Skip to main content

Iteration 2

After the Self-study of Module 2, you apply what you learned to the project. At the end of this iteration you have your own dataset, built from the raw data, and a baseline on it.

In Iteration 1, building the models was quick, because the raw data had already been turned into the clean starter dataset: one row per molecule, with the features and the targets in columns. In this iteration you do that work yourself. You start from the raw HOPV files of 350 molecules and turn them into a dataset in the same form as the starter dataset, with what you learned about data preparation and data understanding in the Module 2 Self-study. Then you train your Iteration 1 models on it as a baseline.

A well-known saying in computing is “garbage in, garbage out”: a model can only be as good as the data it learns from. If the data contains errors, if missing values are handled carelessly, or if the features don’t describe the molecules well, the model learns from those problems, and a better algorithm does not fix that. This is why the data stages get a whole iteration.

The work is still structured: notebooks and resources will guide you through the main steps. However, you will begin making more decisions yourself and applying concepts introduced in the self-study material.

Business Understanding

Read:

Data Understanding

Explore the HOPV data, identify the input and target variables, investigate data quality, and examine important distributions and relationships.

Complete in order:

  1. Understanding HOPV Stability
  2. Understanding HOMO/LUMO Data

Data Preparation

Turn the raw molecular data into data suitable for machine-learning models.

Complete:

Baseline

Train your Iteration 1 classification, regression and clustering models again, in the same way, on your new dataset. These scores are the baseline that you improve in Iteration 3.

Complete:

Keep iterating

Although the notebooks provide a structured path, the machine-learning workflow is not strictly linear.

What you discover in the baseline may give you a reason to return to the data, change its preparation, investigate a new feature, or revisit your understanding of the problem.

In Iteration 3, after the Module 3 Self-study, you improve the models with cross-validation and hyperparameter tuning, take a closer look at the clusters, and score your models on the test molecules for the first time.