This directory contains data, processing, and visualizations I found useful during my graduate research in chemical engineering. It is not a package but a collection of scripts. However, some scripts are used as modules to other ones, in which case they have been defined to allow one to import from anywhere.
There is no standard for processing data with dimensions and units. Here, I either specify the units in the accompanying text, or directly in the column name. One can use any of the many existing data structure for arrays, or lists of arrays, that specifies dimensionality and units in addition to numbers. But I have simply used pandas dataframes with float arrays because the purpose here isn't to perform operations between quantities which requires dimensional consistency, but only to look at distributions and correlations, and the reported units are often most convenient.
The minimum quality level for inclusion is that a script produces some data visualization or processing on the input data. This is a low bar because superior plots are available in the literature, sometimes even on Wikipedia, but there is some advantage in having the data in machine readable form to do calculations on. Some scripts merely visualize a geometric sequence as a plot. Others calculate statistics, evaluate integrals, or do some other computations on the input data.
For the most part, these scripts demonstrate the application of basic statistical reasoning in either purely empirical or mixed theoretical and empirical models for the purpose of prediction or design.
I have tried to cite all data sources but in some cases have lost the original reference materials (such cases tend to be small amounts of data readily available on the internet by search). There is no license and it cannot be licensed because it makes use of copyrighted data. You must comply with any of the original copyrights should you try to extend these scripts in your own applications.
This package extensively uses numpy, scipy, pandas, and matplotlib which are not standard library but considered core numerical and scientific computing. It also uses thermo, ase, and pubchempy, though those packages are usually separated into different scripts from the main one.
These libraries are already high-level, to an extent that can be questioned as good practice, such as plot methods for the DataFrame class. Much of the code in these scripts is redundant because it achieves the same goal, e.g., determining if an array makes a geometric sequence. But building a utility API that do little or nothing more than take inputs and place them into inputs of another high-level API function appears unnecessary. The only advantage it gives is that a change to a common source of those utilities would uniformly change all uses of it, but given the scripts are for the most part standalone, this advantage is not great. This is the justification for redundant code.
The analysis is ad hoc. What might be done is a systematic and exhaustive analysis, by consideration of combinatorics. That is, look into the distribution of every variable (or its logarithm), and the correlation between every pair of variables (or the logarithms thereof), with linear regression done between interesting pairs. In some cases it is done, for example, by using the pandas.DataFrame.corr method. But each variable has a different physical significance, many pairs are not interesting even if they have strong correlations, some pairs are interesting even if they have noisy correlations, some relationships are neither linear nor power law, and so forth. An ad hoc approach is useful when one comes in knowing the relationships, usually derived from physical models, expected between the variables.
Some packages which I found useful for research problems, though only a small number are used here:
For atomistic data including structure:
- ase.data and ase.neighborlist modules of the ase package, among the very many other useful modules of this package.
- molmod (the data modules) from the Center for Molecular Modeling.
- pymatgen, for querying the database of the Materials Project.
- mendeleev by Lukasz Mentel.
- pubchempy package from Matt Swain, python interface to the PubChem API. For unit conversions and calculations with physical quantities:
- pymatgen.core.units module of the pymatgen package.
- python-quantities by Darren Dale et al.
- brian2 by Marcel Stimberg et al. For thermodynamic data:
- thermo package from Caleb Bell, has extensive property data and convenient methods. Dependencies on some of the author packages in the author's Chemical Engineering Design Library, including fluids (modeling fluid dynamics) and chemicals (which has most of the thermodynamic data). External C++ library dependency (through python binding) on CoolProp.
- pyro by Prof. Christopher Martin. For spectroscopic data processing:
- Radis by Erwan Pannier et al.
- rampy by Charles Le Losq et al.
- chemplexity by James Dillon. General chemistry modeling
- ChemPy package from Bjoern Dahlgren, has general solution to the problem of equilibria and kinetics, and some thermodynamic data for water. The author has several other packages for numerical computing and chemistry applications.
- ChemPy package from J W Allen, has molecular structure property calculations, rate constant fitting, rate equation evaluation, and thermodynamic data evaluation, all for molecular scale analysis.
- Reaktoro package by Allan Leal (C++ library with Python binding). A formulation of equilibrium as a minimization of Gibb's free energy, rather than as non-linear system of equations which have fundamental thermodynamic variables implicit in rate constants (which is what ChemPy by Dahlgren formulates reaction equilibria as, solving the nonlinear symbolic equations using its own pyneqsys). Big, multipurpose libraries:
- RDKit, for biological applications but there are many useful utilities for the more general computational chemist, including structure parsing and atom type/functional group classification in its GraphMol sublibrary (C++ library).
- Cantera, actively community maintained with financial support by NUMFocus, has data and processing like in thermo and ChemPy but fuller featured (C++ library).
Some notes:
- I found the thermo package stating that the vapor pressure of silver is 1.e-5 Pa at 1217 K, when it is ~1.e-2 torr at that temperature. Care should be taken in interpreting these results. It appears the thermo package is more for vapor-liquid and other conventional chemical engineering and chemistry applications.
- There is also the ChemPy package by Jan Rybizki, which is for astrophysics.
- Throughout these scripts there is the common, though incorrect, definition of axis labels in logarithmic plots as merely that variable such as 'x', letting the logarithmic scale on the axis imply that the logarithm is taken. Of course, what is plotted is the logarithm of the quantity, so the label is more correctly 'log(x)' .
There are several modules which are erroneous, or sufficiently incomplete as to not be included in here. They may at some time be added. There is no guarantee of correctness for what is at this time published. Please report errors in the issues. Note some errors are already known and stated in the readmes or commit log.