MetabolicRate Predictor

Predictor Algorithm baseed on multi platform machine learning methods to predict tabular data

Amir Kfir • Tomer Keren

Project includes:

1. DataSet preprocess
2. Classic Machine Learning algorithms results
3. Deep Learning Model results

	Data Set Example
	Results Comparison

Training and comparing

The api made simple as possible. Also added Optuna wrapper - Use the relevant switches in order to ask Optuna to find the best parameters for your model

Files in the repository

File name	Purpsoe
dataset_prepare.py	creates DS from CSV file and applying preprocessing
dl_model.py	final dl used on the DS
linear_ml_nodel.py	linear models that were tested before final model was chosen
optuna_search.py	automatic search for best model hyperparameters
rmr_predictor.py	supporting file for running algorithm

Running the algorithm

1. Download repo either by pressing "Download ZIP" under the green button or use clone command

git clone https://github.com/tomerkeren42/MetabolicRate.git

2. Install requriements:

pip install -r requirements.txt

Prerequisites

Library	Version	Library	Version
Python	3.8	torch	1.9.0
pandas	1.1.3	scikit-learn	0.32.2
plotly	5.0.0	scipy	1.5.2
matplotlib	3.3.2	xgboost	1.4.2

3. Add Tabular DataSet to dedicated path:

MetablicRate/dataset/

4. Run python main.py -h and follow instructions for your own use

usage: RMR Predictor -p PATH [-h] [-o] [--study-name STUDY_NAME] [--trials TRIALS]  [--log LOG] [--epochs EPOCHS] [--learning-rate LEARNING_RATE] [--hidden-units HIDDEN_UNITS] [--optimizer-name {Adam,RMSprop,SGD}] [--dropout DROPOUT] [--weights_file WEIGHTS_FILE]

Starting RMR predictor tool - trainable deep learning net, which is compared to other ML algorithms

positional arguments:
  -p PATH, --path   Enter path to dataset
optional arguments:
  -h, --help            show this help message and exit
  -o, --optuna          Run Optuna optimization for detecting best DL model parameters
  --study-name          Run Optuna optimization for detecting best DL model parameters
  --trials              Number of epoch for Deep Learning Model
  --log                 Write output to new log file at logs/ directory
  --epochs              Number of epoch for Deep Learning Model
  --learning-rate       Step size for the optimizer which trains the DL model
  --hidden-units        Number of hidden units in the hidden layer of the DL model
  --optimizer-name {Adam,RMSprop,SGD}
                        Optimizer for training the DL model
  --dropout             Probability of dropout layer for turning off neurons in the DL model
  --weights_file        Path to weight file, if exist and do not want to train new net

5.Enjoy!

For more additional help:

Optuna