ClickHouse/src/AggregateFunctions/AggregateFunctionTopK.cpp

128 lines
4.7 KiB
C++
Raw Normal View History

#include <AggregateFunctions/AggregateFunctionFactory.h>
#include <AggregateFunctions/AggregateFunctionTopK.h>
#include <AggregateFunctions/Helpers.h>
#include <AggregateFunctions/FactoryHelpers.h>
2021-05-03 22:46:51 +00:00
#include <Common/FieldVisitors.h>
#include <DataTypes/DataTypeDate.h>
#include <DataTypes/DataTypeDateTime.h>
2021-05-03 22:46:51 +00:00
#define TOP_K_MAX_SIZE 0xFFFFFF
namespace DB
{
namespace ErrorCodes
{
extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH;
extern const int ARGUMENT_OUT_OF_BOUND;
extern const int ILLEGAL_TYPE_OF_ARGUMENT;
}
namespace
{
/// Substitute return type for Date and DateTime
template <bool is_weighted>
class AggregateFunctionTopKDate : public AggregateFunctionTopK<DataTypeDate::FieldType, is_weighted>
{
using AggregateFunctionTopK<DataTypeDate::FieldType, is_weighted>::AggregateFunctionTopK;
DataTypePtr getReturnType() const override { return std::make_shared<DataTypeArray>(std::make_shared<DataTypeDate>()); }
};
template <bool is_weighted>
class AggregateFunctionTopKDateTime : public AggregateFunctionTopK<DataTypeDateTime::FieldType, is_weighted>
{
using AggregateFunctionTopK<DataTypeDateTime::FieldType, is_weighted>::AggregateFunctionTopK;
DataTypePtr getReturnType() const override { return std::make_shared<DataTypeArray>(std::make_shared<DataTypeDateTime>()); }
};
template <bool is_weighted>
2019-03-10 21:32:42 +00:00
static IAggregateFunction * createWithExtraTypes(const DataTypePtr & argument_type, UInt64 threshold, UInt64 load_factor, const Array & params)
{
2018-09-10 17:09:07 +00:00
WhichDataType which(argument_type);
if (which.idx == TypeIndex::Date)
2019-03-10 21:32:42 +00:00
return new AggregateFunctionTopKDate<is_weighted>(threshold, load_factor, {argument_type}, params);
if (which.idx == TypeIndex::DateTime)
2019-03-10 21:32:42 +00:00
return new AggregateFunctionTopKDateTime<is_weighted>(threshold, load_factor, {argument_type}, params);
/// Check that we can use plain version of AggregateFunctionTopKGeneric
if (argument_type->isValueUnambiguouslyRepresentedInContiguousMemoryRegion())
2019-03-10 21:32:42 +00:00
return new AggregateFunctionTopKGeneric<true, is_weighted>(threshold, load_factor, argument_type, params);
else
2019-03-10 21:32:42 +00:00
return new AggregateFunctionTopKGeneric<false, is_weighted>(threshold, load_factor, argument_type, params);
}
template <bool is_weighted>
AggregateFunctionPtr createAggregateFunctionTopK(const std::string & name, const DataTypes & argument_types, const Array & params)
{
if (!is_weighted)
{
2019-02-03 10:59:27 +00:00
assertUnary(name, argument_types);
}
2019-02-03 10:59:27 +00:00
else
{
2019-02-03 10:59:27 +00:00
assertBinary(name, argument_types);
if (!isInteger(argument_types[1]))
throw Exception("The second argument for aggregate function 'topKWeighted' must have integer type", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
}
UInt64 threshold = 10; /// default values
2019-03-10 21:32:42 +00:00
UInt64 load_factor = 3;
if (!params.empty())
{
if (params.size() > 2)
throw Exception("Aggregate function " + name + " requires two parameters or less.",
2019-02-03 10:59:27 +00:00
ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
UInt64 k = applyVisitor(FieldVisitorConvertToNumber<UInt64>(), params[0]);
2019-03-10 21:32:42 +00:00
if (params.size() == 2)
{
load_factor = applyVisitor(FieldVisitorConvertToNumber<UInt64>(), params[1]);
2019-03-10 21:32:42 +00:00
if (load_factor < 1)
throw Exception("Too small parameter 'load_factor' for aggregate function " + name + ". Minimum: 1",
ErrorCodes::ARGUMENT_OUT_OF_BOUND);
}
if (k > TOP_K_MAX_SIZE || load_factor > TOP_K_MAX_SIZE || k * load_factor > TOP_K_MAX_SIZE)
throw Exception("Too large parameter(s) for aggregate function " + name + ". Maximum: " + toString(TOP_K_MAX_SIZE),
ErrorCodes::ARGUMENT_OUT_OF_BOUND);
if (k == 0)
throw Exception("Parameter 0 is illegal for aggregate function " + name,
ErrorCodes::ARGUMENT_OUT_OF_BOUND);
threshold = k;
}
2020-06-29 23:17:20 +00:00
AggregateFunctionPtr res(createWithNumericType<AggregateFunctionTopK, is_weighted>(
*argument_types[0], threshold, load_factor, argument_types, params));
if (!res)
2019-03-10 21:32:42 +00:00
res = AggregateFunctionPtr(createWithExtraTypes<is_weighted>(argument_types[0], threshold, load_factor, params));
if (!res)
throw Exception("Illegal type " + argument_types[0]->getName() +
" of argument for aggregate function " + name, ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
return res;
}
}
void registerAggregateFunctionTopK(AggregateFunctionFactory & factory)
{
AggregateFunctionProperties properties = { .returns_default_when_only_null = false, .is_order_dependent = true };
factory.registerFunction("topK", { createAggregateFunctionTopK<false>, properties });
factory.registerFunction("topKWeighted", { createAggregateFunctionTopK<true>, properties });
}
}