Naive Level - @Vopaaz

1. Finish extracting naive feature mentioned in #7 (comment) & #7 (comment) & #7 (comment) - More features - 1
2. Apply the traditional machine learning approach and give an initial result.

Deep Learning Approach Preparation - @Xiaochr

1. Search for the line-drawing algorithm and apply the best
2. Explore the https://www.floydhub.com/ thoroughly
3. Learn

DDL

4/21

Work left to do

Docstring - Xiaochr

Feature Extraction

• Markov Chain - Xiaochr
• Search: time-series location data feature extraction - Xiaochr
• Naive feature: max, min, avg distance (l1 distance) to the border of the center area - Vopaaz
• Select the time span where all device has the record and split into arbitrary intervals, on each time point, calculate the position of the device (assume that it's moving at constant velocity and moving along a straight line) - Vopaaz

DDL

4/14 11:00 p.m.

Work left to do

Modify the docstring and squash commit. - Vopaaz

More features

1. The final recorded coordinate. - Vopaaz
2. Convert the path to the image and then apply the neural network. - Suspend
3. Vopaaz and Xiaochr each come up with a feature considering device.

DDL

4/17 24:00, 4/18 Further discussion.

--- DELETE, UNNECESSARY ---

Malkov - Xiaochr

1. Frequency -> Probability
2. Consider time, estimate the position by assuming constant velocity and straight line.
3. Also, write a time-unconsidered version

Classify for Path - Vopaaz

Consider the classification against a path. The (only) feature is the entry point, the prediction is whether the exit is in the center.
Apply generally used classifier.
Actually Malkov is the same idea.

--- DELETE ---

Basic utilities - Vopaaz

• Reading csv file, provide an API that reads the two tables
  • Process the nan value properly
  • Time columns transformed into datetime.datetime data type
• A function that returns whether a point is in the center
• A Framework of submission, where the 16:00:00 - 16:00:00 record can be directly determined, and leave the rest pass, wait until we process.

Exploration - Xiaochr

• Connect all the paths (by inserting records, connecting missed entry and exit points)

Rules

• For each class or function, description or documentation is a MUST.

DDL

1. Vopaaz 4-12 Fri. 24:00
2. Xiaochr 4-13 Sat. 24:00
3. Next Step Discussion 4-14 Sun. 24:00