The .lpm model file

ML-LPA uses its own file type to store the trained Machine Learning models: the .lpm files (Lipid Phase Model). This tutorial explore thoroughly these files.

The .lpm file format

The .lpm are neither text nor binary files: they are in fact .zip compressed files with a custom extension. Inside the archive can be found several text files defining the Machine Learning models and describing the training scores.

There are multiple reasons to explain the choice of this unusual format:

• Using a non-binary file format prevents compatibility issues (e.g. different pickles version).

• Compressing the several files in a single archive file avoids losing some of the file by mistake and prevents modifications of the text files by mistake.

• Using a different file extension prevents confusion with other archive files.

Inside the .lpm file

Opening the .lpm file

The .lpm files can be easily opened and explored. You will first need to modify the file extension and then open the archive file. This can be done manually in the GUI of your operating system, or within the Terminal with the following commands (in which can renaming in not required).

> mkdir model_contents
> unzip model_test.lpm -d model_contents

In this example, the .lpm file model_test.lpm have been created using the function generateModel(). Check the related tutorial for more details.

The training .csv files

The .lpm archive contains three data files in the .csv format:

• model_coordinates.csv

• model_distances.csv

• model_states.csv

To avoid any compatibility issue with different scikit-learn versions, the instance of the scikit-learn model classes are never directly saved in the files. Instead, the training datasets are stored in the file and are used later to re-train the models on them.

The metadata .xml file

The .lpm archive also contains one data files in the .xml format: the model_data.xml metadata file. In this files are stored several types of information collected during the training:

• The information on the systems that have been used to train the models (e.g. name of the molecule type, number of molecules).

• The settings used for the training (e.g. neighbour rank, size of the training subsets).

• The scores obtained during the training, either the final scores for each phases or the scores for each models used in ML-LPA.

• Other general metadata (e.g. ML-LPA and scikit-learn versions, date and time)

Reading the .lpm file

ML-LPA can directly read the model file, either to extract and/or display the content, or to load it into a model dictionary.

Extract and display

To extract the contents of the .lpm file, the function readModelFile()

metadata_dict, coordinates, distances, phases = mllpa.readModelFile('test_model.lpm', train_sets=True)

The function readModelFile() will return the dictionary of the training metadata (metadata_dict), but also the three arrays used to train the models: coordinates, distances and phases.

You can also directly display the metadata in the Terminal, by using the keyword argument display=

mllpa.readModelFile('test_model.lpm', display=True)

Load from file

The file can be loaded in a dictionary using the function loadModels().

trained_models, training_params = mllpa.loadModels('test_file.lpm')

The function loadModels() will return the trained models in a dictionary (trained_models), but also another dictionary containing all the parameters used to train the models (training_params).

What is next?

• Now that you know how what is inside a .lpm file, you can use it to predict the phases in an unknown system.

Check the API

The following elements have been used in this tutorial:

• readModelFile()

• loadModels()