This is a class project for University of Washington CHEME 545/546 Winter 2021.

The purpose of this project is two-fold:

1. Predict the size (nm), max absorbance (nm) and max photoluminescence emission (nm) outcomes of a hot injection synthesis of CdSe Quantum Dots based on a particular set of experimental conditions.

2. Explore 12 different ML models to identify which algorithm yields the most accurate prediction of the desired 3 aforementioned outputs. Moreover, do a compare and contrast analysis for diameter prediction models explored in this project with those discussed in the paper Machine Learning Tools to Predict Hot Injection Syntheses Outcomes for II-VI and IV-VI Quantum Dots.

Hot_injection
    |- doc
        |- comparison paper.pdf
        |- component_specification.txt
        |- Initial Project Idea.pdf
        |- Technology Review.pdf
        |- Use Cases and Comp Spec Illustration.pdf
    |- hotdots
        |- ML models
            |- Datasets
                |- CdSe - Raw.csv
                |- CdSe for Abs (S&E).csv
                |- CdSe for Diameter (S&E).csv
                |- CdSe for PL (S&E).csv
                |- CdSe scaled.csv
                |- Models_Metrics.csv
                |- augmented_data.csv
                |- documentation.txt
            |- Notebooks
                |- Comparison Santos et al
                    |- Analysis_Summary.ipynb
                    |- Models_Metrics.csv
                |- Dataset Cleaning
                   |- Data Augmentation
                       |- Augmenting Data.ipynb
                       |- Optimization ML Models for Data Augmentation.ipynb
                       |- abs_filler.csv
                       |- augmented absorbance.csv
                       |- augmented_data.csv
                       |- pl_filler.csv
                       |- random_forest_ab.joblib
                       |- random_forest_pl.joblib
                   |- Standardization and One-Hot Encoding.ipynb
                |- ML algorithms
                   |- Single-Output Models
                       |- Decision Tree.ipynb
                       |- Extra Trees.ipynb
                       |- Gradient Boosting Machine.ipynb
                       |- KNN.ipynb
                       |- Multilinear Regression.ipynb
                       |- Random Forest.ipynb
                       |- Ridge Regression.ipynb
                       |- SVR (kernel='linear','rbf).ipynb
                   |- Multi-Output Models.ipynb
        |- Streamlit UI
            |- CdSe - BetterthanRaw.csv
            |- Extra_Trees.joblib
            |- main.py
            |- test_main.py
    |- .gitignore
    |- .travis.yml
    |- LICENSE.txt
    |- README.md
    |- environment.yml

There are 2 primary submodules inside the hotdots package: ML Models and Streamlit UI.

Streamlit UI contains the components for the predictive package, which allows a user to enter certain experimental parameters for the hot injection synthesis of CdSe quantum dots after which package will use the extra trees algorithm to predict the synthesized CdSe dots' diameter, absorbance and emission.

ML Models contains the analysis that our team conducted on the various ML models' predictive capabilities in predicting the diameter, absorbance and emission. In total, 12 models were tested. It also includes our comparative analysis with the paper, from which we obtained our initial dataset.

This interface enables the user input certain experimental conditions after which the package will predict the resultant CdSe diameter (nm), absorbance (nm) and PL emission (nm) wavelength. The package utilizes extra trees model to create

Windows Users: Windows preview, Ubuntu

Mac Users: Terminal

• bokeh
• joblib
• numpy
• pandas
• sklearn
• streamlit

1. Enter into your desired directory from your terminal
2. Enter the command git clone https://github.com/UW-DIRECT-2021-NANO/Hot_injection.git
3. Once inside the local, cloned repository, enter the command cd hotdots/Streamlit-User\ Interface/
4. Enter the command streamlit run main.py
5. Copy the URL into your browser and make some quantum dots!

The user interface code is in a file called main.py inside the Streamlit-User Interface. The test file for this file is called test_main.py.

The Datasets submodule contain the datasets that were used to train the various ML models explored. The original dataset was part of the supplementary information of Machine Learning Tools to Predict Hot Injection Syntheses Outcomes for II-VI and IV-VI Quantum Dots. All of the input data were standardized (numerical inputs) and one-hot encoded (categorical inputs). Data augmentation was used to impute the "None" values of some of the absorbance and emission entries. The resultant dataset after these transformations is labelled as augmented_data.csv in this submodule.

The Notebooks submodule contains all of the analytic components. Comparison Santos et al encapsulates our comparative analysis with the aforementioned paper. The Analysis_Summary.ipynb in that submodule outlines all of the 16 models that were investigated in this project, along with a figure comparison with the paper. The Dataset cleaning submodule includes the dataset standardization, one-hot encoding and augmentation process applied to the initial dataset. The Machine Learning Models submodules include individual notebooks for all of the different single-output and multi-output ML models explored in this project.

Tests can be run using the nosetests package.

nosetests Hot_injection/hotdots/Streamlit UI/ test_main.py

This project follows the PEP8 style using the pylint code checker.

• Benedicte Diakubama - MS, Chemical Engineering
• Florence Dou - PhD, Chemistry
• Hao Anh Nguyen - PhD, Chemistry
• Harrison Sarsito - MS, Chemical Engineering