name; }
size_t getNumberOfArguments() const override { return 2; }
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
if (!cur_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 0 of function {} should be tuples, got {}",
getName(), arguments[0].type->getName());
const auto & cur_types = cur_tuple->getElements();
Columns cur_elements;
if (arguments[0].column)
cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_types.size();
if (tuple_size == 0)
return std::make_shared<DataTypeUInt8>();
const auto & p_column = arguments[1];
auto func = FunctionFactory::instance().get(func_name, context);
DataTypes types(tuple_size);
for (size_t i = 0; i < tuple_size; ++i)
{
try
{
ColumnWithTypeAndName cur{cur_elements.empty() ? nullptr : cur_elements[i], cur_types[i], {}};
auto elem_func = func->build(ColumnsWithTypeAndName{cur, p_column});
types[i] = elem_func->getResultType();
}
catch (DB::Exception & e)
{
e.addMessage("While executing function {} for tuple element {}", getName(), i);
throw;
}
}
return std::make_shared<DataTypeTuple>(types);
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto & cur_types = cur_tuple->getElements();
auto cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_elements.size();
if (tuple_size == 0)
return DataTypeUInt8().createColumnConstWithDefaultValue(input_rows_count);
const auto & p_column = arguments[1];
auto func = FunctionFactory::instance().get(func_name, context);
Columns columns(tuple_size);
for (size_t i = 0; i < tuple_size; ++i)
{
ColumnWithTypeAndName cur{cur_elements[i], cur_types[i], {}};
auto elem_func = func->build(ColumnsWithTypeAndName{cur, p_column});
columns[i] = elem_func->execute({cur, p_column}, elem_func->getResultType(), input_rows_count);
}
return ColumnTuple::create(columns);
}
};
using FunctionTupleMultiplyByNumber = FunctionTupleOperatorByNumber<MULTIPLY_NAME>;
using FunctionTupleDivideByNumber = FunctionTupleOperatorByNumber<DIVIDE_NAME>;
2021-08-25 14:02:04 +00:00
class FunctionDotProduct : public TupleIFunction
2021-08-23 13:41:20 +00:00
{
public:
2021-08-25 14:02:04 +00:00
static constexpr auto name = "dotProduct";
2021-08-23 13:41:20 +00:00
2021-08-25 14:02:04 +00:00
explicit FunctionDotProduct(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionDotProduct>(context_); }
2021-08-23 13:41:20 +00:00
String getName() const override { return name; }
size_t getNumberOfArguments() const override { return 2; }
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
const auto * left_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto * right_tuple = checkAndGetDataType<DataTypeTuple>(arguments[1].type.get());
if (!left_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 0 of function {} should be tuples, got {}",
getName(), arguments[0].type->getName());
if (!right_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 1 of function {} should be tuples, got {}",
getName(), arguments[1].type->getName());
const auto & left_types = left_tuple->getElements();
const auto & right_types = right_tuple->getElements();
Columns left_elements;
Columns right_elements;
if (arguments[0].column)
left_elements = getTupleElements(*arguments[0].column);
if (arguments[1].column)
right_elements = getTupleElements(*arguments[1].column);
if (left_types.size() != right_types.size())
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT,
"Expected tuples of the same size as arguments of function {}. Got {} and {}",
getName(), arguments[0].type->getName(), arguments[1].type->getName());
size_t tuple_size = left_types.size();
if (tuple_size == 0)
return std::make_shared<DataTypeUInt8>();
2021-08-25 14:02:04 +00:00
auto multiply = FunctionFactory::instance().get("multiply", context);
auto plus = FunctionFactory::instance().get("plus", context);
DataTypePtr res_type;
2021-08-23 13:41:20 +00:00
for (size_t i = 0; i < tuple_size; ++i)
{
try
{
ColumnWithTypeAndName left{left_elements.empty() ? nullptr : left_elements[i], left_types[i], {}};
ColumnWithTypeAndName right{right_elements.empty() ? nullptr : right_elements[i], right_types[i], {}};
2021-08-25 14:02:04 +00:00
auto elem_multiply = multiply->build(ColumnsWithTypeAndName{left, right});
if (i == 0)
{
res_type = elem_multiply->getResultType();
continue;
}
ColumnWithTypeAndName left_type{res_type, {}};
ColumnWithTypeAndName right_type{elem_multiply->getResultType(), {}};
auto plus_elem = plus->build({left_type, right_type});
res_type = plus_elem->getResultType();
2021-08-23 13:41:20 +00:00
}
catch (DB::Exception & e)
{
e.addMessage("While executing function {} for tuple element {}", getName(), i);
throw;
}
}
2021-08-25 14:02:04 +00:00
return res_type;
2021-08-23 13:41:20 +00:00
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
const auto * left_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto * right_tuple = checkAndGetDataType<DataTypeTuple>(arguments[1].type.get());
const auto & left_types = left_tuple->getElements();
const auto & right_types = right_tuple->getElements();
auto left_elements = getTupleElements(*arguments[0].column);
auto right_elements = getTupleElements(*arguments[1].column);
size_t tuple_size = left_elements.size();
if (tuple_size == 0)
return DataTypeUInt8().createColumnConstWithDefaultValue(input_rows_count);
2021-08-25 14:02:04 +00:00
auto multiply = FunctionFactory::instance().get("multiply", context);
auto plus = FunctionFactory::instance().get("plus", context);
ColumnWithTypeAndName res;
2021-08-23 13:41:20 +00:00
for (size_t i = 0; i < tuple_size; ++i)
{
ColumnWithTypeAndName left{left_elements[i], left_types[i], {}};
ColumnWithTypeAndName right{right_elements[i], right_types[i], {}};
2021-08-25 14:02:04 +00:00
auto elem_multiply = multiply->build(ColumnsWithTypeAndName{left, right});
2021-08-23 13:41:20 +00:00
2021-08-25 14:02:04 +00:00
ColumnWithTypeAndName column;
column.type = elem_multiply->getResultType();
column.column = elem_multiply->execute({left, right}, column.type, input_rows_count);
2021-08-23 13:41:20 +00:00
2021-08-25 14:02:04 +00:00
if (i == 0)
2021-08-23 13:41:20 +00:00
{
2021-08-25 14:02:04 +00:00
res = std::move(column);
2021-08-23 13:41:20 +00:00
}
2021-08-25 14:02:04 +00:00
else
2021-08-23 13:41:20 +00:00
{
2021-08-25 14:02:04 +00:00
auto plus_elem = plus->build({res, column});
auto res_type = plus_elem->getResultType();
res.column = plus_elem->execute({res, column}, res_type, input_rows_count);
res.type = res_type;
2021-08-23 13:41:20 +00:00
}
}
2021-08-25 14:02:04 +00:00
return res.column;
2021-08-23 13:41:20 +00:00
}
};
2021-08-30 12:52:00 +00:00
/// this is for convenient usage in LNormalize
template <const char * func_label>
class FunctionLNorm : public TupleIFunction {};
template <>
class FunctionLNorm<L1_LABEL> : public TupleIFunction
2021-08-25 14:56:31 +00:00
{
public:
static constexpr auto name = "L1Norm";
2021-08-30 12:52:00 +00:00
explicit FunctionLNorm(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionLNorm>(context_); }
2021-08-25 14:56:31 +00:00
String getName() const override { return name; }
size_t getNumberOfArguments() const override { return 1; }
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
if (!cur_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 0 of function {} should be tuples, got {}",
getName(), arguments[0].type->getName());
const auto & cur_types = cur_tuple->getElements();
Columns cur_elements;
if (arguments[0].column)
cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_types.size();
if (tuple_size == 0)
return std::make_shared<DataTypeUInt8>();
auto abs = FunctionFactory::instance().get("abs", context);
auto plus = FunctionFactory::instance().get("plus", context);
DataTypePtr res_type;
for (size_t i = 0; i < tuple_size; ++i)
{
try
{
ColumnWithTypeAndName cur{cur_elements.empty() ? nullptr : cur_elements[i], cur_types[i], {}};
auto elem_abs = abs->build(ColumnsWithTypeAndName{cur});
if (i == 0)
{
res_type = elem_abs->getResultType();
continue;
}
ColumnWithTypeAndName left_type{res_type, {}};
ColumnWithTypeAndName right_type{elem_abs->getResultType(), {}};
auto plus_elem = plus->build({left_type, right_type});
res_type = plus_elem->getResultType();
}
catch (DB::Exception & e)
{
e.addMessage("While executing function {} for tuple element {}", getName(), i);
throw;
}
}
return res_type;
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto & cur_types = cur_tuple->getElements();
auto cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_elements.size();
if (tuple_size == 0)
return DataTypeUInt8().createColumnConstWithDefaultValue(input_rows_count);
auto abs = FunctionFactory::instance().get("abs", context);
auto plus = FunctionFactory::instance().get("plus", context);
ColumnWithTypeAndName res;
for (size_t i = 0; i < tuple_size; ++i)
{
ColumnWithTypeAndName cur{cur_elements[i], cur_types[i], {}};
auto elem_abs = abs->build(ColumnsWithTypeAndName{cur});
ColumnWithTypeAndName column;
column.type = elem_abs->getResultType();
column.column = elem_abs->execute({cur}, column.type, input_rows_count);
if (i == 0)
{
res = std::move(column);
}
else
{
auto plus_elem = plus->build({res, column});
auto res_type = plus_elem->getResultType();
res.column = plus_elem->execute({res, column}, res_type, input_rows_count);
res.type = res_type;
}
}
return res.column;
}
};
2021-08-30 12:52:00 +00:00
using FunctionL1Norm = FunctionLNorm<L1_LABEL>;
2021-08-25 14:56:31 +00:00
2021-08-30 12:52:00 +00:00
template <>
class FunctionLNorm<L2_LABEL> : public TupleIFunction
2021-08-25 15:44:35 +00:00
{
public:
static constexpr auto name = "L2Norm";
2021-08-30 12:52:00 +00:00
explicit FunctionLNorm(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionLNorm>(context_); }
2021-08-25 15:44:35 +00:00
String getName() const override { return name; }
size_t getNumberOfArguments() const override { return 1; }
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
if (!cur_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 0 of function {} should be tuples, got {}",
getName(), arguments[0].type->getName());
const auto & cur_types = cur_tuple->getElements();
Columns cur_elements;
if (arguments[0].column)
cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_types.size();
if (tuple_size == 0)
return std::make_shared<DataTypeUInt8>();
auto multiply = FunctionFactory::instance().get("multiply", context);
auto plus = FunctionFactory::instance().get("plus", context);
DataTypePtr res_type;
for (size_t i = 0; i < tuple_size; ++i)
{
try
{
ColumnWithTypeAndName cur{cur_elements.empty() ? nullptr : cur_elements[i], cur_types[i], {}};
auto elem_multiply = multiply->build(ColumnsWithTypeAndName{cur, cur});
if (i == 0)
{
res_type = elem_multiply->getResultType();
continue;
}
ColumnWithTypeAndName left_type{res_type, {}};
ColumnWithTypeAndName right_type{elem_multiply->getResultType(), {}};
auto plus_elem = plus->build({left_type, right_type});
res_type = plus_elem->getResultType();
}
catch (DB::Exception & e)
{
e.addMessage("While executing function {} for tuple element {}", getName(), i);
throw;
}
}
auto sqrt = FunctionFactory::instance().get("sqrt", context);
return sqrt->build({ColumnWithTypeAndName{res_type, {}}})->getResultType();
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto & cur_types = cur_tuple->getElements();
auto cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_elements.size();
if (tuple_size == 0)
return DataTypeUInt8().createColumnConstWithDefaultValue(input_rows_count);
auto multiply = FunctionFactory::instance().get("multiply", context);
auto plus = FunctionFactory::instance().get("plus", context);
ColumnWithTypeAndName res;
for (size_t i = 0; i < tuple_size; ++i)
{
ColumnWithTypeAndName cur{cur_elements[i], cur_types[i], {}};
auto elem_multiply = multiply->build(ColumnsWithTypeAndName{cur, cur});
ColumnWithTypeAndName column;
column.type = elem_multiply->getResultType();
column.column = elem_multiply->execute({cur, cur}, column.type, input_rows_count);
if (i == 0)
{
res = std::move(column);
}
else
{
auto plus_elem = plus->build({res, column});
auto res_type = plus_elem->getResultType();
res.column = plus_elem->execute({res, column}, res_type, input_rows_count);
res.type = res_type;
}
}
auto sqrt = FunctionFactory::instance().get("sqrt", context);
auto sqrt_elem = sqrt->build({res});
return sqrt_elem->execute({res}, sqrt_elem->getResultType(), input_rows_count);
}
};
2021-08-30 12:52:00 +00:00
using FunctionL2Norm = FunctionLNorm<L2_LABEL>;
2021-08-25 15:44:35 +00:00
2021-08-30 12:52:00 +00:00
template <>
class FunctionLNorm<Linf_LABEL> : public TupleIFunction
2021-08-26 12:34:46 +00:00
{
public:
static constexpr auto name = "LinfNorm";
2021-08-30 12:52:00 +00:00
explicit FunctionLNorm(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionLNorm>(context_); }
2021-08-26 12:34:46 +00:00
String getName() const override { return name; }
size_t getNumberOfArguments() const override { return 1; }
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
if (!cur_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 0 of function {} should be tuples, got {}",
getName(), arguments[0].type->getName());
const auto & cur_types = cur_tuple->getElements();
Columns cur_elements;
if (arguments[0].column)
cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_types.size();
if (tuple_size == 0)
return std::make_shared<DataTypeUInt8>();
auto abs = FunctionFactory::instance().get("abs", context);
auto max = FunctionFactory::instance().get("max2", context);
DataTypePtr res_type;
for (size_t i = 0; i < tuple_size; ++i)
{
try
{
ColumnWithTypeAndName cur{cur_elements.empty() ? nullptr : cur_elements[i], cur_types[i], {}};
auto elem_abs = abs->build(ColumnsWithTypeAndName{cur});
if (i == 0)
{
res_type = elem_abs->getResultType();
continue;
}
ColumnWithTypeAndName left_type{res_type, {}};
ColumnWithTypeAndName right_type{elem_abs->getResultType(), {}};
auto max_elem = max->build({left_type, right_type});
res_type = max_elem->getResultType();
}
catch (DB::Exception & e)
{
e.addMessage("While executing function {} for tuple element {}", getName(), i);
throw;
}
}
return res_type;
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto & cur_types = cur_tuple->getElements();
auto cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_elements.size();
if (tuple_size == 0)
return DataTypeUInt8().createColumnConstWithDefaultValue(input_rows_count);
auto abs = FunctionFactory::instance().get("abs", context);
auto max = FunctionFactory::instance().get("max2", context);
ColumnWithTypeAndName res;
for (size_t i = 0; i < tuple_size; ++i)
{
ColumnWithTypeAndName cur{cur_elements[i], cur_types[i], {}};
auto elem_abs = abs->build(ColumnsWithTypeAndName{cur});
ColumnWithTypeAndName column;
column.type = elem_abs->getResultType();
column.column = elem_abs->execute({cur}, column.type, input_rows_count);
if (i == 0)
{
res = std::move(column);
}
else
{
auto max_elem = max->build({res, column});
auto res_type = max_elem->getResultType();
res.column = max_elem->execute({res, column}, res_type, input_rows_count);
res.type = res_type;
}
}
return res.column;
}
};
2021-08-30 12:52:00 +00:00
using FunctionLinfNorm = FunctionLNorm<Linf_LABEL>;
2021-08-26 12:34:46 +00:00
2021-08-30 12:52:00 +00:00
template <>
class FunctionLNorm<Lp_LABEL> : public TupleIFunction
2021-08-26 18:17:44 +00:00
{
public:
static constexpr auto name = "LpNorm";
2021-08-30 12:52:00 +00:00
explicit FunctionLNorm(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionLNorm>(context_); }
2021-08-26 18:17:44 +00:00
String getName() const override { return name; }
size_t getNumberOfArguments() const override { return 2; }
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
if (!cur_tuple)
throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "Argument 0 of function {} should be tuples, got {}",
getName(), arguments[0].type->getName());
const auto & cur_types = cur_tuple->getElements();
Columns cur_elements;
if (arguments[0].column)
cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_types.size();
if (tuple_size == 0)
return std::make_shared<DataTypeUInt8>();
const auto & p_column = arguments[1];
auto abs = FunctionFactory::instance().get("abs", context);
auto pow = FunctionFactory::instance().get("pow", context);
auto plus = FunctionFactory::instance().get("plus", context);
DataTypePtr res_type;
for (size_t i = 0; i < tuple_size; ++i)
{
try
{
ColumnWithTypeAndName cur{cur_elements.empty() ? nullptr : cur_elements[i], cur_types[i], {}};
auto elem_abs = abs->build(ColumnsWithTypeAndName{cur});
cur.column = elem_abs->execute({cur}, elem_abs->getResultType(), 1);
cur.type = elem_abs->getResultType();
auto elem_pow = pow->build(ColumnsWithTypeAndName{cur, p_column});
if (i == 0)
{
res_type = elem_pow->getResultType();
continue;
}
ColumnWithTypeAndName left_type{res_type, {}};
ColumnWithTypeAndName right_type{elem_pow->getResultType(), {}};
auto plus_elem = plus->build({left_type, right_type});
res_type = plus_elem->getResultType();
}
catch (DB::Exception & e)
{
e.addMessage("While executing function {} for tuple element {}", getName(), i);
throw;
}
}
auto divide = FunctionFactory::instance().get("divide", context);
ColumnWithTypeAndName one{DataTypeFloat64().createColumnConst(1, 1.), std::make_shared<DataTypeFloat64>(), {}};
auto div_elem = divide->build({one, p_column});
ColumnWithTypeAndName inv_p_column;
inv_p_column.type = div_elem->getResultType();
inv_p_column.column = div_elem->execute({one, p_column}, inv_p_column.type, 1);
return pow->build({ColumnWithTypeAndName{res_type, {}}, inv_p_column})->getResultType();
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
const auto * cur_tuple = checkAndGetDataType<DataTypeTuple>(arguments[0].type.get());
const auto & cur_types = cur_tuple->getElements();
auto cur_elements = getTupleElements(*arguments[0].column);
size_t tuple_size = cur_elements.size();
if (tuple_size == 0)
return DataTypeUInt8().createColumnConstWithDefaultValue(input_rows_count);
const auto & p_column = arguments[1];
auto abs = FunctionFactory::instance().get("abs", context);
auto pow = FunctionFactory::instance().get("pow", context);
auto plus = FunctionFactory::instance().get("plus", context);
ColumnWithTypeAndName res;
for (size_t i = 0; i < tuple_size; ++i)
{
ColumnWithTypeAndName cur{cur_elements[i], cur_types[i], {}};
auto elem_abs = abs->build(ColumnsWithTypeAndName{cur});
cur.column = elem_abs->execute({cur}, elem_abs->getResultType(), input_rows_count);
cur.type = elem_abs->getResultType();
auto elem_pow = pow->build(ColumnsWithTypeAndName{cur, p_column});
ColumnWithTypeAndName column;
column.type = elem_pow->getResultType();
column.column = elem_pow->execute({cur, p_column}, column.type, input_rows_count);
if (i == 0)
{
res = std::move(column);
}
else
{
auto plus_elem = plus->build({res, column});
auto res_type = plus_elem->getResultType();
res.column = plus_elem->execute({res, column}, res_type, input_rows_count);
res.type = res_type;
}
}
auto divide = FunctionFactory::instance().get("divide", context);
ColumnWithTypeAndName one{DataTypeFloat64().createColumnConst(input_rows_count, 1.)->convertToFullColumnIfConst(), std::make_shared<DataTypeFloat64>(), {}};
auto div_elem = divide->build({one, p_column});
ColumnWithTypeAndName inv_p_column;
inv_p_column.type = div_elem->getResultType();
inv_p_column.column = div_elem->execute({one, p_column}, inv_p_column.type, input_rows_count);
auto pow_elem = pow->build({res, inv_p_column});
return pow_elem->execute({res, inv_p_column}, pow_elem->getResultType(), input_rows_count);
}
};
2021-08-30 12:52:00 +00:00
using FunctionLpNorm = FunctionLNorm<Lp_LABEL>;
2021-08-26 18:17:44 +00:00
2021-08-27 13:14:32 +00:00
template <const char * func_label>
2021-08-30 12:52:00 +00:00
class FunctionLDistance : public TupleIFunction
2021-08-27 13:14:32 +00:00
{
public:
2021-08-27 14:31:11 +00:00
/// constexpr cannot be used due to std::string has not constexpr constructor in this compiler version
static inline auto name = "L" + std::string(func_label) + "Distance";
2021-08-30 12:52:00 +00:00
explicit FunctionLDistance(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionLDistance>(context_); }
2021-08-27 14:31:11 +00:00
String getName() const override { return name; }
2021-08-27 13:14:32 +00:00
size_t getNumberOfArguments() const override
{
if constexpr (func_label[0] == 'p')
return 3;
else
return 2;
}
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
FunctionTupleMinus tuple_minus(context);
auto type = tuple_minus.getReturnTypeImpl(arguments);
auto column = tuple_minus.executeImpl(arguments, DataTypePtr(), 1);
ColumnWithTypeAndName minus_res{column, type, {}};
auto func = FunctionFactory::instance().get("L" + std::string(func_label) + "Norm", context);
if constexpr (func_label[0] == 'p')
return func->build({minus_res, arguments[2]})->getResultType();
else
return func->build({minus_res})->getResultType();
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
FunctionTupleMinus tuple_minus(context);
auto type = tuple_minus.getReturnTypeImpl(arguments);
auto column = tuple_minus.executeImpl(arguments, DataTypePtr(), input_rows_count);
ColumnWithTypeAndName minus_res{column, type, {}};
auto func = FunctionFactory::instance().get("L" + std::string(func_label) + "Norm", context);
if constexpr (func_label[0] == 'p')
{
auto func_elem = func->build({minus_res, arguments[2]});
return func_elem->execute({minus_res, arguments[2]}, func_elem->getResultType(), input_rows_count);
}
else
{
auto func_elem = func->build({minus_res});
return func_elem->execute({minus_res}, func_elem->getResultType(), input_rows_count);
}
}
};
2021-08-30 12:52:00 +00:00
using FunctionL1Distance = FunctionLDistance<L1_LABEL>;
using FunctionL2Distance = FunctionLDistance<L2_LABEL>;
using FunctionLinfDistance = FunctionLDistance<Linf_LABEL>;
using FunctionLpDistance = FunctionLDistance<Lp_LABEL>;
template <const char * func_label>
class FunctionLNormalize : public TupleIFunction
{
public:
/// constexpr cannot be used due to std::string has not constexpr constructor in this compiler version
static inline auto name = "L" + std::string(func_label) + "Normalize";
explicit FunctionLNormalize(ContextPtr context_) : TupleIFunction(context_) {}
static FunctionPtr create(ContextPtr context_) { return std::make_shared<FunctionLNormalize>(context_); }
String getName() const override { return name; }
size_t getNumberOfArguments() const override
{
if constexpr (func_label[0] == 'p')
return 2;
else
return 1;
}
DataTypePtr getReturnTypeImpl(const ColumnsWithTypeAndName & arguments) const override
{
FunctionLNorm<func_label> norm(context);
auto type = norm.getReturnTypeImpl(arguments);
auto column = norm.executeImpl(arguments, DataTypePtr(), 1);
ColumnWithTypeAndName norm_res{column, type, {}};
FunctionTupleDivideByNumber divide(context);
return divide.getReturnTypeImpl({arguments[0], norm_res});
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const override
{
FunctionLNorm<func_label> norm(context);
auto type = norm.getReturnTypeImpl(arguments);
auto column = norm.executeImpl(arguments, DataTypePtr(), input_rows_count);
ColumnWithTypeAndName norm_res{column, type, {}};
FunctionTupleDivideByNumber divide(context);
return divide.executeImpl({arguments[0], norm_res}, DataTypePtr(), input_rows_count);
}
};
using FunctionL1Normalize = FunctionLNormalize<L1_LABEL>;
2021-08-27 13:14:32 +00:00
2021-08-30 12:52:00 +00:00
using FunctionL2Normalize = FunctionLNormalize<L2_LABEL>;
2021-08-27 13:14:32 +00:00
2021-08-30 12:52:00 +00:00
using FunctionLinfNormalize = FunctionLNormalize<Linf_LABEL>;
2021-08-27 13:14:32 +00:00
2021-08-30 12:52:00 +00:00
using FunctionLpNormalize = FunctionLNormalize<Lp_LABEL>;
2021-08-27 13:14:32 +00:00
void registerVectorFunctions(FunctionFactory & factory)
{
factory.registerFunction<FunctionTuplePlus>();
factory.registerAlias("vectorSum", FunctionTuplePlus::name, FunctionFactory::CaseInsensitive);
factory.registerFunction<FunctionTupleMinus>();
factory.registerAlias("vectorDifference", FunctionTupleMinus::name, FunctionFactory::CaseInsensitive);
factory.registerFunction<FunctionTupleMultiply>();
2021-08-23 13:41:20 +00:00
factory.registerFunction<FunctionTupleDivide>();
factory.registerFunction<FunctionTupleNegate>();
2021-08-27 15:44:04 +00:00
factory.registerFunction<FunctionTupleMultiplyByNumber>();
factory.registerFunction<FunctionTupleDivideByNumber>();
2021-08-23 13:41:20 +00:00
factory.registerFunction<FunctionDotProduct>();
factory.registerAlias("scalarProduct", FunctionDotProduct::name, FunctionFactory::CaseInsensitive);
2021-08-25 14:56:31 +00:00
factory.registerFunction<FunctionL1Norm>();
2021-08-25 15:44:35 +00:00
factory.registerFunction<FunctionL2Norm>();
2021-08-26 12:34:46 +00:00
factory.registerFunction<FunctionLinfNorm>();
2021-08-26 18:17:44 +00:00
factory.registerFunction<FunctionLpNorm>();
2021-08-27 13:14:32 +00:00
factory.registerFunction<FunctionL1Distance>();
2021-08-23 13:41:20 +00:00
factory.registerFunction<FunctionL2Distance>();
factory.registerFunction<FunctionLinfDistance>();
factory.registerFunction<FunctionLpDistance>();
2021-08-30 12:52:00 +00:00
factory.registerFunction<FunctionL1Normalize>();
factory.registerFunction<FunctionL2Normalize>();
factory.registerFunction<FunctionLinfNormalize>();
factory.registerFunction<FunctionLpNormalize>();
2021-08-27 13:14:32 +00:00
//
// factory.registerFunction<FunctionCosineDistance>();
}
}