NVIDIA Morpheus (24.10.01)
(Latest Version)

Program Listing for File table_info.cpp

↰ Return to documentation for file (morpheus/_lib/src/objects/table_info.cpp)

Copy
Copied!
            

/* * SPDX-FileCopyrightText: Copyright (c) 2021-2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved. * SPDX-License-Identifier: Apache-2.0 * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ #include "morpheus/objects/table_info.hpp" #include "morpheus/objects/dtype.hpp" #include <cudf/copying.hpp> #include <cudf/table/table_view.hpp> #include <cudf/types.hpp> #include <glog/logging.h> #include <pybind11/gil.h> // for gil_scoped_acquire #include <pybind11/pybind11.h> // IWYU pragma: keep #include <pybind11/pytypes.h> #include <pybind11/stl.h> // IWYU pragma: keep #include <algorithm> // for find, transform #include <cstddef> // for size_t #include <iterator> // for back_insert_iterator, back_inserter #include <memory> #include <optional> #include <ostream> #include <shared_mutex> #include <stdexcept> #include <utility> // IWYU pragma: no_include <pybind11/cast.h> namespace morpheus { namespace py = pybind11; using namespace py::literals; TableInfoData get_table_info_data_slice(const TableInfoData& table, cudf::size_type start, cudf::size_type stop, std::vector<std::string> column_names) { CHECK_GE(start, 0) << "Start must be >= 0"; if (stop < 0) { stop = table.table_view.num_rows(); } CHECK_GT(stop, 0) << "Stop must be > 0"; CHECK_LE(stop, table.table_view.num_rows()) << "Stop must be less than the number of rows"; CHECK_LE(start, stop) << "Start must be less than stop"; if (column_names.empty()) { column_names = table.column_names; } // Start with our table view auto table_view_out = table.table_view; std::vector<cudf::size_type> col_indices; std::vector<cudf::size_type> col_indices_mapppings; // If the columns are different, calculate the new slice if (column_names != table.column_names) { // Append the indices column idx by default for (cudf::size_type i = 0; i < table.index_names.size(); ++i) { col_indices.push_back(i); } std::transform(column_names.begin(), column_names.end(), std::back_inserter(col_indices), [&table, &col_indices_mapppings](const std::string& c) { auto found_col = std::find(table.column_names.begin(), table.column_names.end(), c); if (found_col == table.column_names.end()) { throw py::key_error("Unknown column: " + c); } auto idx = (found_col - table.column_names.begin()); col_indices_mapppings.push_back(idx); return idx + table.index_names.size(); }); table_view_out = table_view_out.select(col_indices); } // If the start/stop is different, then perform the slice if (start != 0 || stop != table.table_view.num_rows()) { table_view_out = cudf::slice(table_view_out, {start, stop})[0]; } if (col_indices_mapppings.size() > 0) { // Create a new TableInfoData return {table_view_out, table.index_names, column_names, col_indices_mapppings}; } // Create a new TableInfoData return {table_view_out, table.index_names, column_names}; } /****** Component public implementations *******************/ /****** TableInfoBase****************************************/ TableInfoBase::TableInfoBase(std::shared_ptr<const IDataTable> parent, TableInfoData data) : m_parent(std::move(parent)), m_data(std::move(data)) {} const cudf::table_view& TableInfoBase::get_view() const { return m_data.table_view; } std::vector<std::string> TableInfoBase::get_index_names() const { return m_data.index_names; } std::vector<std::string> TableInfoBase::get_column_names() const { return m_data.column_names; } cudf::size_type TableInfoBase::num_indices() const { return this->get_index_names().size(); } cudf::size_type TableInfoBase::num_columns() const { return this->get_column_names().size(); } cudf::size_type TableInfoBase::num_rows() const { return m_data.table_view.num_rows(); } const cudf::column_view& TableInfoBase::get_column(cudf::size_type idx) const { if (idx < 0 || idx >= this->m_data.table_view.num_columns()) { throw std::invalid_argument("idx must satisfy 0 <= idx < num_columns()"); } return this->m_data.table_view.column(this->m_data.index_names.size() + idx); } const std::shared_ptr<const IDataTable>& TableInfoBase::get_parent() const { return m_parent; } TableInfoData& TableInfoBase::get_data() { return m_data; } const TableInfoData& TableInfoBase::get_data() const { return m_data; } bool TableInfoBase::has_sliceable_index() const { py::gil_scoped_acquire gil; auto df = m_parent->get_py_object(); auto index = df.attr("index"); auto is_unique = index.attr("is_unique").cast<bool>(); auto is_monotonic_increasing = index.attr("is_monotonic_increasing").cast<bool>(); auto is_monotonic_decreasing = index.attr("is_monotonic_decreasing").cast<bool>(); // Must be either increasing or decreasing with unique values to slice return is_unique && (is_monotonic_increasing || is_monotonic_decreasing); } TableInfo::TableInfo(std::shared_ptr<const IDataTable> parent, std::shared_lock<std::shared_mutex> lock, TableInfoData data) : TableInfoBase(parent, std::move(data)), m_lock(std::move(lock)) {} TableInfo TableInfo::get_slice(cudf::size_type start, cudf::size_type stop, std::vector<std::string> column_names) const { // Create a new Table info, (cloning the shared_lock) return {this->get_parent(), std::shared_lock<std::shared_mutex>(*m_lock.mutex()), get_table_info_data_slice(this->get_data(), start, stop, column_names)}; } MutableTableInfo::MutableTableInfo(std::shared_ptr<const IDataTable> parent, std::unique_lock<std::shared_mutex> lock, TableInfoData data) : TableInfoBase(parent, std::move(data)), m_lock(std::move(lock)) {} MutableTableInfo::~MutableTableInfo() { if (m_checked_out_ref_count >= 0) { LOG(ERROR) << "Checked out python object was not returned before MutableTableInfo went out of scope"; } } MutableTableInfo MutableTableInfo::get_slice(cudf::size_type start, cudf::size_type stop, std::vector<std::string> column_names) && { // Create a new Table info, (moving the unique_lock) return { this->get_parent(), std::move(m_lock), get_table_info_data_slice(this->get_data(), start, stop, column_names)}; } void MutableTableInfo::insert_columns(const std::vector<std::tuple<std::string, morpheus::DType>>& columns) { const auto num_existing_cols = this->get_data().column_names.size(); const auto num_rows = this->get_data().table_view.num_rows(); // TODO(mdemoret): figure out how to do this without the gil { py::gil_scoped_acquire gil; py::object cudf_scalar = py::module_::import("cudf").attr("Scalar"); auto table = this->get_parent()->get_py_object(); for (std::size_t i = 0; i < columns.size(); ++i) { auto scalar = cudf_scalar(0, std::get<1>(columns[i]).type_str()); table.attr("insert")(num_existing_cols + i, std::get<0>(columns[i]), scalar); this->get_data().column_names.push_back(std::get<0>(columns[i])); } } } void MutableTableInfo::insert_missing_columns(const std::vector<std::tuple<std::string, morpheus::DType>>& columns) { std::vector<std::tuple<std::string, morpheus::DType>> missing_columns; for (const auto& column : columns) { if (std::find(this->get_data().column_names.begin(), this->get_data().column_names.end(), std::get<0>(column)) == this->get_data().column_names.end()) { missing_columns.push_back(column); } } if (!missing_columns.empty()) { insert_columns(missing_columns); } } std::unique_ptr<pybind11::object> MutableTableInfo::checkout_obj() { // Get a copy increasing the ref count py::object checked_out_obj = this->get_parent()->get_py_object(); m_checked_out_ref_count = checked_out_obj.ref_count(); auto ptr = std::make_unique<py::object>(std::move(checked_out_obj)); return ptr; } void MutableTableInfo::return_obj(std::unique_ptr<pybind11::object>&& obj) { obj.reset(nullptr); m_checked_out_ref_count = -1; } std::optional<std::string> MutableTableInfo::ensure_sliceable_index() { std::optional<std::string> old_index_col_name{"_index_"}; auto ptr_df = this->checkout_obj(); { py::gil_scoped_acquire gil; auto& py_df = *ptr_df; auto df_index = py_df.attr("index"); // Check to see if we actually need the change if (df_index.attr("is_unique").cast<bool>() && (df_index.attr("is_monotonic_increasing").cast<bool>() || df_index.attr("is_monotonic_decreasing").cast<bool>())) { // Set the outputname to nullopt old_index_col_name = std::nullopt; } else { auto index_name = df_index.attr("name"); if (!index_name.is_none()) { old_index_col_name = *old_index_col_name + index_name.cast<std::string>(); } df_index.attr("name") = py::str(*old_index_col_name); py_df.attr("reset_index")("inplace"_a = true); } } this->return_obj(std::move(ptr_df)); // If we made a change, update the index and column list if (old_index_col_name.has_value()) { auto& tbl_data = this->get_data(); tbl_data.column_names.insert(tbl_data.column_names.begin(), *old_index_col_name); tbl_data.index_names.clear(); tbl_data.index_names.emplace_back(""); } return old_index_col_name; } } // namespace morpheus

© Copyright 2024, NVIDIA. Last updated on Jul 8, 2024.