Continuous Walk-Forward Optimization (Part 1): Working with Optimization Reports
22 419
3
Introduction
In the previous articles (Optimization Management (Part I) and Optimization Management (Part 2)) we considered a mechanism for launching the optimization in the terminal through a third-party process. This allows creating a certain Optimization Manager which can implement the process similarly to a trading algorithm implementing a specific trading process, i.e. in a fully automated mode without user interference. The idea is to create an algorithm which manages the sliding optimization process, in which forward and historical periods are shifted by a preset interval and overlap each other.
This approach to algorithm optimization can serve as strategy robustness testing rather than pure optimization, although it performs both roles. As a result, we can find out whether a trading system is stable and can determine optimal combinations of indicators for the system. Since the described process can involve different robot coefficient filtration and optimal combination selection methods, which we need to check in each of the time intervals (which can be multiple) the process can hardly be implemented manually. Moreover, thus we can encounter errors connected with data transfer or other errors related to the human factor. Therefore, some tools are needed that would manage the optimization process from the outside without our intervention. The created program meets the set goals. For a more structured presentation, the program creation process has been split into several articles, each of which covers a specific area of the program creation process.
This part is devoted to the creation of a toolkit for working with optimization reports, for importing them from the terminal, as well as for filtering and sorting the obtained data. To provide a better presentation structure, we will use the *xml file format. Data from the file can be read by both humans and programs. Moreover, data can be grouped in blocks inside the file and thus the required information can be accessed faster and easier.
Our program is a third-party process written in C# and it needs to create and read created *xml documents similarly to MQL5 programs. Therefore, the report creation block will be implemented as a DLL which can be used both in MQL5 and C# code. Thus, in order to develop an MQL5 code, we will need a library. We will first describe the library creation process, while the next article will provide description of the MQL5 code that works with the created library and generates optimization parameters. We will consider these parameters in the current article.
Report Structure and Required Ratios
As already shown in previous articles, MetaTrader 5 can independently download the report of optimization passes, however it does not provide as much information as the report generated on the Backtest tab after completion of a test with a specific set of parameters. In order to have greater scope in working with optimization data, the report should include many of the data displayed on this tab, as well as provide for the possibility to add more custom data to the report. For these purposes, we will download our own generated reports instead of the standard one. Let's start with the definition of three data types required for our program:
- Tester settings (the same settings for the whole report)
- Trading Robot settings (unique for each optimization pass)
- Coefficients describing the trading results (unique for each optimization pass)
<Optimisation_Report Created="06.10.2019 10:39:02"> <Optimiser_Settings> <Item Name="Bot">StockFut\StockFut.ex5</Item> <Item Name="Deposit" Currency="RUR">100000</Item> <Item Name="Leverage">1</Item> </Optimiser_Settings>
Parameters are written to the "Item" block, each having its own "Name" attribute. The deposit currency will be written to the "Currency" attribute.
Based on this, the file structure should contain 2 main sections: tester settings and the description of optimization passes. We need to keep
three parameters for the first section:
- Robot Path Relative to the Experts folder
- Deposit Currency and Deposit
- Account Leverage
The second section will contain a sequence of blocks with optimization results, each of which will contain a section with coefficients as well as a set of robot parameters.
<Optimisation_Results> <Result Symbol="SBRF Splice" TF="1" Start_DT="1481327340" Finish_DT="1512776940"> <Coefficients> <VaR> <Item Name="90">-1055,18214207419</Item> <Item Name="95">-1323,65133343373</Item> <Item Name="99">-1827,30841143882</Item> <Item Name="Mx">-107,03475</Item> <Item Name="Std">739,584549199836</Item> </VaR> <Max_PL_DD> <Item Name="Profit">1045,9305</Item> <Item Name="DD">-630</Item> <Item Name="Total Profit Trades">1</Item> <Item Name="Total Lose Trades">1</Item> <Item Name="Consecutive Wins">1</Item> <Item Name="Consecutive Lose">1</Item> </Max_PL_DD> <Trading_Days> <Mn> <Item Name="Profit">0</Item> <Item Name="DD">0</Item> <Item Name="Number Of Profit Trades">0</Item> <Item Name="Number Of Lose Trades">0</Item> </Mn> <Tu> <Item Name="Profit">0</Item> <Item Name="DD">0</Item> <Item Name="Number Of Profit Trades">0</Item> <Item Name="Number Of Lose Trades">0</Item> </Tu> <We> <Item Name="Profit">1045,9305</Item> <Item Name="DD">630</Item> <Item Name="Number Of Profit Trades">1</Item> <Item Name="Number Of Lose Trades">1</Item> </We> <Th> <Item Name="Profit">0</Item> <Item Name="DD">0</Item> <Item Name="Number Of Profit Trades">0</Item> <Item Name="Number Of Lose Trades">0</Item> </Th> <Fr> <Item Name="Profit">0</Item> <Item Name="DD">0</Item> <Item Name="Number Of Profit Trades">0</Item> <Item Name="Number Of Lose Trades">0</Item> </Fr> </Trading_Days> <Item Name="Payoff">1,66020714285714</Item> <Item Name="Profit factor">1,66020714285714</Item> <Item Name="Average Profit factor">0,830103571428571</Item> <Item Name="Recovery factor">0,660207142857143</Item> <Item Name="Average Recovery factor">-0,169896428571429</Item> <Item Name="Total trades">2</Item> <Item Name="PL">415,9305</Item> <Item Name="DD">-630</Item> <Item Name="Altman Z Score">0</Item> </Coefficients> <Item Name="_lot_">1</Item> <Item Name="USymbol">SBER</Item> <Item Name="Spread_in_percent">3.00000000</Item> <Item Name="UseAutoLevle">false</Item> <Item Name="max_per">174</Item> <Item Name="comission_stock">0.05000000</Item> <Item Name="shift_stock">0.00000000</Item> <Item Name="comission_fut">4.00000000</Item> <Item Name="shift_fut">0.00000000</Item> </Result> </Optimisation_Results> </Optimisation_Report>
Inside the Optimisation_Results block, we will have repeated Result blocks, each of which contains the i-th optimization pass. Each of the Result blocks contains 4 attributes:
- Symbol
- TF
- Start_DT
- Finish_DT
These are the tester settings which will vary depending on the time interval in which the optimization is performed. Each of the robot parameters is written to the Item block with the Name attribute as a unique value, based on which it can be identified. Robot coefficients are written to the Coefficients block. Coefficients which cannot be grouped are enumerated directly in the Item block. Other coefficients are divided into blocks:
- VaR
- 90 - quantile 90
- 95 - quantile 95
- 99 - quantile 99
- Mx - math expectation
- Std - standard deviation
- Max_PL_DD
- Profit - total profit
- DD - total drawdown
- Total Profit Trades - total number of profitable trades
- Total Lose Trades - total number of losing trades
- Consecutive Wins - winning trades in a row
- Consecutive Lose - losing trades in a row
- Trading_Days - trading reports by days
- Profit - average profit per day
- DD - average losses per day
- Number Of Profit Trades - number of profitable trades
- Number Of Lose Trades - number of losing trades
As a result, we receive a list with the coefficients of optimization results, which fully describe test results. Now, to filter and select robot parameters, there is a complete list of required coefficients which enable us to efficiently evaluate the robot performance.
The wrapper class of the optimizations report, the class storing optimization dates, as well as the structure of optimizations results in C#.
Let's start with the structure storing data for a specific optimization pass.
public struct ReportItem { public Dictionary<string, string> BotParams; // List of robot parameters public Coefficients OptimisationCoefficients; // Robot coefficients public string Symbol; // Symbol public int TF; // Timeframe public DateBorders DateBorders; // Date range }
All robot coefficients are stored in the dictionary in a string format. The file with robot parameters does not save the type of data, therefore the string format suites best here. The list of robot coefficients is provided in a different structure, similarly to other blocks which are grouped in the *xml optimizations report. Trading reports by days are also stored in the dictionary.
public Dictionary<DayOfWeek, DailyData> TradingDays;
The DayOfWeek and the dictionary must always contain the enumeration of 5 days (from Monday to Friday) as a key, similarly to the *xml file. The most interesting class in the data storing structure is DateBorders. Similar to the data being grouped within a structure containing fields which describe each of the date parameters, date ranges are also stored in the DateBorders structure.
public class DateBorders : IComparable { /// <summary> /// Constructor /// </summary> /// <param name="from">Range beginning date</param> /// <param name="till">Range ending date</param> public DateBorders(DateTime from, DateTime till) { if (till <= from) throw new ArgumentException("Date 'Till' is less or equal to date 'From'"); From = from; Till = till; } /// <summary> /// From /// </summary> public DateTime From { get; } /// <summary> /// To /// </summary> public DateTime Till { get; } }
For a fully-featured operation with the date range, we need the possibility to create two date ranges. For this purpose, overwrite 2 operators "==" and "!=".
Equality criteria are determined by the equality of both dates in the two passed ranges, i.e. the beginning date matches the trading beginning of the second range (while the same also applies to the trading end). However, since the object type is 'class', it can be equal to null, and thus we first need to provide for the ability to compare to null. Let's use the is keyword for that purpose. After that we can compare parameters with each other, otherwise, if we try to compare to null, "null reference exception" will be returned.
#region Equal /// <summary> /// The equality comparison operator /// </summary> /// <param name="b1">Element 1</param> /// <param name="b2">Element 2</param> /// <returns>Result</returns> public static bool operator ==(DateBorders b1, DateBorders b2) { bool ans; if (b2 is null && b1 is null) ans = true; else if (b2 is null || b1 is null) ans = false; else ans = b1.From == b2.From && b1.Till == b2.Till; return ans; } /// <summary> /// The inequality comparison operator /// </summary> /// <param name="b1">Element 1</param> /// <param name="b2">Element 2</param> /// <returns>Comparison result</returns> public static bool operator !=(DateBorders b1, DateBorders b2) => !(b1 == b2); #endregion
To overload the inequality operator, we no longer need to write the above described procedures, while all of them are already written in operator "==". The next feature we need to implement is data sorting by time periods, that is why we need to overload operators ">", "<", ">=", "<=".
#region (Grater / Less) than /// <summary> /// Comparing: current element is greater than the previous one /// </summary> /// <param name="b1">Element 1</param> /// <param name="b2">Element 2</param> /// <returns>Result</returns> public static bool operator >(DateBorders b1, DateBorders b2) { if (b1 == null || b2 == null) return false; if (b1.From == b2.From) return (b1.Till > b2.Till); else return (b1.From > b2.From); } /// <summary> /// Comparing: current element is less than the previous one /// </summary> /// <param name="b1">Element 1</param> /// <param name="b2">Element 2</param> /// <returns>Result</returns> public static bool operator <(DateBorders b1, DateBorders b2) { if (b1 == null || b2 == null) return false; if (b1.From == b2.From) return (b1.Till < b2.Till); else return (b1.From < b2.From); } #endregion
If any of the parameters passed to the operator is equal to null, the comparison becomes impossible, therefore return False. Otherwise compare step by step. If the first time interval matches, compare by the second time interval. If they are not equal, compare by the first interval. Thus, if we describe the comparison logic based om the "Greater" operator example, the greater interval is the one which is older in time than the previous one, either by the start date or by the end date (if start dates are equal). The "less" comparison logic is similar to the "greater" comparison.
The next operators to be overloaded to enable the sorting option are 'Greater Than Or Equal' and 'Less Than Or Equal'.
#region Equal or (Grater / Less) than /// <summary> /// Greater than or equal comparison /// </summary> /// <param name="b1">Element 1</param> /// <param name="b2">Element 2</param> /// <returns>Result</returns> public static bool operator >=(DateBorders b1, DateBorders b2) => (b1 == b2 || b1 > b2); /// <summary> /// Less than or equal comparison /// </summary> /// <param name="b1">Element 1</param> /// <param name="b2">Element 2</param> /// <returns>Result</returns> public static bool operator <=(DateBorders b1, DateBorders b2) => (b1 == b2 || b1 < b2); #endregion
As can be seen, the operator overload does not require the description of the internal comparison logic. Instead, we use the already overloaded operators == and >, <. However, as Visual Studio suggests during compilation, in addition to the overloading of these operators we need to overload some functions inherited from the "object" base class.
#region override base methods (from object) /// <summary> /// Overloading of equality comparison /// </summary> /// <param name="obj">Element to compare to</param> /// <returns></returns> public override bool Equals(object obj) { if (obj is DateBorders other) return this == other; else return base.Equals(obj); } /// <summary> /// Cast the class to a string and return its hash code /// </summary> /// <returns>String hash code</returns> public override int GetHashCode() { return ToString().GetHashCode(); } /// <summary> /// Convert the current class to a string /// </summary> /// <returns>String From date - To date</returns> public override string ToString() { return $"{From}-{Till}"; } #endregion /// <summary> /// Compare the current element with the passed one /// </summary> /// <param name="obj"></param> /// <returns></returns> public int CompareTo(object obj) { if (obj == null) return 1; if (obj is DateBorders borders) { if (this == borders) return 0; else if (this < borders) return -1; else return 1; } else { throw new ArgumentException("object is not DateBorders"); } }
Equals method: overload it either using the overloaded operator == (if the passed object has type DateBorders) or the basic implementation of the method.
ToString method: overload it as a string representation of two dates separated by a hyphen. This will help us to overload the GetHashCode method.
GetHashCode method: overload it by first casting the object to a string and then returning the hash code of this string. When a new class instance is created in C#, its hash code is unique regardless of the class content. That is, if we do not overload the method and create two instances of the DateBorders class with the same From and To dates inside, they will have different hash codes despite identical contents. This rule does not apply to strings, because C# provides a mechanism which prevents from the creation of new instances of the String class if the string was previously created — thus their hash codes for identical strings will match. Using the ToString method overloading and using the string hash code, we provide the behavior of our class hash codes similar to those of String. Now, when using the IEnumerable.Distinct method, we can guarantee that the logic of receiving the unique list of date ranges will be correct, as this method is based on the hash codes of the compared objects.
Implementing the IComparable interface, from which our class is inherited, we implement the CompareTo method which compares the current class instance with the passed one. Its implementation is easy and it uses overloads of previously overloaded operators.
Having implemented the required overloads, we can work with this class more efficiently. We can:
- Compare two instances for equality
- Compare two instances for greater than/less than
- Compare two instances for greater than or equal/less than or equal
- Sort ascending/descending
- Get unique values from a list of date ranges
- Use the IEnumerable.Sort method which sorts the list in the descending order and uses the IComparable interface.
Since we are implementing a rolling optimization, which will have backtests and forward tests, we need to create a method to compare historic and forward intervals.
/// <summary> /// Method for comparing forward and historical optimizations /// </summary> /// <param name="History">Array of historical optimization</param> /// <param name="Forward">Array of forward optimizations</param> /// <returns>Sorted list historical - forward optimization</returns> public static Dictionary<DateBorders, DateBorders> CompareHistoryToForward(List<DateBorders> History, List<DateBorders> Forward) { // array of comparable optimizations Dictionary<DateBorders, DateBorders> ans = new Dictionary<DateBorders, DateBorders>(); // Sort the passed parameters History.Sort(); Forward.Sort(); // Create a historical optimization loop int i = 0; foreach (var item in History) { if(ans.ContainsKey(item)) continue; ans.Add(item, null); // Add historical optimization if (Forward.Count <= i) continue; // If the array of forward optimization is less than the index, continue the loop // Forward optimization loop for (int j = i; j < Forward.Count; j++) { // If the current forward optimization is contained in the results array, skip if (ans.ContainsValue(Forward[j]) || Forward[j].From < item.Till) { continue; } // Compare forward and historical optimization ans[item] = Forward[j]; i = j + 1; break; } } return ans; }
As you can see, the method is static. This is done to make it available as a regular function, without binding to a specific class instance. First of all it sorts the passed time intervals in the ascending order. Thus, in the next loop we can know for sure that all the previously passed intervals are less than or equal to the next ones. Then implement two loops: foreach for historical intervals, nested loop for forward intervals.
At the beginning of the historical data loop, we always add historical ranges (key) to the collection with results, as well as temporarily set null in place of forward intervals. Forward results loop starts with the i-th parameter. This prevents from repeating the loop with already used elements of the forward list. That is the forward interval should always follow the historical one, i.e. it should be > than the historical. That is why we implement the loop by forward intervals, in case in the passed list there is a forward period for the very first historical interval, which precedes the very first historical interval. It is better to visualize the idea in a table:
| Historical | Forward | ||
|---|---|---|---|
| From | To | From | To |
| 10.03.2016 | 09.03.2017 | 12.12.2016 | 09.03.2017 |
| 10.06.2016 | 09.06.2017 | 10.03.2017 | 09.06.2017 |
| 10.09.2016 | 09.09.2017 | 10.06.2017 | 09.09.2017 |
So the first historical interval ends on 09.03.2017, and the first forward interval starts on 12.12.2016, which is not correct. That is why we skip it in the forward intervals loop, due to the condition. Also, skip the forward interval, which is contained in the resulting dictionary. If the j-th forward data does not yet exists in the resulting dictionary and the forward interval beginning date is >= current historical interval end date, save the received value and exit the forward intervals loop as the required value has already been found. Before exiting, assign the value of the forward interval following the selected one to the i variable (the variable which means the forward list iterations beginning. This is done because the current interval will no longer be needed (due to the initial data sorting).
A check before the historical optimization ensures that all historical optimizations are unique. Thus, the following list is obtained in the resulting dictionary:
| Key | Value |
|---|---|
| 10.03.2016-09.3.2017 | 10.03.2017-09.06.2017 |
| 10.06.2016-09.06.2017 | 10.06.2017-09.09.2017 |
| 10.09.2016-09.09.2017 | null |
As can be seen from the presented data, the first forward interval is discarded and no interval is found for the last historical one, as no such interval has been passed. Based on this logic, the program will compare data of the historic and forward intervals and will understand which of the historical intervals should provide optimization parameters for forward tests.
To enable efficient operation with a specific optimization result, I have created a wrapper structure for the ReportItem structure which contains a number of additional methods and overloaded operators. Basically, the wrapper contains two fields:
/// <summary> /// Optimization pass report /// </summary> public ReportItem report; /// <summary> /// Sorting factor /// </summary> public double SortBy;
The first field was described above. The second field is created to enable sorting by multiple values, for example profit and recovery factor. The sorting mechanism will be described later, but the idea is to convert these values to one and to store it in this variable.
The structure also contains type conversion overloads:
/// <summary> /// The operator of implicit type conversion from optimization pass to the current type /// </summary> /// <param name="item">Optimization pass report</param> public static implicit operator OptimisationResult(ReportItem item) { return new OptimisationResult { report = item, SortBy = 0 }; } /// <summary> /// The operator of explicit type conversion from current to the optimization pass structure /// </summary> /// <param name="optimisationResult">current type</param> public static explicit operator ReportItem(OptimisationResult optimisationResult) { return optimisationResult.report; }
As a result, we can implicitly cast ReportItem type to its wrapper, and then explicitly cast the ReportItem wrapper to the trading report element. This can be more efficient than sequential filling of fields. Since all fields in the ReportItem structure are divided into categories, we may sometimes need a lengthy code in order to receive a desired value. A special method has been created to save space and to create a more universal getter. It receives the requested robot ratios data via the passed enum SourtBy from the above GetResult(SortBy resultType) code. The implementation is simple but too long and it is therefore not provided here. The method iterates over the passed enum in the switch constuct and returns the value of the requested coefficient. Since most of the coefficients have type double and since this type can contain all other numeric types, coefficient values are converted to double.
Comparison operator overloads have also been implemented for this wrapper type:
/// <summary> /// Overloading of the equality comparison operator /// </summary> /// <param name="result1">Parameter 1 to compare</param> /// <param name="result2">Parameter 2 to compare</param> /// <returns>Comparison result</returns> public static bool operator ==(OptimisationResult result1, OptimisationResult result2) { foreach (var item in result1.report.BotParams) { if (!result2.report.BotParams.ContainsKey(item.Key)) return false; if (result2.report.BotParams[item.Key] != item.Value) return false; } return true; } /// <summary> /// Overloading of the inequality comparison operator /// </summary> /// <param name="result1">Parameter 1 to compare</param> /// <param name="result2">Parameter 2 to compare</param> /// <returns>Comparison result</returns> public static bool operator !=(OptimisationResult result1, OptimisationResult result2) { return !(result1 == result2); } /// <summary> /// Overloading of the basic type comparison operator /// </summary> /// <param name="obj"></param> /// <returns></returns> public override bool Equals(object obj) { if (obj is OptimisationResult other) { return this == other; } else return base.Equals(obj); }
The elements of optimizations containing the same names and values of robot parameters will be considered equal. Thus, if we need to compare two optimization passes, we already have the ready-to-use overloaded operators. This structure also contains a method which writes data to a file. If it exists, data is simply added to the file. Explanation of the data writing element and method implementation will be provided below.
Creating a File to Store the Optimization Report
We will work with the optimization reports and will write them not only in the terminal, but also in the created program. That is why let us add the optimization report creating method to this DLL. Let us also provide several methods for data writing to a file, i.e. enable writing of a data array to a file as well as enable the addition of a separate element to existing file (if the file does not exist, it should be created). The last method will be imported to the terminal and will be used in C# classes. Let's start considering the implemented report file writing methods with the functions connected with the addition of data to a file. The ReportWriter class was created for this purpose. The full class implementation is available in the attached project file. Here I will only show the most interesting methods. Let's first describe how this class works.
It contains only static methods: this enables exporting of its methods to MQL5. For the same purpose, the class is marked with a public access modifier. This class contains a static field of ReportItem type and a number of methods which alternately add coefficients and EA parameters to it.
/// <summary> /// temporary data keeper /// </summary> private static ReportItem ReportItem; /// <summary> /// clearing the temporary data keeper /// </summary> public static void ClearReportItem() { ReportItem = new ReportItem(); }
Another method is ClearReportItem(). It recreates the field instance. In this case we lose access to the previous instance of this object: it is erased and data saving process starts again. Data adding methods are grouped by blocks. Here are the signatures of these methods.
/// <summary> /// Add robot parameters /// </summary> /// <param name="name">Parameter name</param> /// <param name="value">Parameter value</param> public static void AppendBotParam(string name, string value); /// <summary> /// Add the main list of coefficients /// </summary> /// <param name="payoff"></param> /// <param name="profitFactor"></param> /// <param name="averageProfitFactor"></param> /// <param name="recoveryFactor"></param> /// <param name="averageRecoveryFactor"></param> /// <param name="totalTrades"></param> /// <param name="pl"></param> /// <param name="dd"></param> /// <param name="altmanZScore"></param> public static void AppendMainCoef(double payoff, double profitFactor, double averageProfitFactor, double recoveryFactor, double averageRecoveryFactor, int totalTrades, double pl, double dd, double altmanZScore); /// <summary> /// Add VaR /// </summary> /// <param name="Q_90"></param> /// <param name="Q_95"></param> /// <param name="Q_99"></param> /// <param name="Mx"></param> /// <param name="Std"></param> public static void AppendVaR(double Q_90, double Q_95, double Q_99, double Mx, double Std); /// <summary> /// Add total PL / DD and associated values /// </summary> /// <param name="profit"></param> /// <param name="dd"></param> /// <param name="totalProfitTrades"></param> /// <param name="totalLoseTrades"></param> /// <param name="consecutiveWins"></param> /// <param name="consecutiveLose"></param> public static void AppendMaxPLDD(double profit, double dd, int totalProfitTrades, int totalLoseTrades, int consecutiveWins, int consecutiveLose); /// <summary> /// Add a specific day /// </summary> /// <param name="day"></param> /// <param name="profit"></param> /// <param name="dd"></param> /// <param name="numberOfProfitTrades"></param> /// <param name="numberOfLoseTrades"></param> public static void AppendDay(int day, double profit, double dd, int numberOfProfitTrades, int numberOfLoseTrades);
The method adding trading statistics broken down by days should be called for each of the 5 trading days. If we do not add it for one of the days, the written file will not be read in the future. Once data is added to the data storage field, we can proceed to recording the field. Before this, check if the file exists and create it if necessary. A few methods have been added for creating the file.
/// <summary> /// The method creates the file if it has not been created /// </summary> /// <param name="pathToBot">Path to the robot</param> /// <param name="currency">Deposit currency</param> /// <param name="balance">Balance</param> /// <param name="leverage">Leverage</param> /// <param name="pathToFile">Path to file</param> private static void CreateFileIfNotExists(string pathToBot, string currency, double balance, int leverage, string pathToFile) { if (File.Exists(pathToFile)) return; using (var xmlWriter = new XmlTextWriter(pathToFile, null)) { // set document format xmlWriter.Formatting = Formatting.Indented; xmlWriter.IndentChar = '\t'; xmlWriter.Indentation = 1; xmlWriter.WriteStartDocument(); // Create document root #region Document root xmlWriter.WriteStartElement("Optimisation_Report"); // Write the creation date xmlWriter.WriteStartAttribute("Created"); xmlWriter.WriteString(DateTime.Now.ToString("dd.MM.yyyy HH:mm:ss")); xmlWriter.WriteEndAttribute(); #region Optimiser settings section // Optimizer settings xmlWriter.WriteStartElement("Optimiser_Settings"); // Path to the robot WriteItem(xmlWriter, "Bot", pathToBot); // Deposit WriteItem(xmlWriter, "Deposit", balance.ToString(), new Dictionary<string, string> { { "Currency", currency } }); // Leverage WriteItem(xmlWriter, "Leverage", leverage.ToString()); xmlWriter.WriteEndElement(); #endregion #region Optimization results section // the root node of the optimization results list xmlWriter.WriteStartElement("Optimisation_Results"); xmlWriter.WriteEndElement(); #endregion xmlWriter.WriteEndElement(); #endregion xmlWriter.WriteEndDocument(); xmlWriter.Close(); } } /// <summary> /// Write element to a file /// </summary> /// <param name="writer">Writer</param> /// <param name="Name">Element name</param> /// <param name="Value">Element value</param> /// <param name="Attributes">Attributes</param> private static void WriteItem(XmlTextWriter writer, string Name, string Value, Dictionary<string, string> Attributes = null) { writer.WriteStartElement("Item"); writer.WriteStartAttribute("Name"); writer.WriteString(Name); writer.WriteEndAttribute(); if (Attributes != null) { foreach (var item in Attributes) { writer.WriteStartAttribute(item.Key); writer.WriteString(item.Value); writer.WriteEndAttribute(); } } writer.WriteString(Value); writer.WriteEndElement(); }
I also provide here the implementation of the WriteItem method which contains the repeating code for adding a final element with data and element-specific attributes to a file. The file creating method CreateFileIfNotExists checks whether the file exists, creates the file and starts forming the minimum required file structure.
Firstly, it creates the file root, i.e. the <Optimization_Report/> tag, inside which all the child structures of the file are located. Then file creation data is filled — this is implemented for further convenient work with files. After that we create a node with unchanged optimizer settings and specify them. Then create a section which will store optimization results and immediately close it. As a result we have an empty file with the minimum required formatting.
<Optimisation_Report Created="24.10.2019 19:10:08"> <Optimiser_Settings> <Item Name="Bot">Path to bot</Item> <Item Name="Deposit" Currency="Currency">1000</Item> <Item Name="Leverage">1</Item> </Optimiser_Settings> <Optimisation_Results /> </Optimisation_Report>
Thus we will be able to read this file using the XmlDocument class. This is the most useful class for reading and editing existing Xml documents. We will use exactly this class to add data to existing documents. Repeated operations are implemented as separate methods and thus we will be able to add data to an exiting document more efficiently:
/// <summary> /// Writing attributes to a file /// </summary> /// <param name="item">Node</param> /// <param name="xmlDoc">Document</param> /// <param name="Attributes">Attributes</param> private static void FillInAttributes(XmlNode item, XmlDocument xmlDoc, Dictionary<string, string> Attributes) { if (Attributes != null) { foreach (var attr in Attributes) { XmlAttribute attribute = xmlDoc.CreateAttribute(attr.Key); attribute.Value = attr.Value; item.Attributes.Append(attribute); } } } /// <summary> /// Add section /// </summary> /// <param name="xmlDoc">Document</param> /// <param name="xpath_parentSection">xpath to select parent node</param> /// <param name="sectionName">Section name</param> /// <param name="Attributes">Attribute</param> private static void AppendSection(XmlDocument xmlDoc, string xpath_parentSection, string sectionName, Dictionary<string, string> Attributes = null) { XmlNode section = xmlDoc.SelectSingleNode(xpath_parentSection); XmlNode item = xmlDoc.CreateElement(sectionName); FillInAttributes(item, xmlDoc, Attributes); section.AppendChild(item); } /// <summary> /// Write item /// </summary> /// <param name="xmlDoc">Document</param> /// <param name="xpath_parentSection">xpath to select parent node</param> /// <param name="name">Item name</param> /// <param name="value">Value</param> /// <param name="Attributes">Attributes</param> private static void WriteItem(XmlDocument xmlDoc, string xpath_parentSection, string name, string value, Dictionary<string, string> Attributes = null) { XmlNode section = xmlDoc.SelectSingleNode(xpath_parentSection); XmlNode item = xmlDoc.CreateElement(name); item.InnerText = value; FillInAttributes(item, xmlDoc, Attributes); section.AppendChild(item); }
The first method FillInAttributes fills attributes for the passed node, WriteItem writes an item to the section specified via XPath, while AppendSection adds a section inside another section, which is specified via a path passed using Xpath. These code blocks are often used when adding data to a file. The data writing method is quite lengthy and is divided into blocks.
/// <summary> /// Write trading results to a file /// </summary> /// <param name="pathToBot">Path to the bot</param> /// <param name="currency">Deposit currency</param> /// <param name="balance">Balance</param> /// <param name="leverage">Leverage</param> /// <param name="pathToFile">Path to file</param> /// <param name="symbol">Symbol</param> /// <param name="tf">Timeframe</param> /// <param name="StartDT">Trading start dare</param> /// <param name="FinishDT">Trading end date</param> public static void Write(string pathToBot, string currency, double balance, int leverage, string pathToFile, string symbol, int tf, ulong StartDT, ulong FinishDT) { // Create the file if it does not yet exist CreateFileIfNotExists(pathToBot, currency, balance, leverage, pathToFile); ReportItem.Symbol = symbol; ReportItem.TF = tf; // Create a document and read the file using it XmlDocument xmlDoc = new XmlDocument(); xmlDoc.Load(pathToFile); #region Append result section // Write a request to switch to the optimization results section string xpath = "Optimisation_Report/Optimisation_Results"; // Add a new section with optimization results AppendSection(xmlDoc, xpath, "Result", new Dictionary<string, string> { { "Symbol", symbol }, { "TF", tf.ToString() }, { "Start_DT", StartDT.ToString() }, { "Finish_DT", FinishDT.ToString() } }); // Add section with optimization results AppendSection(xmlDoc, $"{xpath}/Result[last()]", "Coefficients"); // Add section with VaR AppendSection(xmlDoc, $"{xpath}/Result[last()]/Coefficients", "VaR"); // Add section with total PL / DD AppendSection(xmlDoc, $"{xpath}/Result[last()]/Coefficients", "Max_PL_DD"); // Add section with trading results by days AppendSection(xmlDoc, $"{xpath}/Result[last()]/Coefficients", "Trading_Days"); // Add section with trading results on Monday AppendSection(xmlDoc, $"{xpath}/Result[last()]/Coefficients/Trading_Days", "Mn"); // Add section with trading results on Tuesday AppendSection(xmlDoc, $"{xpath}/Result[last()]/Coefficients/Trading_Days", "Tu"); // Add section with trading results on Wednesday AppendSection(xmlDoc, $"{xpath}/Result[last()]/Coefficients/Trading_Days", "We"); // Add section with trading results on Thursday AppendSection(xmlDoc, $"{xpath}/Result[last()]/Coefficients/Trading_Days", "Th"); // Add section with trading results on Friday AppendSection(xmlDoc, $"{xpath}/Result[last()]/Coefficients/Trading_Days", "Fr"); #endregion #region Append Bot params // Iterate through bot parameters foreach (var item in ReportItem.BotParams) { // Write the selected robot parameter WriteItem(xmlDoc, "Optimisation_Report/Optimisation_Results/Result[last()]", "Item", item.Value, new Dictionary<string, string> { { "Name", item.Key } }); } #endregion #region Append main coef // Set path to node with coefficients xpath = "Optimisation_Report/Optimisation_Results/Result[last()]/Coefficients"; // Save coefficients WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.Payoff.ToString(), new Dictionary<string, string> { { "Name", "Payoff" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.ProfitFactor.ToString(), new Dictionary<string, string> { { "Name", "Profit factor" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.AverageProfitFactor.ToString(), new Dictionary<string, string> { { "Name", "Average Profit factor" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.RecoveryFactor.ToString(), new Dictionary<string, string> { { "Name", "Recovery factor" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.AverageRecoveryFactor.ToString(), new Dictionary<string, string> { { "Name", "Average Recovery factor" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.TotalTrades.ToString(), new Dictionary<string, string> { { "Name", "Total trades" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.PL.ToString(), new Dictionary<string, string> { { "Name", "PL" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.DD.ToString(), new Dictionary<string, string> { { "Name", "DD" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.AltmanZScore.ToString(), new Dictionary<string, string> { { "Name", "Altman Z Score" } }); #endregion #region Append VaR // Set path to node with VaR xpath = "Optimisation_Report/Optimisation_Results/Result[last()]/Coefficients/VaR"; // Save VaR results WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.VaR.Q_90.ToString(), new Dictionary<string, string> { { "Name", "90" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.VaR.Q_95.ToString(), new Dictionary<string, string> { { "Name", "95" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.VaR.Q_99.ToString(), new Dictionary<string, string> { { "Name", "99" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.VaR.Mx.ToString(), new Dictionary<string, string> { { "Name", "Mx" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.VaR.Std.ToString(), new Dictionary<string, string> { { "Name", "Std" } }); #endregion #region Append max PL and DD // Set path to node with total PL / DD xpath = "Optimisation_Report/Optimisation_Results/Result[last()]/Coefficients/Max_PL_DD"; // Save coefficients WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.MaxPLDD.Profit.Value.ToString(), new Dictionary<string, string> { { "Name", "Profit" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.MaxPLDD.DD.Value.ToString(), new Dictionary<string, string> { { "Name", "DD" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.MaxPLDD.Profit.TotalTrades.ToString(), new Dictionary<string, string> { { "Name", "Total Profit Trades" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.MaxPLDD.DD.TotalTrades.ToString(), new Dictionary<string, string> { { "Name", "Total Lose Trades" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.MaxPLDD.Profit.ConsecutivesTrades.ToString(), new Dictionary<string, string> { { "Name", "Consecutive Wins" } }); WriteItem(xmlDoc, xpath, "Item", ReportItem.OptimisationCoefficients.MaxPLDD.DD.ConsecutivesTrades.ToString(), new Dictionary<string, string> { { "Name", "Consecutive Lose" } }); #endregion #region Append Days foreach (var item in ReportItem.OptimisationCoefficients.TradingDays) { // Set path to specific day node xpath = "Optimisation_Report/Optimisation_Results/Result[last()]/Coefficients/Trading_Days"; // Select day switch (item.Key) { case DayOfWeek.Monday: xpath += "/Mn"; break; case DayOfWeek.Tuesday: xpath += "/Tu"; break; case DayOfWeek.Wednesday: xpath += "/We"; break; case DayOfWeek.Thursday: xpath += "/Th"; break; case DayOfWeek.Friday: xpath += "/Fr"; break; } // Save results WriteItem(xmlDoc, xpath, "Item", item.Value.Profit.Value.ToString(), new Dictionary<string, string> { { "Name", "Profit" } }); WriteItem(xmlDoc, xpath, "Item", item.Value.DD.Value.ToString(), new Dictionary<string, string> { { "Name", "DD" } }); WriteItem(xmlDoc, xpath, "Item", item.Value.Profit.Trades.ToString(), new Dictionary<string, string> { { "Name", "Number Of Profit Trades" } }); WriteItem(xmlDoc, xpath, "Item", item.Value.DD.Trades.ToString(), new Dictionary<string, string> { { "Name", "Number Of Lose Trades" } }); } #endregion // Rewrite the file with the changes xmlDoc.Save(pathToFile); // Clear the variable which stored results written to a file ClearReportItem(); }
First we load the entire document to memory and then add sections. Let us consider the Xpath request format which passes the path to the root node.
$"{xpath}/Result[last()]/Coefficients"
The xpath variable contains the path to the node in which the optimization pass elements are stored. This node stores optimization results nodes which can be presented as an array of structures. The Result[last()] construct selects the last element of the array, after which the path is passed to the nested /Coefficients node. Following the described principle, we select the required node with the results of optimizations.
The next step is adding of robot parameters: in the loop we add parameters directly to the results directory. Then add a number of coefficients into the coefficients directory. This addition is divided into blocks. As a result we save results and clear the temporary storage. As a result we get a file with the list of parameters and optimization results. To separate threads during asynchronous operations launched from different processes (this is how optimization in the tester is performed when using multiple processors), another writing method has been created, which separates threads using named mutexes.
/// <summary> /// Write to file while locking using a named mutex /// </summary> /// <param name="mutexName">Mutex name</param> /// <param name="pathToBot">Path to the bot</param> /// <param name="currency">Deposit currency</param> /// <param name="balance">Balance</param> /// <param name="leverage">Leverage</param> /// <param name="pathToFile">Path to file</param> /// <param name="symbol">Symbol</param> /// <param name="tf">Timeframe</param> /// <param name="StartDT">Trading start dare</param> /// <param name="FinishDT">Trading end date</param> /// <returns></returns> public static string MutexWriter(string mutexName, string pathToBot, string currency, double balance, int leverage, string pathToFile, string symbol, int tf, ulong StartDT, ulong FinishDT) { string ans = ""; // Mutex lock Mutex m = new Mutex(false, mutexName); m.WaitOne(); try { // write to file Write(pathToBot, currency, balance, leverage, pathToFile, symbol, tf, StartDT, FinishDT); } catch (Exception e) { // Catch error if any ans = e.Message; } // Release the mutex m.ReleaseMutex(); // Return error text return ans; }
This method writes data using the previous method but the writing process is wrapped by a mutex and in a try-catch block. The last one enables mutex release even on case of an error. Otherwise the process may freeze and optimization may fail to continue. These methods are also used in the OptimisationResult structure in the WriteResult method.
/// <summary> /// The method adds current parameter to the existing file or creates a new file with the current parameter /// </summary> /// <param name="pathToBot">Relative path to the robot from the Experts folder</param> /// <param name="currency">Deposit currency</param> /// <param name="balance">Balance</param> /// <param name="leverage">Leverage</param> /// <param name="pathToFile">Path to file</param> public void WriteResult(string pathToBot, string currency, double balance, int leverage, string pathToFile) { try { foreach (var param in report.BotParams) { ReportWriter.AppendBotParam(param.Key, param.Value); } ReportWriter.AppendMainCoef(GetResult(ReportManager.SortBy.Payoff), GetResult(ReportManager.SortBy.ProfitFactor), GetResult(ReportManager.SortBy.AverageProfitFactor), GetResult(ReportManager.SortBy.RecoveryFactor), GetResult(ReportManager.SortBy.AverageRecoveryFactor), (int)GetResult(ReportManager.SortBy.TotalTrades), GetResult(ReportManager.SortBy.PL), GetResult(ReportManager.SortBy.DD), GetResult(ReportManager.SortBy.AltmanZScore)); ReportWriter.AppendVaR(GetResult(ReportManager.SortBy.Q_90), GetResult(ReportManager.SortBy.Q_95), GetResult(ReportManager.SortBy.Q_99), GetResult(ReportManager.SortBy.Mx), GetResult(ReportManager.SortBy.Std)); ReportWriter.AppendMaxPLDD(GetResult(ReportManager.SortBy.ProfitFactor), GetResult(ReportManager.SortBy.MaxDD), (int)GetResult(ReportManager.SortBy.MaxProfitTotalTrades), (int)GetResult(ReportManager.SortBy.MaxDDTotalTrades), (int)GetResult(ReportManager.SortBy.MaxProfitConsecutivesTrades), (int)GetResult(ReportManager.SortBy.MaxDDConsecutivesTrades)); foreach (var day in report.OptimisationCoefficients.TradingDays) { ReportWriter.AppendDay((int)day.Key, day.Value.Profit.Value, day.Value.Profit.Value, day.Value.Profit.Trades, day.Value.DD.Trades); } ReportWriter.Write(pathToBot, currency, balance, leverage, pathToFile, report.Symbol, report.TF, report.DateBorders.From.DTToUnixDT(), report.DateBorders.Till.DTToUnixDT()); } catch (Exception e) { ReportWriter.ClearReportItem(); throw e; } }
In this method, we alternately add optimization results to a temporary storage, then call the Write method to save them in an existing file or create a new file if it has not yet been created.
The described method for writing obtained data is required for adding information to a prepared file. There is another method which is better suitable when a data series needs to be written. The method has been developed as an extension for the IEnumerable<OptimisationResult> interface. Now we can save data for all lists inherited from the appropriate interface.
public static void ReportWriter(this IEnumerable<OptimisationResult> results, string pathToBot, string currency, double balance, int leverage, string pathToFile) { // Delete the file if it exists if (File.Exists(pathToFile)) File.Delete(pathToFile); // Create writer using (var xmlWriter = new XmlTextWriter(pathToFile, null)) { // Set document format xmlWriter.Formatting = Formatting.Indented; xmlWriter.IndentChar = '\t'; xmlWriter.Indentation = 1; xmlWriter.WriteStartDocument(); // The root node of the document xmlWriter.WriteStartElement("Optimisation_Report"); // Write attributes WriteAttribute(xmlWriter, "Created", DateTime.Now.ToString("dd.MM.yyyy HH:mm:ss")); // Write optimizer settings to file #region Optimiser settings section xmlWriter.WriteStartElement("Optimiser_Settings"); WriteItem(xmlWriter, "Bot", pathToBot); // path to the robot WriteItem(xmlWriter, "Deposit", balance.ToString(), new Dictionary<string, string> { { "Currency", currency } }); // Currency and deposit WriteItem(xmlWriter, "Leverage", leverage.ToString()); // Leverage xmlWriter.WriteEndElement(); #endregion // Write optimization results to the file #region Optimisation result section xmlWriter.WriteStartElement("Optimisation_Results"); // Loop through optimization results foreach (var item in results) { // Write specific result xmlWriter.WriteStartElement("Result"); // Write attributes of this optimization pass WriteAttribute(xmlWriter, "Symbol", item.report.Symbol); // Symbol WriteAttribute(xmlWriter, "TF", item.report.TF.ToString()); // Timeframe WriteAttribute(xmlWriter, "Start_DT", item.report.DateBorders.From.DTToUnixDT().ToString()); // Optimization start date WriteAttribute(xmlWriter, "Finish_DT", item.report.DateBorders.Till.DTToUnixDT().ToString()); // Optimization end date // Write optimization result WriteResultItem(item, xmlWriter); xmlWriter.WriteEndElement(); } xmlWriter.WriteEndElement(); #endregion xmlWriter.WriteEndElement(); xmlWriter.WriteEndDocument(); xmlWriter.Close(); } }
The method writes optimization reports into a file one by one until the array has no more data. If the file already exists at the passed path, it will be replaced with a new one. First we create a file writer and configure it. Then, following the already known file structure, we write optimizer settings and optimization results one by one. As can be seen from the above code extract, the results are written in a loop, which loops through the elements of the collection, at the instance of which the described method was called. Inside the loop, data writing is delegated to the method created for writing data of a specific element into the file.
/// <summary> /// Write a specific optimization pass /// </summary> /// <param name="resultItem">Optimization pass value</param> /// <param name="writer">Writer</param> private static void WriteResultItem(OptimisationResult resultItem, XmlTextWriter writer) { // Write coefficients #region Coefficients writer.WriteStartElement("Coefficients"); // Write VaR #region VaR writer.WriteStartElement("VaR"); WriteItem(writer, "90", resultItem.GetResult(SortBy.Q_90).ToString()); // Quantile 90 WriteItem(writer, "95", resultItem.GetResult(SortBy.Q_95).ToString()); // Quantile 95 WriteItem(writer, "99", resultItem.GetResult(SortBy.Q_99).ToString()); // Quantile 99 WriteItem(writer, "Mx", resultItem.GetResult(SortBy.Mx).ToString()); // Average for PL WriteItem(writer, "Std", resultItem.GetResult(SortBy.Std).ToString()); // Standard deviation for PL writer.WriteEndElement(); #endregion // Write PL / DD parameters - extreme points #region Max PL DD writer.WriteStartElement("Max_PL_DD"); WriteItem(writer, "Profit", resultItem.GetResult(SortBy.MaxProfit).ToString()); // Total profit WriteItem(writer, "DD", resultItem.GetResult(SortBy.MaxDD).ToString()); // Total loss WriteItem(writer, "Total Profit Trades", ((int)resultItem.GetResult(SortBy.MaxProfitTotalTrades)).ToString()); // Total number of winning trades WriteItem(writer, "Total Lose Trades", ((int)resultItem.GetResult(SortBy.MaxDDTotalTrades)).ToString()); // Total number of losing trades WriteItem(writer, "Consecutive Wins", ((int)resultItem.GetResult(SortBy.MaxProfitConsecutivesTrades)).ToString()); // Winning trades in a row WriteItem(writer, "Consecutive Lose", ((int)resultItem.GetResult(SortBy.MaxDDConsecutivesTrades)).ToString()); // Losing trades in a row writer.WriteEndElement(); #endregion // Write trading results by days #region Trading_Days // The method writing trading results void AddDay(string Day, double Profit, double DD, int ProfitTrades, int DDTrades) { writer.WriteStartElement(Day); WriteItem(writer, "Profit", Profit.ToString()); // Profits WriteItem(writer, "DD", DD.ToString()); // Losses WriteItem(writer, "Number Of Profit Trades", ProfitTrades.ToString()); // Number of profitable trades WriteItem(writer, "Number Of Lose Trades", DDTrades.ToString()); // Number of losing trades writer.WriteEndElement(); } writer.WriteStartElement("Trading_Days"); // Monday AddDay("Mn", resultItem.GetResult(SortBy.AverageDailyProfit_Mn), resultItem.GetResult(SortBy.AverageDailyDD_Mn), (int)resultItem.GetResult(SortBy.AverageDailyProfitTrades_Mn), (int)resultItem.GetResult(SortBy.AverageDailyDDTrades_Mn)); // Tuesday AddDay("Tu", resultItem.GetResult(SortBy.AverageDailyProfit_Tu), resultItem.GetResult(SortBy.AverageDailyDD_Tu), (int)resultItem.GetResult(SortBy.AverageDailyProfitTrades_Tu), (int)resultItem.GetResult(SortBy.AverageDailyDDTrades_Tu)); // Wednesday AddDay("We", resultItem.GetResult(SortBy.AverageDailyProfit_We), resultItem.GetResult(SortBy.AverageDailyDD_We), (int)resultItem.GetResult(SortBy.AverageDailyProfitTrades_We), (int)resultItem.GetResult(SortBy.AverageDailyDDTrades_We)); // Thursday AddDay("Th", resultItem.GetResult(SortBy.AverageDailyProfit_Th), resultItem.GetResult(SortBy.AverageDailyDD_Th), (int)resultItem.GetResult(SortBy.AverageDailyProfitTrades_Th), (int)resultItem.GetResult(SortBy.AverageDailyDDTrades_Th)); // Friday AddDay("Fr", resultItem.GetResult(SortBy.AverageDailyProfit_Fr), resultItem.GetResult(SortBy.AverageDailyDD_Fr), (int)resultItem.GetResult(SortBy.AverageDailyProfitTrades_Fr), (int)resultItem.GetResult(SortBy.AverageDailyDDTrades_Fr)); writer.WriteEndElement(); #endregion // Write other coefficients WriteItem(writer, "Payoff", resultItem.GetResult(SortBy.Payoff).ToString()); WriteItem(writer, "Profit factor", resultItem.GetResult(SortBy.ProfitFactor).ToString()); WriteItem(writer, "Average Profit factor", resultItem.GetResult(SortBy.AverageProfitFactor).ToString()); WriteItem(writer, "Recovery factor", resultItem.GetResult(SortBy.RecoveryFactor).ToString()); WriteItem(writer, "Average Recovery factor", resultItem.GetResult(SortBy.AverageRecoveryFactor).ToString()); WriteItem(writer, "Total trades", ((int)resultItem.GetResult(SortBy.TotalTrades)).ToString()); WriteItem(writer, "PL", resultItem.GetResult(SortBy.PL).ToString()); WriteItem(writer, "DD", resultItem.GetResult(SortBy.DD).ToString()); WriteItem(writer, "Altman Z Score", resultItem.GetResult(SortBy.AltmanZScore).ToString()); writer.WriteEndElement(); #endregion // Write robot coefficients #region Bot params foreach (var item in resultItem.report.BotParams) { WriteItem(writer, item.Key, item.Value); } #endregion }
The implementation of the method that writes data to a file is very simple, although it is quite long. After creating appropriate sections and filling the attributes the method adds data on VaR of the performed optimization pass and values characterizing the maximum profit and drawdown. A nested function has been created to write optimization results for a specific date, which is called 5 times, for each of the days. After that coefficients without grouping and root parameters are added. Since the described procedure is performed in one loop for each of the elements, the data are not written to the file until the xmlWriter.Close() method is called (this is done in the main writing method). Thus, this is the fastest extension method for writing a data array, as compared to previously considered methods. We have considered procedures related to data writing into a file. Now let us move on to the next logical part of the description, i.e. data reading from the resulting file.
Reading the optimization report file
We need to read the files in order to process the received information and to display it. Therefore, an appropriate file reading mechanism is required. It is implemented as a separate class:
public class ReportReader : IDisposable { /// <summary> /// Constructor /// </summary> /// <param name="path">Path to file</param> public ReportReader(string path); /// <summary> /// Binary number format provider /// </summary> private readonly NumberFormatInfo formatInfo = new NumberFormatInfo { NumberDecimalSeparator = "." }; #region DataKeepers /// <summary> /// Presenting the report file in OOP format /// </summary> private readonly XmlDocument document = new XmlDocument(); /// <summary> /// Collection of document nodes (rows in excel table) /// </summary> private readonly System.Collections.IEnumerator enumerator; #endregion /// <summary> /// The read current report item /// </summary> public ReportItem? ReportItem { get; private set; } = null; #region Optimiser settings /// <summary> /// Path to the robot /// </summary> public string RelativePathToBot { get; } /// <summary> /// Balance /// </summary> public double Balance { get; } /// <summary> /// Currency /// </summary> public string Currency { get; } /// <summary> /// Leverage /// </summary> public int Leverage { get; } #endregion /// <summary> /// File creation date /// </summary> public DateTime Created { get; } /// <summary> /// File reader method /// </summary> /// <returns></returns> public bool Read(); /// <summary> /// The method receiving the item by its name (the Name attribute) /// </summary> /// <param name="Name"></param> /// <returns></returns> private string SelectItem(string Name) => $"Item[@Name='{Name}']"; /// <summary> /// Get the trading result value for the selected day /// </summary> /// <param name="dailyNode">Node of this day</param> /// <returns></returns> private DailyData GetDay(XmlNode dailyNode); /// <summary> /// Reset the quote reader /// </summary> public void ResetReader(); /// <summary> /// Clear the document /// </summary> public void Dispose() => document.RemoveAll(); }
Let's view the structure in more detail. The class is inherited from the iDisposable interface. This is not a required condition, but is done for precaution. Now the describe class contains the required Dispasable method which clears the document object. The object stores the optimization results file loaded to memory.
The approach is convenient because when creating an instance, the class inherited from the above mentioned interface should be wrapped to the 'using' construct, which automatically calls the specified method when it goes beyond the 'using' structure block boundaries. It means that the read document will not be kept long in memory and thus the loaded memory amount is reduced.
The row-wise document reader class uses Enumerator received from the read document. The read values are written to the special property and thus we provide access to data. Also, the following data is filled during class instantiation: properties specifying the main optimizer settings, file creation date and time. To eliminate the influence of OS localization settings (both when writing and when reading the file) the double-precision number delimiter format is indicated. When reading the file for the first time, the class should be reset to list beginning. For this purpose we use the ResetReader method which resets Enumerator to the list beginning. The class constructor is implemented so as to fill in all the required properties and to prepare the class for further use.
public ReportReader(string path) { // load the document document.Load(path); // Get file creation date Created = DateTime.ParseExact(document["Optimisation_Report"].Attributes["Created"].Value, "dd.MM.yyyy HH:mm:ss", null); // Get enumerator enumerator = document["Optimisation_Report"]["Optimisation_Results"].ChildNodes.GetEnumerator(); // Parameter receiving function string xpath(string Name) { return $"/Optimisation_Report/Optimiser_Settings/Item[@Name='{Name}']"; } // Get path to the robot RelativePathToBot = document.SelectSingleNode(xpath("Bot")).InnerText; // Get balance and deposit currency XmlNode Deposit = document.SelectSingleNode(xpath("Deposit")); Balance = Convert.ToDouble(Deposit.InnerText.Replace(",", "."), formatInfo); Currency = Deposit.Attributes["Currency"].Value; // Get leverage Leverage = Convert.ToInt32(document.SelectSingleNode(xpath("Leverage")).InnerText); }
First of all it loads the passed document and fills its creation date. Enumerator obtained during class instantiation belongs to the document child nodes located under section Optimisation_Report/Optimisation_Results, i.e. to the nodes having tag <Result/>. To get the desired optimizer configuration parameters, path to the required document node is specified using xpath markup. An analogue of this built-in function having a shorter path is the SelectItem method, which indicates the path to an item among document nodes having tag <Item/> according to its Name attribute. The GetDay method converts the passed document node to the appropriate structure of the daily trading report. The last method in this class is the data reader method. Its implementation in brief form is shown below.
public bool Read() { if (enumerator == null) return false; // Read the next item bool ans = enumerator.MoveNext(); if (ans) { // Current node XmlNode result = (XmlNode)enumerator.Current; // current report item ReportItem = new ReportItem[...] // Fill the robot parameters foreach (XmlNode item in result.ChildNodes) { if (item.Name == "Item") ReportItem.Value.BotParams.Add(item.Attributes["Name"].Value, item.InnerText); } } return ans; }
The hidden code part contains the optimization report instantiation operation and filling of the report fields with the read data. This operation includes similar actions, which convert the string format to the required one. Further loop fills in the robot parameters using data read row by row from the file. This operation is only performed if the completing file line has not been reached. The result of the operation is the returning of an indication of whether the row was read or not. It also serves as an indication of reaching the end of the file.
Multifactor Filtering and Sorting of the Optimization Report
To meet the objectives, I created two enumerations that indicated the sorting direction (SortMethd and OrderBy). They are similar and probably only one of them could be enough. However, in order to separate filtering and sorting methods, two enumerations were created instead of one. The purpose of the enumerations is to show the ascending or descending order. The ratio type of the coefficients with the passed value is indicated by flags. The purpose is to set the comparison condition.
/// <summary> /// Filtering type /// </summary> [Flags] public enum CompareType { GraterThan = 1, // greater than LessThan = 2, // less than EqualTo = 4 // equal }
Type of coefficients by which the data can be filtered and sorted are described by the aforementioned enumeration OrderBy. Sorting and filtering methods are implemented as methods expanding collections inherited from the IEnumerable<OptimisationResult> interface. In the filtering method, we check each of the coefficients item by item, whether it meets the specified criteria, and reject the optimization passes in which any of the coefficients does not meet the criteria. To filter data we use the Where loop contained in the IEnumerable interface. The method is implemented as follows.
/// <summary> /// Optimization filtering method /// </summary> /// <param name="results">Current collection</param> /// <param name="compareData">Collection of coefficients and filtering types</param> /// <returns>Filtered collection</returns> public static IEnumerable<OptimisationResult> FiltreOptimisations(this IEnumerable<OptimisationResult> results, IDictionary<SortBy, KeyValuePair<CompareType, double>> compareData) { // Result sorting function bool Compare(double _data, KeyValuePair<CompareType, double> compareParams) { // Comparison result bool ans = false; // Comparison for equality if (compareParams.Key.HasFlag(CompareType.EqualTo)) { ans = compareParams.Value == _data; } // Comparison for 'greater than current' if (!ans && compareParams.Key.HasFlag(CompareType.GraterThan)) { ans = _data > compareParams.Value; } // Comparison for 'less than current' if (!ans && compareParams.Key.HasFlag(CompareType.LessThan)) { ans = _data < compareParams.Value; } return ans; } // Sorting condition bool Sort(OptimisationResult x) { // Loop through passed sorting parameters foreach (var item in compareData) { // Compare the passed parameter with the current one if (!Compare(x.GetResult(item.Key), item.Value)) return false; } return true; } // Filtering return results.Where(x => Sort(x)); }
Two functions are implemented inside the method, each of them performs its own part of data filtering task. Let's view them, starting with the final function:
- Compare — its purpose is to compare the passed value presented as KeyValuePair and the value specified in the method. In addition to greater/less and equality comparison, we may need to check other conditions. For this purpose we will utilize flags. A flag is one bit, while the int field stores 8 bits. Thus we can have up to 8 simultaneously set or removed flags for the int field. Flags can be checked sequentially, without the need to create multiple loops or huge conditions, and thus we only have three conditions. Moreover, in the graphical interface which we will consider later it is also convenient to use flags to set required comparison parameters. We sequentially check the flags in this function and also check whether data correspond to these flags.
- Sort: unlike the previous method, this one is designed to check multiple written parameters instead of one. We run an item-wise loop through all flags passed for filtering and use the previously described function to find out whether the selected parameter meets the specified criteria. To enable access to a value of the specific selected item in the loop without using the "Switch case" operator, the aforementioned OptimisationResult.GetResult(OrderBy item) method is used. If the passed value does not match the requested one, return false and thus discard unsuitable values.
The 'Where' method is used to sort data. It automatically generates a list of suitable conditions, which is returned as the extension method execution result.
Data filtering is quite easy to understand. Difficulties may occur with sorting. Let us consider the sorting mechanism using an example. Suppose we have Profit Factor and Recovery Factor parameters. We need to sort data by these two parameters. If we perform two sorting iterations one after another, we will still receive data sorted by the last parameter. We need to compare these values in some way.
| Profit | Profit factor | Recovery factor |
|---|---|---|
| 5000 | 1 | 9 |
| 15000 | 1.2 | 5 |
| -11000 | 0.5 | -2 |
| 0 | 0 | 0 |
| 10000 | 2 | 5 |
| 7000 | 1 | 4 |
These two coefficients are not normalized within their boundary values. Also they have very wide range of values relative to each other. Logically, we first need to normalize them while preserving their sequence. The standard way to bring data to a normalized form is to divide each of them by the maximum value in the series: thus we will obtain a series of values that vary in the range [0;1]. But first, we need to find the extreme points of this series of values presented in the table.
| Profit factor | Recovery factor | |
|---|---|---|
| Min | 0 | -2 |
| Max | 2 | 9 |
As can be seen from the table, Recovery factor has negative values and thus the above approach is not suitable here. In order to eliminate this effect, we simply shift the entire series by a negative value taken modulo. Now we can calculate the normalized value of each of the parameters.
| Profit | Profit factor | Recovery factor | Normalized sum |
|---|---|---|---|
| 5000 | 0.5 | 1 | 0.75 |
| 15000 | 0.6 | 0.64 | 0.62 |
| -11000 | 0.25 | 0 | 0.13 |
| 0 | 0 | 0.18 | 0.09 |
| 10000 | 1 | 0.64 | 0.82 |
| 7000 | 0.5 | 0.55 | 0.52 |
Now that we have all the coefficients in the normalized form, we can use the weighted sum, in which the weight is equal to one divided by n (here n is the number of factor being weighted). As a result we obtain a normalized column which can be used as the sorting criteria. If any of the coefficients should be sorted in the descending order, we need to subtract this parameter from one and thus to swap the largest and the smallest coefficients.
The code implementing this mechanism is presented as two methods, the first one of which indicates the sorting order (ascending or descending), and the second method implements the sorting mechanism. The first of the methods, SortMethod GetSortMethod(SortBy sortBy), is quite simple, so let's move on to the second method.
public static IEnumerable<OptimisationResult> SortOptimisations(this IEnumerable<OptimisationResult> results, OrderBy order, IEnumerable<SortBy> sortingFlags, Func<SortBy, SortMethod> sortMethod = null) { // Get the unique list of flags for sorting sortingFlags = sortingFlags.Distinct(); // Check flags if (sortingFlags.Count() == 0) return null; // If there is one flag, sort by this parameter if (sortingFlags.Count() == 1) { if (order == OrderBy.Ascending) return results.OrderBy(x => x.GetResult(sortingFlags.ElementAt(0))); else return results.OrderByDescending(x => x.GetResult(sortingFlags.ElementAt(0))); } // Form minimum and maximum boundaries according to the passed optimization flags Dictionary<SortBy, MinMax> Borders = sortingFlags.ToDictionary(x => x, x => new MinMax { Max = double.MinValue, Min = double.MaxValue }); #region create Borders min max dictionary // Loop through the list of optimization passes for (int i = 0; i < results.Count(); i++) { // Loop through sorting flags foreach (var item in sortingFlags) { // Get the value of the current coefficient double value = results.ElementAt(i).GetResult(item); MinMax mm = Borders[item]; // Set the minimum and maximum values mm.Max = Math.Max(mm.Max, value); mm.Min = Math.Min(mm.Min, value); Borders[item] = mm; } } #endregion // The weight of the weighted sum of normalized coefficients double coef = (1.0 / Borders.Count); // Convert the list of optimization results to the List type array // Since it is faster to work with List<OptimisationResult> listOfResults = results.ToList(); // Loop through optimization results for (int i = 0; i < listOfResults.Count; i++) { // Assign value to the current coefficient OptimisationResult data = listOfResults[i]; // Zero the current sorting factor data.SortBy = 0; // Loop through the formed maximum and minimum borders foreach (var item in Borders) { // Get the current result value double value = listOfResults[i].GetResult(item.Key); MinMax mm = item.Value; // If the minimum is below zero, shift all data by the negative minimum value if (mm.Min < 0) { value += Math.Abs(mm.Min); mm.Max += Math.Abs(mm.Min); } // If the maximum is greater than zero, calculate if (mm.Max > 0) { // Calculate the coefficient according to the sorting method if ((sortMethod == null ? GetSortMethod(item.Key) : sortMethod(item.Key)) == SortMethod.Decreasing) { // Calculate the coefficient to sort in descending order data.SortBy += (1 - value / mm.Max) * coef; } else { // Calculate the coefficient to sort in ascending order data.SortBy += value / mm.Max * coef; } } } // Replace the value of the current coefficient with the sorting parameter listOfResults[i] = data; } // Sort according to the passed sorting type if (order == OrderBy.Ascending) return listOfResults.OrderBy(x => x.SortBy); else return listOfResults.OrderByDescending(x => x.SortBy); }
If sorting is to be performed by one parameter, execute sorting without resorting to the normalization of the series. Then immediately return the result. If sorting is to be performed by several parameters, we first generate a dictionary consisting of maximum and minimum values of the considered series. This allows accelerating the calculations, since otherwise we would need to request parameters during each iteration. This would generate much more loops than we have considered in this implementation.
Then, weight is formed for the weighted summation, and an
operation is performed to normalize a series to its sum. Here two loops are used again, the above described operations are performed in the
internal loop. The resulting weighted sum is added to the
SortBy variable of the appropriate array element. At the end of this
operation, when the resulting coefficient to be used for sorting has already been formed, use the previously described sorting method via
the standard
List<T>.OrderBy or List<T> array method. OrderByDescending — when
descending sorting is needed. Sorting method for separate members of the weighted sum is set by a
delegate passed as one of the function parameters.
If this delegate is left as a default parametrized value, the earlier mentioned method is used; otherwise the passed delegate is
used.
Conclusion
We have created a mechanism that will be actively used within our application in the future. In addition to the unloading and reading of xml files of a custom format, which store structured information about performed tests, the mechanism contains C# collection expanding methods, which are used to sort and filter data. We have implemented the multi-factor sorting mechanism, which is not available in the standard terminal tester. One of the advantages of the sorting method is the ability to account for a series of factors. However, its disadvantage is that the results can only be compared within the given series. It means that the weighted sum of the selected time interval cannot be compared with other intervals, because each of them uses an individual series of coefficients. In the next articles, we will consider the algorithm conversion method to enable the application or an automated optimizer for the algorithms, as well as the creation of such an automated optimizer.
Translated from Russian by MetaQuotes Ltd.
Original article: https://www.mql5.com/ru/articles/7290
Library for easy and quick development of MetaTrader programs (part XXIV): Base trading class - auto correction of invalid parameters
Library for easy and quick development of MetaTrader programs (part XXIII): Base trading class - verification of valid parameters
Extending Strategy Builder Functionality
Library for easy and quick development of MetaTrader programs (part XXII): Trading classes - Base trading class, verification of limitations
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
So before I can use this program, will I have to convert using Visual Studio?
Hello!
So before I can use this program, will I have to convert using Visual Studio?
Hello. This is the first article from 5 parts that are already published. At the 4rth part - there is full program. And yes, you have to compile some code it visual studio. If say about code that where ateched to the current article - so yes, it must be compiled in visual studio.