Description
Description
In this assignment, you will write a program that plays the part of a “smart building”. This program will be
given a real-time stream of sensor data from throughout the building, and use this data to decide whether
to turn on the lights in each room. Your goal is to minimise the cost of lighting in the building, while also
trying to make sure that the lights stay on if there are people in a room. Every 15 seconds, you will receive a
new data point and have to decide whether each light should be turned on or off. There are several types
of sensors in the building, each with different reliability and data output. You will be given a file called
data.csv containing one day of complete data with all sensor values and the number of people in each room.
This assignment can be approached in many different ways. We will not be giving any guidance on what
algorithms are most appropriate.
Your solution must include a Probabilistic Graphical Model as the core component. Other than that you are
free to use any algorithm as part of your approach, including any algorithm available in Python’s sklearn
library.
It is recommended you start this assignment by discussing several different possible approaches with your
partner. Make sure you discuss what information you have available, what information is uncertain, and
what assumptions it may be reasonable to make.
Every area on the floor plan is named with a string of the form ‘r’, ‘c’, ‘o’, or ‘outside’. ‘r’,‘c’ and ‘o’ stand
for room, corridor, and open area respectively.
Data
The file data.csv contains complete data that is representative of a typical weekday in the office building.
This data includes the output of each sensor as well as the true number of people in each room. This data
was generated using a simulation of the building, and your program will be tested against many days of data
generated by the same simulation. Because this data would be expensive to collect, you are only given 2400
complete data points, from a single workday. The simulation attempts to be a realistic approximation to
reality, so it includes many different types of noise and bias. You should treat this project as if the data came
from a real office building, and is to be tested on real data from that building. You can make any assumptions
that you think would be reasonable in the real world, and you should describe all assumptions in the report.
Part of your mark will be determined by the feasibility of your assumptions, if applied to the real world.
Added Nov 2nd: [** The number of people who come to the office each day varies according to
this distribution: num_people = round(Normal(mean=20, stddev=1)). This information was
obtained from records of the number of workers present each day, and the empirical distribution of num_people was found to be identical to round(Normal(mean=20, stddev=1)).**]
Data format specification
Sensor data
Your submission file must contain a function called get_action(sensor_data), which receives sensor data
in the following format:
sensor_data = {‘reliable_sensor1’: ‘motion’, ‘reliable_sensor2’: ‘motion’,
‘reliable_sensor3’: ‘motion’, ‘reliable_sensor4’: ‘motion’,
‘unreliable_sensor1’: ‘motion’, ‘unreliable_sensor2’: ‘motion’,
‘unreliable_sensor3’: ‘motion’, ‘unreliable_sensor4’: ‘motion’,
‘door_sensor1’: 0, ‘door_sensor2’: 0, ‘door_sensor3’: 0, ‘door_sensor4’:0,
‘robot1’: (‘r1’, 0), ‘robot2’: (‘r16’, 0), ‘time’: datetime.time(8, 0), ‘electricity_price’: 0.81}
Added Oct 30th: [The motion and door sensors report on motion from the entire previous 15
seconds, but the robot reports an instantaneous count of the number of people.]
2
The possible values of each field in sensor_data are:
• reliable_sensors and unreliable_sensors can have the values [‘motion’, ‘no motion’]. All reliable_sensors
are of the same brand and are usually quite accurate. unreliable_sensors are a different type of motion
sensor, which you know tends to be a little less accurate.
• door_sensors count how many people passed through a door (in either direction), so it can be any
integer.
• The robot sensors are robots that wander around the building and count the number of people in each
room. The value is a 2-tuple of the current room, and the number of people counted. I.e. if the robot
goes into r4 and counts 8 people, it would have the value (‘r4’,8). If it goes into room ‘c2’ and no one is
present, it would have value (‘c2’,0).
• Any of the sensors may fail at any time, in which case they will have the value None. They may start
working again.
The value of time is a datetime.time object representing the current time. Datapoints will be provided in 15
second resolution, i.e., your function will be fed data points from 15 second intervals from 8 am – 6 pm.
Training data
The file data.csv contains a column for each of the above sensors, as well as columns for each room, which
tell you the current number of people in that room. The columns of data.csv are the following and can be
divided into two groups:
1. Columns that represent readings from sensors, as described in the previous section: reliable_sensor1,
reliable_sensor2, reliable_sensor3, reliable_sensor4, unreliable_sensor1, unreliable_sensor2, unreliable_sensor3, unreliable_sensor4, robot1, robot2, door_sensor1, door_sensor2, door_sensor3,
door_sensor4, time, electricity_price.
2. Columns that are present only in the training data and provide the ground truth with the number
of people in each room, corridor, open area, and outside the building: r1, r2, r3, r4, r5, r6, r7, r8,
r9, r10, r11, r12, r13, r14, r15, r16, r17, r18, r19, r20, r21, r22, r23, r24, r25, r26, r27, r28, r29, r30,
r31, r32, r33, r34, r35, c1, c2, c3, c4, o1, outside. Added Oct 30th: [This ground truth data
provides the instantaneous count of people per room (i.e. Every 15 seconds, a snapshot
of each room is magically taken at exactly the same time, and the number of people in
each room is counted. If someone passes through multiple rooms within 15 seconds, they
will not increment the count in multiple rooms, only in one room) ].
Note that the first column of data.csv is the index, and has no name.
You should use this data to learn the parameters of your model. Also, you can save the parameters to csv
files that can be loaded during testing.
Action data
get_action() must return a dictionary with the following format. Note that every numbered room named
“r” in the building has lights that you can turn on or off. All other rooms/corridors have lights that are
permanently on, which you have no control over, and which do not affect the cost.
actions_dict = {‘lights1’: ‘off’, ‘lights2’: ‘off’, ‘lights3’: ‘off’,
‘lights4’: ‘off’, ‘lights5’: ‘off’, ‘lights6’: ‘off’, ‘lights7’: ‘off’,
‘lights8’: ‘off’, ‘lights9’: ‘off’, ‘lights10’: ‘off’, ‘lights11’: ‘off’,
‘lights12’: ‘off’, ‘lights13’: ‘off’, ‘lights14’: ‘off’, ‘lights15’: ‘off’,
‘lights16’: ‘off’, ‘lights17’: ‘off’,’lights18′: ‘off’, ‘lights19’: ‘off’,
‘lights20’: ‘off’, ‘lights21’: ‘off’, ‘lights22’: ‘off’, ‘lights23’: ‘off’,
‘lights24’: ‘off’, ‘lights25’: ‘off’, ‘lights26’: ‘off’, ‘lights27’: ‘off’,
‘lights28’: ‘off’, ‘lights29’: ‘off’,’lights30′: ‘off’, ‘lights31’: ‘off’,
‘lights32’: ‘off’, ‘lights33’: ‘off’, ‘lights34’: ‘off’,’lights35′: ‘off’}
3
The outcome space of all actions is (‘on’,’off’).
In the provided example_solution.py, there is an example code stub that shows an example of how to set
up your code.
Figure 1 shows the floor plan specification.
Cost specification
If a light is on in a room for 15 seconds, it usually costs you about 1 cent. The exact price of electricity goes
up and down, but luckily, the electricity provider lists the current price online, and this price is included
in the sensor_data. If there are people in a room and there is no light on, it costs you 4 cents per person
every 15 seconds, because of lost productivity. Added Oct 30: [The cost can be calculated exactly
using the complete training data, so it is also based on an instantaneous count of the number
of people in each room.]
Your goal is to minimise the total cost of lighting plus lost productivity, added up over the whole day. You do
not need to calculate this cost, the testing code will calculate it using the actions returned by your function,
and the true locations of people (unavailable to you). The file example_test.py shows exactly how the cost
is calculated.
Testing specification
Your program must be submitted as a python file called solution.py. During testing, solution.py will be
placed in a folder with test.py. A simpler version of test.py has been provided (called example_test.py),
so you can confirm that testing will work. A more elaborate version of test.py will be used to grade your
solution.
Report
Your report should cover the following points:
• What algorithms you used, a brief description of how they work, and their time complexity.
• A short justification of the methods you used (if you tried different variations, describe them).
• Any assumptions you made when creating your model.
The report must be less than 2000 words (around 4 pages of text). The only accepted format is PDF.
Marking Criteria
This assignment will be marked according to the following criteria:
1. 50% of the mark will be determined by the cost incurred by your code after several days of simulated
data. The mapping from cost to marks will be determined after the assignment has been submitted.
2. 20% of the mark will be determined by the description of the algorithms used, and a short justification
of the methods used.
3. 10% of the mark will be determined by a description of the assumptions and/or simplifications you
made in your model, and whether those assumptions would be effective in the real-world.
4. 20% of the mark will be determined by the quality, readability and efficiency of the code.
Items 2 and 3 will be assessed using the report. Items 1 and 4 will be assessed using python file.
4
Figure 1: Floor plan (note that dotted grey lines denote the boundaries of areas when the boundary is
unclear).
5
Bonus Marks
Bonus marks will be given to the top 10 performing programs (10 percentage points for 1st place, 1 percentage
point for 10th place).
6
