Machine Learning Features

This code is intended to make using CRIPT for gathering polymer data for machine learning simple and easy.

Note: As of right now, only supervised learning support is being developed. Unsupervised learning support may be added in the future.

How To Use: Returning Data Object

To return a data object, the user will call the ML_Data_Process function. This function has 6 parameters:

1. FeatureProperties

These are the properties the user wants to use as the features of their data. To continue the example from before, they may only want materials whose color property is red but want the properties that serve as the features for their model to be something such as molar mass, stress, solubility, etc. They should specify those properties here.

2. LabelProperties

These are the properties that serve as the labels for the model, similar to the features.

3. Library

This is where the user specifies which of the supported machine learning libraries they are using. As of now, support is being developed for Scikit-learn, Tensorflow, and Pytorch. This is specified using an enum named ML_Library. For example, if the user is using pytorch, they would pass in the value ML_Library.PYTORCH

4. SearchParameter

This should be a substring present in the names of all the materials the user wants to go through. For example, if you were only interested in materials that contained "carb" in their name, you would set SearchParamter to "carb". The default value for this parameter is an empty string ''.

5. Preprocess

This parameter is where the user can specify if they want the function to preprocess the data for them. The default value for this parameter is False, meaning no preprocessing will be done if not specified and it will only return a data object compatible with the user's machine learning library (if supported). If the user wants the function to preprocess the data for them, they can set Preprocess to a list of Boolean variables representing what types of preprocessing they want to do. As of right now, this list is of the form [Normalization, Outlier Removal, One Hot Encoding]. In other words, if the user is dealing with numeric data and wants it to be normalized and the outliers to be removed, they would set it to [True, True, False]. Or if they were dealing with categorical data and needed it to be one hot encoded, they would set it to [False, False, True].

6. Limit

This parameter controls how many materials the search function will go through, not how many final materials there are. After the search function gathers that many materials, many have to be removed due to lacking the specified properties. By default this is set to 10000.

How to Use: Exporting

There is another function in this code called ML_Data_Save. Instead of returning a data object compatible with a machine learning library, this saves the data into a csv file on the user's machine. This is useful if they wanted to instead analyze the data in R.

This function has 3 parameters:

1. Properties

These are the properties that the user wants to save as part of their data, similar to FeatureProperties and LabelProperties

3. Limit

This is the same as in ML_Data_Process

5. SearchParameter

This is the same as in ML_Data_Process