https://github.com/masanorihirano/pams/blob/main/pams/runners/sequential.py#L420

https://github.com/masanorihirano/plhamJ/tree/master/samples/src/main/java/plham/samples

• CI2002.ipynb
• cancel_test.ipynb
• dark_pool.ipynb
• dist_parallel.ipynb -> skip
• fat_finger.ipynb
• invest_div.ipynb
• market_share.ipynb
• price_limit.ipynb
• shock_transfer.ipynb
• trading_halt.ipynb
• black_scholes.ipynb
• multi_thread.ipynb -> skip

plots or other analysis is also necessary.

equivalent to https://github.com/plham/plhamJ/blob/master/core/src/main/java/plham/core/event/OrderMistakeShock.java

Let's consider the plan for checking stylized fact in financial market as validations of this package.

equivalent to https://github.com/plham/plhamJ/blob/master/core/src/main/java/plham/core/event/PriceLimitRule.java

see https://github.com/masanorihirano/pams/actions/runs/3428731428/jobs/5713969066
currently, excludes python 3.11 on windows

maxHifreqOrders was replaced to maxHighFrequencyOrders in pams. But, we should also support "maxHifreqOrders"

equivalent to https://github.com/plham/plhamJ/blob/master/core/src/main/java/plham/core/agent/TestAgent.java

We can drop python 3.7 support because google colab was updated to python 3.8
After dropping python 3.7,

• we can use Final typing
• possibly, we can follow PEP585 . It mean that we should use tuple, list dict, set, and type instead of Tuple, List, Dict, Set, Type.

Because of python, we cannot protect orders submitted once.
Therefore, agents can modify the submitted orders if they store submitted order instances.

pams simulation architecture limit the chance agent can modify the orders because usually agents can act something only in submit_orders methods.
However, we should consider the processing of modified orders.

Due to the original plhamJ implementation, currently there is no way to modify the submitted orders and only thing agents can do is cancel orders and submit new orders.

Note: todo is written here:

In the original plhamJ package, some getXXXXByName methods are implemented to simulator.

https://github.com/plham/plhamJ/blob/7702c3ac6372766fa3acdfc2870384ae941ff9cc/core/src/main/java/plham/core/main/Simulator.java#L216-L234

In pams, I think simulator.name2market can be a replacement.

But, if those methods are necessary for the certain purpose, we should consider the implementation again.
If not, we should drop the following ToDo:

for efficient usage of GPU, we should carefully consider the fork strategy or others.

• fundamentals.py
• index_market.py
• market.py
• order.py
• order_book.py
• session.py
• simulator.py
• agents/arbitrage_agent.py
• agents/base.py
• agents/fcn_agent.py
• agents/high_frequency_agent.py
• events/base.py
• events/funamental_price_shock.py
• logs/base.py
• logs/market_steps_loggers
• runners/base.py
• runners/sequential.py
• utils/class_finder.py
• utils/extends.py
• utils/json_random.py

• fundamentals.py
• index_market.py
• market.py
• order.py
• order_book.py
• session.py
• simulator.py
• agents/arbitrage_agent.py
• agents/base.py
• agents/fcn_agent.py
• agents/high_frequency_agent.py
• events/base.py
• events/funamental_price_shock.py
• logs/base.py
• logs/market_steps_loggers
• runners/base.py
• runners/sequential.py
• utils/class_finder.py
• utils/extends.py
• utils/json_random.py

Currently, we support the class finder in pams package.
However, for users' convenience, we should support this also for user-defined classes in the main python files such as main.py in samples.

The current implementation might support user-defined classes because we also look into the global and local scopes.
But, I'm not sure about this. Therefore, further investigation is required.

In the simulations, we should consider the rare case that cannot occur in actual markets.
In this case, we faced the case that the orders are placed at the totally same time.
This occurs only when the max_normal_orders is more than 1 or one agent submits the buy and sell order at the same time and executed them, aka. wash sales (仮装売買 in japanese), which is strictly prohibitted.

if init is missing in some components, the same code assigning instance does not work properly. It seems occurring this issue when google colab notebook run twice.

https://github.com/masanorihirano/plhamJ/tree/master/samples/src/main/java/plham/samples

• CI2002
• cancel_test
• dark_pool
• dist_parallel -> skip
• fat_finger
• invest_div
• market_share
• price_limit
• shock_transfer
• trading_halt
• black_scholes
• multi_thread -> skip

equivalent to https://github.com/plham/plhamJ/blob/master/core/src/main/java/plham/core/event/TradingHaltRule.java

Currently, classes are set by a string in config JSON/python dict.

When python dict is used for setting, we should support Type direction in the setting.

For example,

class UserDefinedAgent(Agent):
    pass

setting = {
    ....
    "Agents": {
        "class": UserDefinedAgent,
        ....
    }
}

runner = SequentialRunner(
    settings=config,
    ...
)

...

if it is possible, class_registration to the runner is not necessary.