Backpropagation in MQL4 Neural Network EA

工作已完成

执行时间2 天
客户反馈
It was a pleasure working with Mehmet. Nothing was too much trouble. Professional and punctual and understands the subject (Neural networks and backpropagation in my case) I would recommend him.

指定

I need the backpropagation fixed in a Neural Network of an existing EA that will be used to take trades.


* Neural Network experience is essential * 
* Do not apply if you do not understand Backpropagation, Calculus (the chain rule), (Stochastic) Gradient descent and error functions *

95% of the code described below already exists.

The input features already exist in the code.

Neural network structure:
One input layer
Three hidden layers
One Output layer (2 to 4 outputs depending on activation)

REQUIREMENTS:
Backpropagation using stochastic gradient descent
Loss/Error function
Update gradients
Update weights and biases in file
The network must learn
The weights and biases gradients must not explode or vanish
possible regularisation and normalization adjustments
The MQL4 code must compile with no errors and no warnings (property strict) 

Clean existing code:
No libraries
No includes
No Errors
No Warnings
Clear well documented code
Variable names: use readable like trainingInputs[] rather than ti[] or x1[]

-------
The Existing code works like this:
Initialize:
Create random weights & biases in arrays
Write weights and biases to csv files

Train:
1) Load weights and biases from files into arrays (on each iteration)
2) Feed forward through NN (process input features)
3) Get a Buy or Sell signal (0 - 1) - EA takes a trade
4) Copy Feed Forward arrays to use at close of trade
(nothing happens until the trade closes)
5) When the trade closes, get the trade History Classify/Label the trade as buy/sell, profit/loss
6) Calculate error/cost function (using copied arrays)
7) Backpropagation (overwrite new adjusted weights and biases from updated arrays into files)
Repeat the process.

Trade:
Take trades using trained network
-------

Display on screen:
Trade quantity (considered as one labeled training example)
Last trade Ticket No.
Network output (predicted) eg. (0.2 | 0.8)
Actual (expected) output eg. ( 0 | 1) 
Trade type: Buy/Sell/Hold
Trade result: M15 - GBPUSD | win/loss (1.2% SL and TP = 2.6 * SL) (12 pips SL and 32 pips TP)
% Error over the epoch - Loss/Trade quantity (I want to see if the network is learning)

Write to file:
Write the cost/iteration No. to file stats.csv (new line for each with epoch No.)  so I can see or use the data to plot a graph in excel of the learn rate

I would like the code written in the same or similar style to mine.. (so I can understand the code and make any small changes in the future)

95% of the code already exists.

Timeframe: 2 days from start of job. Only take the job if and when you can start and complete in 2 days.
The budget is $150 - $200.
PAYMENT ON COMPLETION OF WORKING CODE ONLY.
Request the MQL4 file

反馈

1
开发者 1
等级
(30)
项目
55
22%
仲裁
12
67% / 8%
逾期
2
4%
空闲
相似订单
I am looking for someone who has experience working with the ZigZag indicator. I want to create a multi timeframe dashboard scanner based on ZigZag. I will explain more in private messages. if you feel capable & interest with that,please pm me for more details. thanks
The ZONE RECOVERY/HEDGING STRATEGY will be on a timeframe where each strategy is independent and pending orders should be placed at the high and low of the candlestick. If it is a buy, then the pending order at the high of the candlestick will be triggered, and if it is a sell, then the pending order at the low of the candlestick will be triggered. If the market breaks the high of the candlestick, then the pending
Hi, i want to create EA that using martingale and hedge system working on metatrader 4 and has some options to be manually changed in the bot options i need expert in this strategy to do this job note that strategy is very simple no indicator needed unless moving average
I am looking for someone to Optimize and Backtest my robot on a monthly basis. Each month, I would expect: 4 new SET Files Each SET file would be based on recent history, so the tests would generally look at 3-6 months of data. Can be more or less that's up to you. You will test many settings and combinations of settings to find effective settings for the next month. Can test, different timesframes, different pairs
I have two not very sophisticated bots that need to be converted from python to MQL5. The bots are very similar, they use 99% of the same methodology, the only difference being that one is a trend-following bot, and the other is a mean reversion bot. I need the parameters to be at the top of the script so that they can be easily changed and optimised. The entire codebase is already done in python (sl,tp, trailing
The specifications cover 3 requests: 1- Addition of a Trailing Take Profit function 2- Addition of opening orders according to a decreasing distance (20, 15, 10, 5...pips) 3- Addition of a function that allows choosing a fixed, increasing, or decreasing take profit
Job Description: We are seeking an experienced EA programmer to create an EA that utilizes SnR + Trendlines + Multi timeframe confluence Project Requirements: - Support and Resistance, Open Close levels/kissing candles/Rejection block, Support broken becomes Resistance(SbR), Resistance broken becomes Support(RbS), Quasimodo Levels, Asian Range, London Killzone, London Open, New York Killzone, New York Open
I have an ea which is working according to ict liquidity sweep strategy . With sl and tp but I test it in strategy tester but the result is awful the sl is hitting more than tp so i want to change its sl to tp and tp to sl to set the result to its normal form is there any professional programmer to solve my inquiry ? If yes so bet for it thnk alot
Dears, i am looking for developer to make expert advisor in forex to me and i have a strategy , i need the ea to be customize and has some options to be changeable
The specifications cover 4 requests: 1- Addition of a Trailing Take Profit function 2- Addition of a opening orders according to a decreasing distance (20, 15, 10, 5...pips) 3- Addition of a function that allows choosing a fixed, increasing, or decreasing take profit. 4- E rror s correction in coding

项目信息

预算
150 - 200 USD
开发人员
135 - 180 USD
截止日期
 1  3 天