Create Initial Platform for model training

[hooman007]

Make a python platform that handles the workflow of machine learning.

Based on the Apnea dataset kernels on kaggle, make the first platform for our model. Input data would be from our fitbit output csv. Ticket #3. Based on Ticket #4, (Data Collection - Creating Scenarios) , the input data would focus on several features. Focus the model input on those. Also consider other features taken from fitbit as they might have hidden value.

stages of workflow to include:

• [x] data reading from csv
• [x] data processing and preparing for training
• [ ] framwork to create model architecture
• [ ] training and testing model

Note on data split for train-valid-test: we can have all the data as train first. then if model converges and learns something, we split the data.

[MasoudMo]

Looking at the kernels available on ECG data points, I found some very useful and interesting data pre-processing techniques.

Firstly, the best possible input type we can obtain is .CSV files since Panda has some very useful libraries for loading and manipulating the data.

Secondly, after turning our data into numpy arrays from Panda dataframes, we can augment the data to have equal number of test cases for different outputs. This will be explained in more details in tomorrow's meeting.

Thirdly, the kernel I looked at has a useful neural network architecture that we can study in details in Saturday's meeting.

In general, if the following challenges are overcome, the initial model can be completed: Manage to turn the data from Fit Bit into .CSV files Figure a way to use transfer learning to incorporate data from the Fit Bit into a pre-trained model Determine the types of inputs obtained from Fit Bit Collect a high number of readings from Fit Bit

[hooman007]

fitbit data collection is now active. Please try to use the cSV output of the data collection scripts to start traiing models.

Note:

• either take new data from the short experiments that are going to be done on Jan 12,
• or get the old data of Daniel's fitbit , if he and his Fitbit are there, to extract the data he has labeled in long term and scored an axiety level for them.