Phi toolbox for multivariate analysis by Sal Garcia ([email protected], [email protected])

Version 2.0 includes: Batch Analysis

Version 1.0 includes: Principal Components Analysis (PCA), Projection to Latent Structures (PLS), Locally Weighted PLS (LWPLS), Savitzy-Golay derivative and Standard Normal Variate pre-processing for spectra.

A variety of plotting tools for models created with pyphi.

Pyphi requires the following python packages: numpy, scipy, pandas, xlrd, bokeh, matplotlib, pyomo. These can be installed via setup.py below or manually using pip/conda and the requirements.txt file.

1. Ensure you have Python 3 installed and accessible via your terminal ("python" command).
   • It's strongly encouraged you create a virtual environment using anaconda (conda create -n your_pyphienv python) or venv (pip -m venv your_pyphienv). You can then activate your environment conda activate your_pyphienv or venv Windows yourenv\Scripts\activate.bat or venv Linux/mac source yourenv/bin/activate ) and then install everything into a sandboxed environment.
2. Download this repository via git clone or manually using the download zip button at the top of the page.
3. Install the pyphi and pyphi_plots modules by opening a terminal window, navigating to the root of this repository, and typing pip install -r requirements.txt.

To confirm you have a working installation, navigate to the Examples folder and copy the Example_Script_testing_MD_by_NLP.py to the directory of your choice. Run python Example_Script_testing_MD_by_NLP.py, verifying there are no errors logged to the console.

• IPOPT as an executable in your system path or GAMS python module or GAMS executable in yoru system path.
  • Windows: conda install -c conda-forge IPOPT=3.11.1 or download from IPOPT releases page, extract and add the IPOPT\bin folder to your system path or add all files to your working directory.
  • Mac/Linux: conda install -c conda-forge IPOPT, download from IPOPT releases page, or Compile using coinbrew.
• libhsl with ma57 within library loading path or in the same directory as IPOPT executable.
  • Speeds up IPOPT for large problems but requires a free academic or paid industrial license and a local IPOPT installation.
  • Must request in advance and building the source code is nontrivial. Expert use only.
• If IPOPT is not detected, pyphi will submit the pyomo models to the NEOS server to solve them remotely.
  • To use the NEOS server, the environment variable "NEOS_EMAIL" must be assigned a valid email. This can be done outside of python using set/set/export or use import os os.environ["NEOS_EMAIL"] = [email protected] in your code.

Adding a folder to your system path:

• Windows: temporary set PATH=C:\Path\To\ipopt\bin;%PATH% or persistent setx PATH=C:\Path\To\ipopt\bin;%PATH%.
• Mac/Linux: export PATH=/path/to/ipopt:$PATH, add to .profile/.*rc file to make persistent.
• Both via Conda: after activating your environment, use conda env config vars set and your OS-specific set or export command.

• Added batch predict and data replicator to phi
• I realize I had not updated this section in a long while (sorry)

• Added routine cat_2_matrix to conver categorical classifiers to matrices
• Added Multi-block PLS model

• Fixed small bug un clean_low_variances routine

• Added rotation of loadings so that var(t) for ti>=0 is always larger than var(t) for ti<0

• Enhanced clean_low_variances function to return a list with columns removed from dataframe.

• PLS model estimation using Non-linear programming as described in Journal of Chemometrics, 28(7), pp.575-584.

• PCA model estimation using Non-linear programming as described in Lopez-Negrete et al. J. Chemometrics 2010; 24: 301–311.