Overview

Teaching: min
Exercises: min

Questions

• What capabilities does ChemML provide across data, representation, modeling, optimization, and explainability?

• How can I practice ChemML tools through guided exercises?

• How do I run a full end-to-end workflow from data to AutoML comparison?

• How do I summarize model performance clearly and reproducibly?

Objectives

• Use ChemML notebooks to explore core capabilities

• Complete a capability-focused hands-on activity

• Execute a full modeling workflow with the post pipeline

• Complete a workflow sprint and report results

1. Welcome and Session Plan

This lesson has four blocks in the following order:

• Demo 1 (capabilities): 50 minutes
• Hands-on 1 (capabilities): 50 minutes
• Demo 2 (overall workflow with post): 50 minutes
• Hands-on 2 (workflow sprint): 40 minutes

1.1. What you will do

1. Follow a guided tour of ChemML capabilities.
2. Practice those capabilities in short station-style tasks.
3. Follow a full workflow demo using the post pipeline.
4. Complete a short workflow sprint and report your results.

1.2. Materials location

All lesson resources are in the GitHub repo

Core subfolders:

• /notebooks/ (tutorial notebooks)
• /data/ (dataset)

1.3. Environment setup

If ChemML is already available in your environment, skip to Section 2.

conda create -n chemml_demo python=3.12 -y      # Skip if you already have an env set up
conda activate chemml_demo                      # Replace with existing env if applicable
conda install -c conda-forge openbabel nb_conda_kernels python-graphviz

git clone https://github.com/hachmannlab/chemml.git
cd chemml
pip install -e .
pip install tqdm

2. Demo 1: ChemML Capabilities (50 Min)

2.1. Timeline

• 0-5 min: module map and goals
• 5-15 min: data and preprocessing
• 15-25 min: representation and baseline modeling
• 25-38 min: optimization and AutoML overview
• 38-45 min: explainability and visualization
• 45-50 min: recap and transition

2.2. Follow-along notebooks

Use these notebooks during the live walkthrough (all notebooks are available on the GitHub repo as well as the ChemML website for a reference on how they should look):

2.2.1. Data ingestion and preprocessing

• Missing data handling: missing_values.ipynb
• Data cleaning utilities (optional if short on time): purge.ipynb

Outcome:

• Move from raw tables to clean, ML-ready data.

2.2.2. Chemical representations and features

• Molecule representations/descriptors: Molecule.ipynb
• Feature engineering/representation: feature_representation.ipynb

Outcome:

• Generate chemistry-aware features for modeling.

2.2.3. Baseline modeling and visualization

• Quick baseline model: simple_ml_model.ipynb
• Extended model workflow: ml_model.ipynb
• Plotting/visual summaries: visualize.ipynb

Outcome:

• Build a baseline and inspect model behavior.

2.2.4. Optimization, AutoML, and explainability

• GA concepts: GA_tutorial.ipynb
• GA feature selection: ga_feature_selection.ipynb
• GA hyperparameter optimization: ga_hyper_opt.ipynb
• AutoML regression workflow: autoML_regression.ipynb
• Explainability tools: Explain_visualizations.ipynb

Outcome:

• Understand the full loop: prepare -> represent -> model -> optimize -> explain.

2.2.5. Optional advanced notebooks

• Deep learning example: Convo_nets.ipynb
• Active learning/model-based sampling: active_model_based.ipynb
• Wrapper/PyScript usage: pyscript_usage.ipynb

3. Hands-on 1: Capabilities Practice (50 Min)

3.1. Time plan

• 0-10 min: open notebooks and verify environment
• 10-35 min: complete required capability stations
• 35-50 min: extension station and report-out

3.2. Required stations

Students complete four short stations:

1. Preprocessing station:
   • missing_values.ipynb
2. Representation station:
   • Molecule.ipynb
   • feature_representation.ipynb
3. Baseline model station:
   • simple_ml_model.ipynb
4. Optimization or explainability station (choose one):
   • ga_feature_selection.ipynb
   • Explain_visualizations.ipynb

Required outcomes:

1. Show one cleaned intermediate dataset
2. Generate at least one molecular or engineered feature set
3. Train one baseline model and report one metric
4. Complete one optimization or explainability action

3.3. Optional extension

1. Compare at least two model families from ml_model.ipynb and summarize one key difference.

3.4. What to submit for Hands-on 1

Each student team submits:

• One short table with: station completed, artifact generated, metric or plot
• One figure from any station
• One sentence on the most useful ChemML capability for their own research

4. Demo 2: Full Workflow with post (50 Min)

4.1. Timeline

• 0-5 min: bridge from capabilities to integrated workflow
• 5-15 min: raw data curation and filtering
• 15-25 min: descriptors and feature cleanup
• 25-35 min: baseline training and evaluation
• 35-45 min: transition to AutoML and model comparison
• 45-50 min: recap and transition

4.2. Workflow assets

• Notebook: post_automl_pipeline.ipynb
• Dataset: data/BradleyMeltingPointDataset.csv

5. Hands-on 2: Workflow Sprint (40 Min)

5.1. Time plan

• 0-5 min: open workflow notebook and data
• 5-25 min: execute core workflow steps
• 25-35 min: compare baseline vs AutoML/fallback artifacts
• 35-40 min: rapid report-out

5.2. Required workflow sprint task

Use:

• post_automl_pipeline.ipynb
• data/BradleyMeltingPointDataset.csv

Required outcomes:

1. Run the workflow through baseline model output
2. Execute at least one AutoML-related step or load one artifact file
3. Report one baseline metric and one optimized/compared metric
4. Identify one bottleneck and one improvement

5.3. What to submit for Hands-on 2

• One mini-summary slide or markdown cell with: model choice, metric, bottleneck, next step
• One figure (parity plot, error comparison, or metric table snapshot)

6. Future Direction (Optional)

Potential mini-project directions for interested students:

• Predict excited-state properties from molecular descriptors
• Build surrogate models that connect lower-cost and higher-level excited-state estimates
• Explore time-indexed regression setups for dynamics-inspired datasets

7. Resources

• ChemML docs notebooks source (local copy for this lesson): notebooks
• Workflow notebook: post_automl_pipeline.ipynb
• Workflow dataset: data/BradleyMeltingPointDataset.csv

8. Presentations and Videorecordings

8.1. Presentations

CyberTraining presnetation

Python presnetation

ML process presnetation

ChemML overview

ChemML presentation

8.2. Classroom recording

8.3. Zoom recordings

Key Points

• ChemML capabilities can be taught as a connected stack: data -> representation -> modeling -> optimization -> explainability

• A two-demo format separates breadth (capabilities) from depth (full workflow)

• Precomputed artifacts support comparison when AutoML runtime is limited