Outline
- ● Objectives
- ● Task Description
- ● Data
- ● Evaluation Metric
- ● Kaggle
- ● Grading
- ● Code Submission
- ● Deadlines
- ● Hints
- ● Regulations again
- ● Useful Links
Objectives
- ● Solve a regression problem with deep neural networks (DNN).
- ● Understand basic DNN training tips
e.g. hyper-parameter tuning, feature selection, regularization, … - ● Get familiar with PyTorch.
Task Description
- ● COVID-19 Cases Prediction
- ● Source: Delphi group @ CMU
○ A daily survey since April 2020 via facebook.
Do not attempt to find any related data! Using additional data is prohibited and your final grade x 0.9 !
Task Description
● Given survey results in the past 3 days in a specific state in U.S., then predict the percentage of new tested positive cases in the 3rd day.
survey positive cases
Day 1
survey
positive cases
survey
positive cases
Day 3
Day 2
Data — Delphi’s COVID-19 Surveys
Conducted surveys via facebook (every day & every state)
Survey: symptoms, COVID-19 testing, social distancing, mental health, demographics, economic effects, …
Data — Delphi’s COVID-19 Surveys
All population in a some samples survey certain state of the U.S.
estimation for all population in that state
(data we are using)
Data — Delphi’s COVID-19 Surveys
● States (40, encoded to one-hot vectors) ○ e.g. AL, AK, AZ, …
● COVID-like illness (4)
○ e.g. cli,ili (influenza-like illness), …
● Behavior Indicators (8)
○ e.g. wearing_mask, travel_outside_state, …
● Mental Health Indicators (5) ○ e.g. anxious, depressed, …
● Tested Positive Cases (1)
○ tested_positive (this is what we want to predict)
Percentage
Data — One-hot Vector
● One-hot vectors:
Vectors with only one element equals to one while others are zero. Usually used to encode discrete values.
If state code = AZ (Arizona)
one-hot encoding
AL (Alabama) AK (Alaska) AZ (Arizona) AR (Arkansas)
WI (Wisconsin)
Data — Training
covid.train.csv (2700 samples)
state one-hot Day 1 features Day 2 features
encoding (40) (18) (18)
Day 3 features (18)
1 row = 1 sample
tested positive
Data–Testing
covid.test.csv (893 samples)
state one-hot Day 1 features Day 2 features Day 3 features
encoding (40) (18) (18) (17)
1 row = 1 sample
Evaluation Metric
● Root Mean Squared Error (RMSE)
your model
input features (testing data)
ground truth label (correct answer)
Kaggle
- ● Link: https://www.kaggle.com/c/ml2021spring-hw1
- ● Displayed name: <student ID>_<anything>
- ○ e.g. b06901020_puipui
- ○ For auditing, don’t put student ID in your displayed name.
- ● Submission format: .csv file
○ See sample code
Kaggle — Submission
- ● You may submit up to 5 results each day (UTC).
- ● Up to 2 submissions will be considered for the private leaderboard.
remember to select 2 results for your final scores before the competition ends!
Grading
- ● Simple baseline (public)
- ● Simple baseline (private)
- ● Medium baseline (public)
- ● Medium baseline (private)
- ● Strong baseline (public)
- ● Strong baseline (private)
- ● Upload code to NTU COOL
+1 pt (sample code) +1 pt (sample code) +1 pt
+1 pt +1 pt +1 pt
+4 pts
Total: 10 pts
Grading — Kaggle
● We might change the strong baseline if it’s too hard.
Grading — Bonus
- ● If you got 10 points, we make your code public to the whole class.
- ● In this case, if you also submit a PDF report briefly describing your methods (<100 words in English), you get a bonus of 0.5 pt.
(your report will also be available to all students) - ● Report template
