ECON3203 Group Assignment (25 marks)
October 4, 2021
1 Background
Developing a predictive model for ATM cash demand is an important task for every bank. Sup-
pose that you are employed by a bank, and your task is to optimise the bank’s cash management
by making smarter decisions about reloading its ATM network.
The variable Withdraw in the dataset ATM_training.csv is the total cash amount withdrawn
per day from an ATM, recorded from the ATM network of a bank. The response variable and
covariate variables are described in the following table.
Variable Description
Withdraw The total cash withdrawn a day (in 1000 local currency)
Shops Number of shops/restaurants within a walkable distance (in 100)
ATMs Number of other ATMs within a walkable distance (in 10)
Downtown =1 if the ATM is in downtown, 0 if not
Weekday = 1 if the day is weekday, 0 if not
Center =1 if the ATM is located in a center (shopping, airport, etc), 0 if not
High =1 if the ATM has a high cash demand in the last month, 0 if not
Your task is to develop a model for predicting the cash demand Withdraw based on the covari-
ates.
The test dataset ATM_test.csv (not provided) has the same structure as the training data
ATM_training.csv.
1.1 Test error
For the measure of prediction accuracy, please use mean squared error (MSE), computed on the
test data. Let ŷi be the prediction of yi where yi is the i-th withdraw in the test data. The test error
is computed as follows
Test_error =
1
ntest
∑
yi∈test data
(ŷi − yi)2,
where ntest is the number of observations in the test data.
2 Submission Instructions
1. Each group needs to submit TWO files (or more if necessary) via the Moodle site (to avoid
repeated submissions, one and only one member of your group should be responsible for
submitting).
1
• A document file, named Group_xxx_document.pdf, that reports your data analysis
procedure and results. You should replace the xxx in the file name with your group ID.
• A Python file, named Group_xxx_implementation.ipynb that implements your data
analysis procedure and produces the test error. You might submit additional files that
are needed for your implementation, the names of these files must follow the same
format Group_xxx_.
2. About your document file Group_xxx_document.pdf
• Describe your data analysis procedure in detail: how the Exploratory Data Analysis
(EDA) step is done, what and why models/methods are used, how the models are
trained, etc. with sufficient justifications. The description should be detailed enough so
that other data scientists, who are supposed to have background in your field, under-
stand and are able to implement the task. All the numerical results are reported up to
four decimal places.
• Clearly and appropriately present any relevant graphs and tables.
• The page limit is 20 pages including EVERYTHING: appendix, computer output,
graphs, tables, etc.
3. The Python file is written using Jupyter Notebook, with the assumption that all the necessary
data files (ATM_training.csv and ATM_test.csv) are in the same folder as the Python file.
If you use deep learning models, then please assume that Keras (with Tensorflow backend)
has been installed.
• If the training of your model involves generating random numbers, the random seed
in Group_xxx_implementation.ipynb must be fixed, e.g. np.random.seed(0), so that
the marker expects to have the same results as you had.
• The Python file Group_xxx_implementation.ipynb must include the following code
import pandas as pd
ATM_test = pd.read_csv('ATM_test.csv')
# YOUR CODE HERE: code that produces the test error test_error
print(test_error)
The idea is that, when the marker runs Group_xxx_implementation.ipynb, with the
test data ATM_test.csv in the same folder as the Python file, he/she expects to see the
same test error as you would if you were provided with the test data. The file should
contain sufficient explanations so that the marker knows how to run your code.
• In case you want to test your code to see if a test error is produced, a “fake” test data is
provided. This data set has the same format as the real test data ATM_test.csv, except
that the values in there are not the actual values. Don’t worry about the test error
produced with this “fake” test data - this is just to test if your code runs smoothly.
• You should ONLY use the methods covered in the lectures and tutorials in this assign-
ment. You are free to use any Python libraries to implement your models as long as
these libraries are be publicly available on the web.
4. Your group is required to submit meeting minutes (using the same submission link, but
seperated from Group_xxx_document.pdf). You may use the template provided for prepar-
ing meeting minutes. The more detailed the meeting minutes, the better (who does what
2
next, what has been done by whom, etc.). In case of a dispute within a group, we will use
the meeting minutes and/or request for more information to make adjustment to the in-
dividual marks. Should a dispute occurs, please treat each other in a professional and
respectful manner.
3 Marking Criteria
This assignment weighs 25 marks in total. The content in Group_xxx_document.pdf contributes
10 marks, and the Python implementation contributes 15 marks. The marking is structured as
follows.
1. The accuracy of your forecast: Your test error will be compared against the small-
est test error (among all groups including the teaching team). The marker first runs
Group_xxx_implementation.ipynb
• Given that this file runs smoothly and a test error is produced, the 15 marks will be
allocated based on your prediction accuracy, compared to the smallest MSE produced
by the best group, and the appropriateness of your implementation.
• If the marker cannot get Group_xxx_implementation.ipynb run or a test error isn’t
produced, some partial marks (maximum 5) will be allocated based on the appropri-
ateness of Group_xxx_implementation.ipynb.
2. Your report described in Group_xxx_document.pdf: The maximum 10 marks are allocated
based on
• the appropriateness of the chosen forecasting method.
• the details, discussion and explanation of your data analysis procedure.
4 Errors
If you believe there are errors with this assignment please contact the teaching team.
3









































.
2. About your document file Group_xxx_document.pdf
• Describe your data analysis procedure in detail: how the Exploratory Data Analysis
(EDA) step is done, what and why models/methods are used, how the models are
trained, etc. with sufficient justifications. The description should be detailed enough so
that other data scientists, who are supposed to have background in your field, under-
stand and are able to implement the task. All the numerical results are reported up to
four decimal places.
• Clearly and appropriately present any relevant graphs and tables.
• The page limit is 20 pages including EVERYTHING: appendix, computer output,
graphs, tables, etc.
3. The Python file is written using Jupyter Notebook, with the assumption that all the necessary
data files (ATM_training.csv and ATM_test.csv) are in the same folder as the Python file.
If you use deep learning models, then please assume that Keras (with Tensorflow backend)
has been installed.
• If the training of your model involves generating random numbers, the random seed
in Group_xxx_implementation.ipynb must be fixed, e.g. np.random.seed(0), so that
the marker expects to have the same results as you had.
• The Python file Group_xxx_implementation.ipynb must include the following code
import pandas as pd
ATM_test = pd.read_csv('ATM_test.csv')
# YOUR CODE HERE: code that produces the test error test_error
print(test_error)
The idea is that, when the marker runs Group_xxx_implementation.ipynb, with the
test data ATM_test.csv in the same folder as the Python file, he/she expects to see the
same test error as you would if you were provided with the test data. The file should
contain sufficient explanations so that the marker knows how to run your code.
• In case you want to test your code to see if a test error is produced, a “fake” test data is
provided. This data set has the same format as the real test data ATM_test.csv, except
that the values in there are not the actual values. Don’t worry about the test error
produced with this “fake” test data - this is just to test if your code runs smoothly.
• You should ONLY use the methods covered in the lectures and tutorials in this assign-
ment. You are free to use any Python libraries to implement your models as long as
these libraries are be publicly available on the web.
4. Your group is required to submit meeting minutes (using the same submission link, but
seperated from Group_xxx_document.pdf). You may use the template provided for prepar-
ing meeting minutes. The more detailed the meeting minutes, the better (who does what
2
next, what has been done by whom, etc.). In case of a dispute within a group, we will use
the meeting minutes and/or request for more information to make adjustment to the in-
dividual marks. Should a dispute occurs, please treat each other in a professional and
respectful manner.
3 Marking Criteria
This assignment weighs 25 marks in total. The content in Group_xxx_document.pdf contributes
10 marks, and the Python implementation contributes 15 marks. The marking is structured as
follows.
1. The accuracy of your forecast: Your test error will be compared against the small-
est test error (among all groups including the teaching team). The marker first runs
Group_xxx_implementation.ipynb
• Given that this file runs smoothly and a test error is produced, the 15 marks will be
allocated based on your prediction accuracy, compared to the smallest MSE produced
by the best group, and the appropriateness of your implementation.
• If the marker cannot get Group_xxx_implementation.ipynb run or a test error isn’t
produced, some partial marks (maximum 5) will be allocated based on the appropri-
ateness of Group_xxx_implementation.ipynb.
2. Your report described in Group_xxx_document.pdf: The maximum 10 marks are allocated
based on
• the appropriateness of the chosen forecasting method.
• the details, discussion and explanation of your data analysis procedure.
4 Errors
If you believe there are errors with this assignment please contact the teaching team.
3

学霸联盟