fennec-ml 0.4.0

Machine Learning tools for LETU FENNEC senior design team

fennec-ml

A suite of data science and machine learning tools for use by the LeTourneau University FENNEC senior design team.

Instalation

pip install fennec-ml

Quick Start

Data Utils

import os
import fennec_ml as fn

# setup
root_dir = os.getcwd()
excel_dir = os.path.join(root_dir, "Raw_Data")
csv_dir = os.path.join(root_dir, "Proccessed_Data")
timesteps = 60

# excel to csv
fn.folder_cleaner(excel_dir)

# normalize and get labels
norm_data = fn.normalize(csv_dir)
labels = fn.get_CG_labels(csv_dir)

# segment and sort into train, validate, and test datasets
dataset_dict = fn.segment_and_split(norm_data, labels, timesteps)

# use dataset_dict
training_sets = dataset_dict['Training_Set']['sets']
training_labels = dataset_dict['Training_Set']['labels']

Project Structure

myproject/
├── data/
│   ├── raw_data/
│   │   ├── flight123_AA_L.xlxs
│   │   └── ...
│   └── proccessed_data/
│       ├── flight123_AA_L.csv
│       └── ...
├── saved_models/
│   └── ...
├── vars_of_interest.json
├── project_dev.ipynb
└── project_training.py

Features

data_cleaner()

Preprocesses .xlsx files into fennec question-usefull .csv files.

Args:

• filepath (string): The .xlsx file to process.
• savepath (string): The folder to save the .csv file.
• overwrite (bool): Skips the overwrite checker if true.
• skip (bool): Skips duplicate files instead of checking or overwriting if true.
• varspath (string): The vars-of-interest.json path. Defaults to same folder as THIS script.

Relies on the vars_of_interest.json file to determine what data is wanted

fn.data_cleaner(excel_filename, overwrite= True)

---

folder_cleaner()

Preprocesses a folder of .xlsx files into fennec question-usefull .csv files.

Args:

• excel_dir (string): The folder of .xlsx files to process.
• savepath (string): The folder to save the .csv file.
• overwrite (bool): Skips the overwrite checker if true.
• skip (bool): Skips duplicate files instead of checking or overwriting if true.
• varspath (string): The vars-of-interest.json path. Defaults to same folder as this script.

Relies on the vars_of_interest.json file to determine what data is wanted

fn.folder_cleaner(excel_dir, skip= True)

---

normalize()

Return a 3D array of normalized data from cleaned csv's
Note: Normalizing means scaling the data between the min and max values

Args:

• csv_dir (string): The path (including the folder name) of cleaned data
• weights (list): An optional list of weights corresponding to each column
• offsets (list): An optional list of offsets corresponding to column

Returns:

• norm_data (list): A list of numpy arrays holding normalized data

scaled_data = fn.normalize(csv_dir)

---

standardize()

Return a 3D array of standardized data from cleaned csv's
Note: Standarizing means scaling the data so the mean = 0 and the std deviation = 1

Args:

• csv_dir (string): The path (including the folder name) of cleaned data
• weights (list): An optional list of weights corresponding to each column
• offsets (list): An optional list of offsets corresponding to column

Returns:

• stand_data (list): A list of numpy arrays holding STANDARDIZED data

scaled_data = fn.standardize(csv_dir)

---

get_CG_labels()

Reads all filenames in a folder and returns characterization labels

Args:

• csv_dir (string): Directory of .csv files from which to get labels

Returns:

• labels (list): A list of all the characterization labels

cg_labels = fn.get_CG_labels(csv_dir)

---

segment_and_split()

Segments, labels, and sorts data into dataset dictionary

Args:

• input_data (list): List of numpy arrays, 1 per proccessed and scaled files
• input_labels (list): List of characterization lables, 1 per file (should correspond to input_data)
• timesteps (int): length of desired segments
• train_split (float): Percentage of segments to save as training segments
• validate_split (float): Percentage of segments to save as validation segments

Returns:

• output (dict): 3 labels: "Training_Set", "Validation_Set", and "Testing_Set"
  Each set has the follwing labels: "sets" and "labels"
  • "sets" : list of sets, corresponds to "labels"
  • "labels" : list of labels, corresponds to "sets"

data_dict = fn.segment_and_split(all_data, all_labels, 60)

train_set = data_dict['Training_Set']['sets']
train_labels = data_dict['Training_Set']['labels']

Contributers

• Luke Fagg (Team Lead)
• Micah Yarbrough (Pilot ID)
• Wills Kookogey (Fault ID)
• Justin Hawk (3D CG)