MLRF

Machine Learning and Shape Recognition

Project Organization

├── LICENSE
├── README.md          <- The top-level README for developers using this project.
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks          <- Jupyter notebooks.
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   ├── figures        <- Generated graphics and figures to be used in reporting
|   └── results        <- Generated models results
|
└── src                <- Source code for use in this project.
    │
    ├── data           <- Scripts to download or generate data
    │
    ├── features       <- Scripts to turn raw data into features for modeling
    │
    ├── models         <- Scripts to train models and then use trained models to make predictions
    |
    └── visualization  <- Scripts to create exploratory and results oriented visualizations

---

Models pre-trained are available here : https://drive.google.com/file/d/129uGZ0P8uAIr4rGbLuZmc40LpUzzWMB1/view?usp=sharing

Make sure to put them in ./models folder.

All commands must be done from the root path mlrf/>

Python setup

First, you will have to install python and setup a virtual environment.

python -m venv venv

Then you will be able to activate it and install packages :

venv/Scripts/activate
pip install -r requirements.txt

Compute all

python ./src/process_all.py

In depth commands

Data part

First let's download the data :

python ./src/data/get.py

You can then retreat it to make a temporary dataset :

• train_batches=5: Train batches amount to take

python ./src/data/make.py --train_batches=5

Let's have a look of what data looks like images in the train set.

• label=0 : Images with this label will be displayed
• amount=9 : Size of tile (9 means 81 images)

python ./src/visualization/mosaic.py --label=5 --amount=500

All figures are saved in ./reports/figures

You can also visualize labels repartition :

• dataset=train : Display train or test dataset.

python ./src/visualization/repartition.py

Features part

Now you can process features. You can compute Flatten, HOG and LBP :

• features=all : Choose features to build ex : flatten,hog

python ./src/features/build.py

To check features correlation :

python ./src/visualization/corr.py

Models training

Build and train your models :

• hyper_params={"svm": {"tol": 1e-4, "C": 1.0, "max_iter": 50}, "k-means": {"n_neighbors": 10}, "xg-boost": {"max_depth": 15, "epochs": 50, "learning_rate": 0.1}} : Dictionary of hyper-parameters
• override=False : Override model, else doesn't compute it

python ./src/models/train.py

leaf_size for KMeans will be adjusted with DE algorithm.

Notice it will display loss history for XG-Boost model

Models testing

You can now tests your models to get performance metrics. Each test will be save with date as key in ./reports/results

• delete_hist=False: Remove results history

python ./src/models/test.py

If you have downloaded pre-trained models and set override to False, it will not compute them.

After models testing, you have a look on figures :

python ./src/visualization/performances.py

Additional commands

• Clean interim data

python ./src/data/clean.py