Intelligent Systems 2020

This is the practical material for the Intelligent Systems course, based on the card based strategy game Schnapsen.

Getting started

To get to know the concept of the game, please visit this web page.

Your job is to make a bot that will play this game.

Particularities of our implementation

For this implementation of the game, we followed the rules described in the above link, with a few caveats:

• Partial information about the state of the deck is updated automatically, so you don't have to keep track of it yourself.
• In the same vein, it is not the player's responsibility to keep track of their (and their enemy's) points, as it would be in regular Schnapsen. This is done automatically through the game engine, which also removes the aspect of having to declare that you have reached 66 points in order to win the game.
• "Closing the talon" is not implemented in order to have a clear separation between the perfect and imperfect information parts of the game, and also in order to avoid further increasing the branching factor.
• Scoring is implemented as the rules would suggest. A player can receive 1-3 points for winning a round, depending on the score differential. However, play.py and tournament.py play the game only in terms of these rounds, not in terms of a "full game" described as playing rounds until one player reaches 7 points.

Technical requirements

You require a working Python 3.x environment and a good text editor or an IDE. You can download Python 3.7 for your machine from the following sources:

• Windows Note that the code runs significantly slower on Windows due to the way process management is implemented.
• MacOS
• Linux
• Other

For the regular Windows Python installation, be sure to select the "Add Python 3.x to PATH" option in the installation dialogue, so that you can access Python through your command line.

For more advanced Windows users that want to use the Linux bash without a virtual machine or dual-booting, consider using Windows Subsystem for Linux. Once set up, this lets you use the command line interface of the respective Linux distribution you chose, with access to the Windows File System. For Linux newcomers, we suggest using Ubuntu. You can then install Python3 and pip through the following command, in case they are not pre-installed. For other distributions, use the equivalent package manager.

sudo apt install python3 python3-pip

The core game engine runs on pure Python, however you will need to install a few additional packages for tasks throughout the course. This is most easily done through the standard Python package manager, pip. You will most likely already have pip after the Python 3.x installation. You can check by running the following command in your command line interface.

pip -V

If this fails, you can find installation instructions here. Once installed, pip can be accessed from the command line interface, and the required packages can be obtained through the following command:

pip install sklearn matplotlib flask

Visual interface

We provide a visual interface which you can use to familiarize yourself with the game, and to play against your own bots later on.

The following example starts a local server and sets rdeep, an already implemented bot, as the opponent.

python visual/server.py --opponent rdeep

The "--opponent" flag is followed by the name of the bot you want to play against.

To see a full list of parameters and their usage, run:

python visual/server.py --help

Once your server is up and running, you can fire up your favorite (modern) web browser and visit http://127.0.0.1:5000/, the local address where your server is listening.

The controls are fairly straightforward; click on the card that you want to play to place it on the table. You can decide to play this move by clicking the "Submit move" button on the top bar, or you can change your mind with "Reset move". Whenever you have a complex move available, such as a marriage or a trump jack exchange, the corresponding buttons on the top bar will become available. Finally, you can use SHIFT + R to start a new game at any time.

Python knowledge

You will of course also need a working knowledge of python. If you're looking to brush up on your skills, there are many good tutorials available. For instance:

• https://www.learnpython.org/
• https://www.codecademy.com/

You do not need to be an expert in python to write a functioning bot. If you already know another programming language, you should be able to get going within a day. You'll pick up the details as the project progresses. However, there are a few things that are important to understand. Check if you know what the following mean. If not, take some time to google them and read up:

Call-by-reference (and "call-by-value")

What happens if I pass a function a 'State' object, and the function changes the object? Do I keep an unchanged state, or does my state change as well?

Object oriented programming

What's the difference between a class and an object? How are these expressed in python? What does the self keyword do?

Recursion

Briefly: a method calling itself. Why would this useful, and how does it work?

List comprehensions

Advanced python, but they occur occasionally in the code. Useful to know.

Examples

Here are some quick use cases and solutions to help you get a feel for the code.

Get the size of the stock

Let 'state' be the state you're given and let's say you want the size of the stock. Then the following a should do the trick:

size_of_stock = state.get_stock_size()

Find out if I'm player 1 or 2

me = state.whose_turn()

Print the (abbreviated) cards in your hand

cards_hand = state.hand()

for i, card in enumerate(cards_hand):

	rank, suit = util.get_card_name(card)

	print('Card {} in the hand is {} of {}'.format(i, rank,suit))

The deck of cards is represented through a list. Each index corresponds to a different card, as per the table below.

	Aces	10s	Kings	Queens	Jacks
Clubs	0	1	2	3	4
Diamonds	5	6	7	8	9
Hearts	10	11	12	13	14
Spades	15	16	17	18	19

Generate a random state

state = State.generate()

# To deterministically generate the same state each time, the generate method can also take a seed, like so:

state = State.generate(25)
# This will always generate the same starting state, to make testing/debugging your bots easier.
# Note that any two states generated with the same seed will be identical, and 25 is only used here as an example.

Check if two states are identical

state = State.generate(1)

# same_state is not the same object as state,
# but all attributes are identical.
same_state = State.generate(1)

diff_state = State.generate(2)

# The equality and inequality operators are overridden for State
# objects, so you can check if all parameters of two states match.

state == same_state # Evaluates to True
state == diff_state # Evaluates to False

Print a representation of the generated state

>>> print(state)

The game is in phase: 1
Player 1's points: 0, pending: 0
Player 2's points: 0, pending: 0
The trump suit is: C
Player 1's hand: QC JD 10H JH 10S
Player 2's hand: 10C AD 10D KH JS
There are 10 cards in the stock

Get own/opponent's points

me = state.whose_turn()
opponent = util.other(me)

own_points = state.get_points(me)
opponents_points = state.get_points(opponent)

Get familiar with the State API

Every state-related function you will use when building your bot can be found, fully documented, in the State class, located in api/_state.py. We highly recommend that you read through this class to understand the capabilities available to you when writing your bots.

Note that you only have access to public functions. Private functions, i.e. functions whose name starts with two underscores "__" are used for the internal implementation of the game and are abstracted away from the player.

Reading the code itself in addition to the documentation can help you get acquainted with the internals of the game engine, however this is not obligatory in order to be able to complete the course.

FAQ

I found a bit that could be implemented much better/more efficiently.

Our main goal was to write code that was easy to read and to understand. To achieve this, we've made many methods less efficient than they need to be. This is especially important for a project like this where many of the students are novice programmers. It is also a good principle in general, at least when you write the first version of your code.

You may feel that your bot is too slow with our State objects, for instance if you're creating an evaluating lots of State objects in a deep tree. Luckily, you're not tied to our API: simply take the State object you're given and copy it to your own, more efficient, implementation. This may get you another plie or two in the search tree, so if you really want to win the competition it might be worth it.

I found a bug/improvement. Can I fork the project and send a pull request?

Sure! Just remember this is not a regular project: we've tried to minimize the amount of advanced python, and the number of dependencies. So, it might be that we're aware of the potential improvement, but we haven't used it just to keep the code simple for novice programmers.

The command-line scripts (play.py, tournament.py) make it difficult to do X

The command line scripts provide a convenient starting point, but if you want to do something more complex (like try a range of parameters for your bot), they are probably too limited.

Your best bet is to write your own script that does what you want, and have it call the engine. Have a look at the function play(...) in api/engine.py, or have it run a by itself. See experiment.py for an example.

Changes from last year's challenge

Bots from last year can be used this year as well.

Attributions

Vectorized Playing Card Graphics 2.0 - http://sourceforge.net/projects/vector-cards/
Copyright 2015 - Chris Aguilar - [email protected]
Licensed under LGPL 3 - www.gnu.org/copyleft/lesser.html