Calculation performance of Deedle Frame significantly slower than that of Series?

Question

I am currently evaluating Deedle for my project, and regarding that I have done some rudimentary performance tests. I have attached the code I used to do this. Basically it's doing two operations:

generate two 1000x10 data frames and multiplying with each other once
generate two series (length 1000) series and multiply them with each other 10 times

Execution time of these operations (only the calculation, not the generation) is measured with StopWatch. I would expect the execution time of these operations to be somewhat similar. However, on my machine the execution time for series ~10ms and for data frames it's ~200ms, so calculating with data frames is ~20x slower. I'm running this in .NET Core 2.1, Deedle 2.0.0-beta01, FSharp.Core 4.5.2, I got similar results with Deedle 1.2.5 as well.

Is there something that I'm doing wrong or is this just an issue with the library or its C# interface, or could there be some other reason for this?

using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Globalization;
using System.IO;
using System.Linq;
using System.Threading.Tasks;
using Deedle;

namespace DeedleTest
{
    class Program
    {
        static void Main(string[] args)
        {
            const int countCols = 10;
            const int countRows = 1000;

            Frame<DateTime, int> df1 = GenerateFrame(countCols, countRows, false);
            Frame<DateTime, int> df2 = GenerateFrame(countCols, countRows, false);

            var stopwatch = Stopwatch.StartNew();

            df1 = df1 * df2;

            stopwatch.Stop();

            Console.WriteLine(stopwatch.ElapsedMilliseconds);

            Series<DateTime, double> ser1 = GenerateSeries(countRows);
            Series<DateTime, double> ser2 = GenerateSeries(countRows);

            stopwatch.Reset();
            stopwatch.Start();

            foreach (int i in Enumerable.Range(0, countCols))
            {
                ser1 = ser1 * ser2;
            }

            stopwatch.Stop();

            Console.WriteLine(stopwatch.ElapsedMilliseconds);

            Console.ReadKey();
        }

        private static Frame<DateTime, int> GenerateFrame(int countCols, int countRows, bool randomNumbers = true)
        {
            var seriesList = new List<Series<DateTime, double>>();

            foreach (int col in Enumerable.Range(0, countCols))
            {
                Series<DateTime, double> series = GenerateSeries(countRows, randomNumbers ? null : (double?)col);

                seriesList.Add(series);
            }
            return Frame.FromColumns(seriesList);
        }

        private static Series<DateTime, double> GenerateSeries(int countRows, double? number = null)
        {
            var randgen = new Random();
            var startDate = DateTime.Now;

            var builder = new SeriesBuilder<DateTime, double>();
            foreach (int row in Enumerable.Range(0, countRows))
            {
                builder.Add(startDate.AddSeconds(row), number == null ? randgen.NextDouble() : (double)number);
            }
            return builder.Series;
        }
    }
}

Calculation performance of Deedle Frame significantly slower than that of Series?

0 Answers0