Description
In this project you are asked to run experiments on the Wisconsin Breast Cancer dataset. There are 569
examples, each labeled as 0 or 1. Classical approaches achieve accuracy of over 98%. You are asked to train
two classifier for this problem using TensorFlow. The first classifier should be trained using only 8% of the
data (46 examples). The second classifier should be trained using only 16% of the data (91 examples).
1. Your programs must set the random seeds of python and tensorflow to 1 to make sure that your results
are reproducible.
2. Your programs will be tested by training them on randomly selected fractions of the dataset. The testing
data will be the entire dataset.
3. The training and testing of each program should not take more than 5 minutes.
Provided programs and data
1. The dataset is given in the files x test.csv and y test.csv
2. A random subset of 46 training examples in x train8.csv and y train8.csv
3. A random subset of 91 training examples in x train16.csv and y train16.csv
4. An example program proj example.py.
5. A program that can extract a random fraction from the training data is available as fraction xy.py.
What you need to do
Design a network to solve this problem. You can use all the functionality of tensorflow, not only the parts
that were described in class. But your programs cannot read additional material the hard drive.
Grading
We will generate random subsets of training examples by running the program fraction xy.py with a seed
that is kept secret. If, for example, the seed is 7, generating a fraction of 8% can be done as follows:
python3 fraction_xy.py x_test.csv y_test.csv 0.08 7
This creates the files x_test_7_8.csv and y_test_7_8.csv
that should be renamed to x_train.csv and y_train.csv
Your grade will be based on the accuracy of your models trained with the generated examples and tested on
the entire testing data.
What you need to submit
1. Source code of the python script. They should be named as follows:
userid-8.py
userid-16.py
where userid is your user id.
2. Documentation describing your network, and the results of experiments/accuracy that your program
achieves on the provided data.
1
