Lab Assignment 2

Part 1 [7 points]
CSE422: Artificial Intelligence [C02]

You are developing an AI-driven cryptocurrency trading bot that optimizes its
trading strategy using a Genetic Algorithm (GA). The goal is to maximize profit
while minimizing risk by evolving key trading parameters such as:
●​ Stop-Loss (%) – The percentage at which a position is automatically closed
to prevent further loss.
●​ Take-Profit (%) – The percentage at which a position is closed to secure
profits.
●​ Trade Size (%) – The portion of available capital allocated per trade.

Your task is to implement a GA-based approach to find the optimal set of trading
parameters that generate the highest profit over a given set of historical price
movements.

Chromosome Representation (Encoding):

{"stop_loss": 2, "take_profit": 5, "trade_size": 20},

{"stop_loss": 3, "take_profit": 7, "trade_size": 30},
{"stop_loss": 07, "take_profit": 4, "trade_size": 25},
{"stop_loss": 08, "take_profit": 6, "trade_size": 15}

String Representation of the chromosomes:

020520 [Explanation: 02 represents 2% stop_loss, 05 represents 5% take_profit
and 20 represents 20% of the total capital amount]
030730
070425
080615
 Fitness Calculation:
Initial Capital: $1000
Let's evaluate the chromosome 020520 which stands for:
Chromosome (Trading Strategy):

●​ Stop-Loss: 2%
●​ Take-Profit: 5%
●​ Trade Size: 20% of capital per trade

on the following historical price changes in percentage:

[-1.2, 3.4, -0.8, 2.1, -2.5, 1.7, -0.3, 5.8, -1.1, 3.5]
Day Price Trade Size Exit Profit/Loss Updated Capital
Change (%) ($) Condition ($) ($)

1 -1.2 200.00 No SL/TP hit -2.40 997.60

2 3.4 199.52 No SL/TP hit +6.78 1004.38

3 -0.8 200.88 No SL/TP hit -1.61 1002.77

4 2.1 200.55 No SL/TP hit +4.21 1006.98

5 -2.5 201.39 Stop-Loss hit -4.03 1002.95

6 1.7 200.59 No SL/TP hit +3.41 1006.36

7 -0.3 201.27 No SL/TP hit -0.60 1005.76

8 5.8 201.15 Take-Profit hit +10.06 1015.82

9 -1.1 203.164 No SL/TP hit -2.23 1013.59

10 3.5 202.72 No SL/TP hit +7.10 1020.69

[Note: We start our trading on Day-01 with $1000 capital and invest $200 since
the trade size is 20%. Then after incurring a loss of -$2.40, our capital reduces to
997.60 ($1000-$2.40). Then for the 2nd day we again start our trade with 20% of
the updated capital which is 997.60*20% = $199.52
Then on Day-5 we see the price change(%) is -2.5 which is greater than the
stop_loss threshold, so we cap the loss at 2% instead of 2.5%, thus the loss for
Day-05 is 201.39 * 0.02 = 4.03
On Day -8, we see the price change(%) is +5.8% which is greater than the
take-profit threshold, so we cap the profit at 5% instead of 5.8%, thus the profit for
Day-08 is 201.15*0.05 = 10.06]

Final Fitness Score

The final capital after all trades: $1020.69

Fitness Score = (Final Capital - Initial Capital)

Fitness = $1020.69 - $1000 = 20.69
This chromosome has a fitness score of 20.69, which represents its profitability.

Summary

●​ The fitness function is the total profit after simulated trading.

●​ A chromosome represents a trading strategy with stop-loss, take-profit, and
trade size.
●​ Each chromosome is evaluated by simulating trading over historical price
data.
●​ The fitness score is the profit gained from the starting capital.
●​ Higher fitness values indicate better trading strategies.

Task Breakdown:

Step 1: Define the Chromosome Structure

Each chromosome represents a trading strategy consisting of three key
parameters:

Gene Description Range

Stop-Loss (%) Maximum loss before auto-closing a trade 01% - 99%

Take-Profit (%) Profit threshold before closing a trade 01% - 99%

Trade Size (%) Percentage of capital allocated per trade 01% - 99%

Step 2: Initialize Population

Generate an initial population of 4 random chromosomes.

Each chromosome is created with random values within the defined ranges.

The population size should be the same for every generation.

Step 3: Evaluate Fitness (Profit Calculation)

The fitness function determines how profitable a strategy is over historical price
movements.

Fitness Function Calculation:

For each chromosome:

1.​ Start with $1000 initial capital.

2.​ Simulate trades using historical price movements.
3.​ Apply stop-loss and take-profit rules.
4.​ Calculate final capital and return profit as fitness score.

Step 4: Select Parents (Random Selection)

Pick two random individual chromosome to produce two offspring

Step 5: Crossover (Recombine Parent Genes)

We combine genes from two parents to create an offspring.

●​ Use single-point crossover:

○​ Pick a random split point.
○​ Mix genes from both parents.

Step 6: Mutation (Introduce Random Changes)

Mutation ensures genetic diversity and prevents the algorithm from getting stuck
in local optima.

●​ Small random changes to genes with a low probability (e.g., 5% mutation

rate).

Step 7: Generate New Population (Next Generation)

●​ Select the best individuals (elitism).

●​ Crossover + Mutation to create new individuals.
●​ Repeat until reaching a termination condition (e.g., max generations=10).

Input
“Capital to Start With”: $1000
"historical_prices": [-1.2, 3.4, -0.8, 2.1, -2.5, 1.7, -0.3, 5.8, -1.1, 3.5],
"initial_population": [
{"stop_loss": 2, "take_profit": 5, "trade_size": 20}, {"stop_loss": 3, "take_profit": 7,
"trade_size": 30},
{"stop_loss": 1.5, "take_profit": 4, "trade_size": 25},
{"stop_loss": 2.5, "take_profit": 6, "trade_size": 15} ],
"generations": 10

Output

"best_strategy":
{ "stop_loss": 2.3, "take_profit": 6.2, "trade_size": 22 },
"Final_profit" : 12.5

Part 2 [3 points]

For this part randomly select two parents from the initial population of your
problem statement. Then perform a two-point crossover to generate two
children. The two points have to be chosen randomly, but it has to be made sure
the second point always comes after the first point.

Here is an example of how two-point crossover works:

Parent 1: 000111000
Parent 2: 111000111

For two points crossover, we have randomly chosen the following points:
1st point:- between index 2 and index 3
2nd point:- between index 6 and index 7

So the two resultant offsprings are, 000000100 & 111111011

[In this part, you just need to iterate once and print the resultant offspring after
doing the crossover]

Part 3 [0 points]

In part 1, you selected parents through random sampling from the initial
population. Another advanced technique for parent selection is known as
Tournament Selection. Please take some time to research and understand this
method at home. Might be helpful in the near future!