kev/Drawer/DataValidator/Rule/Data/MCDValidationRule.cs

using MathNet.Numerics.LinearAlgebra;
using MathNet.Numerics.LinearAlgebra.Double;
using System;
using System.Collections.Generic;
using System.Data;
using System.Diagnostics;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
using Validation.Algorithms;
using Validation.Core;

namespace Validation.Rule.Data
{
    /// <summary>
    /// 
    /// </summary>
    public class MCDValidationRule : IValidationRule
    {
        string IValidationRule.RuleName => "MCD";

        string IValidationRule.Description => "检测 x y z 是否有异常值";

        private readonly string xColumn;

        private readonly string yColumn;

        private readonly string zColumn;

        public MCDValidationRule(string xColumn, string yColumn, string zColumn)
        {
            this.xColumn = xColumn;
            this.yColumn = yColumn;
            this.zColumn = zColumn;
        }

        ValidationResult IValidationRule.Validate(DataTable dataTable)
        {
            if (!ValidateRequiredColumns(dataTable))
            {
                var result = new ValidationResult();
                result.AddError(-1, "x y z 列都不能为空");
                return result;
            }

            var dataMatrix = BuildDataMatrix(dataTable);
            Debug.Assert(dataMatrix.RowCount >= 3);

            var calculator = new RobustMahalanobis(numRandomStarts: 100);
            var distances = calculator.Calculate(dataMatrix);

            var outliers = IdentifyOutliers(distances);

            return BuildValidationResult(outliers, distances);
        }

        /// <summary>
        /// 验证必需的列是否存在
        /// </summary>
        private bool ValidateRequiredColumns(DataTable dataTable)
        {
            return dataTable.Columns.Contains(xColumn)
                && dataTable.Columns.Contains(yColumn)
                && dataTable.Columns.Contains(zColumn);
        }

        bool IsValidRow(DataRow row)
        {
            var columns = new[] { xColumn, yColumn, zColumn };

            foreach (var col in columns)
            {
                var value = row[col];

                if (value == DBNull.Value)
                {
                    return false;
                }

                if (!double.TryParse(value.ToString(), out _))
                {
                    return false;
                }
            }

            return true;
        }

        /// <summary>
        /// 构建数据矩阵
        /// </summary>
        private Matrix<double> BuildDataMatrix(DataTable dataTable)
        {
            var columns = new[] { xColumn, yColumn, zColumn };

            // 确定有效列
            var validRows = dataTable.AsEnumerable()
                .Where(IsValidRow)
                .ToList();

            if (validRows.Count == 0)
            {
                return DenseMatrix.Create(0, columns.Length, 0.0);
            }

            var matrix = DenseMatrix.Create(validRows.Count, columns.Length, 0.0);

            for (int i = 0; i < validRows.Count; i++)
            {
                for (int j = 0; j < columns.Length; j++)
                {
                    matrix[i, j] = Convert.ToDouble(validRows[i][columns[j]]);
                }
            }

            return matrix;
        }

        private int[] IdentifyOutliers(double[] distances)
        {
            if (distances.Length == 0)
            {
                return Array.Empty<int>();
            }

            double threshold = Math.Sqrt(7.81);
            var outliers = distances
                .Select((d, idx) => new { d, idx })
                .Where(x => x.d > threshold)
                .Select(x => x.idx)
                .ToArray();

            return outliers;
        }

        private ValidationResult BuildValidationResult(int[] outlierIndices, double[] distances)
        { 
            var result = new ValidationResult();

            for (int i = 0; i < outlierIndices.Length; i++)
            {
                result.AddError(outlierIndices[i], $"行 {i} 为异常值");
            }

            return result;
        }

        public ValidationResult Validate(string[] headers, int rowIndex, object[] values)
        {
            throw new NotImplementedException();
        }
    }
}