Fourth_Day

Authors:

1. Stephan Meighen-Berger, developed the Fourth Day Code
2. Li Ruohan, implemented the detector simulation
3. Golo Wimmer, developed the Navier-Stokes code

Table of contents

1. Introduction

2. Citation

3. Documentation

4. Installation

5. Emission PDFs

6. Code Example

7. Calibration mode

8. Code structure

9. BETA

   1. Probabilistic Modeling

   2. VEGAN

Introduction

A python package to simulate the bioluminescence in the deep sea. It calculates the light emissions and progates it to a detector. The detector response and properties can be (rudementarily) modelled using this code as well.

Citation

Please cite our work arXiv:2103.03816.

Documentation

The package provides automatically generated documentation under Documentation.

Installation

The easiest way to install the package is to use pip install:

pip install fourth_day

Then simply run

from fourth_day import Fourth_Day, config
# Initialize the object
fd = Fourth_Day()
# Fetch water current data. This may take a while
fd.load_data()

to get the necessary data sets (this requires an internet connection). Please note this requires pyDataverse, which you may not find using Anaconda (we suggest pip instead if it hasn't already been installed).

Another method is: To install please clone the repository or download the latest release. Then follow the instructions given in INSTALL.txt. Note this should install all necessary components except for the beta developments and the Navier_Stokes_code. Additionally, basic water current simulations can be downloaded under [https://doi.org/10.7910/DVN/CNMW2S]. The location of these files needs to be specified by setting

config['water']['model']['directory'] = "../PATH/TO/FOLDER/"

example_dataverse_downloader.ipynb shows an example how to download the dataset using the pyDataverse package.

Emission PDFs

The emission pdfs are constructed from data taken from Latz, M.I., Frank, T.M. & Case, J.F. "Spectral composition of bioluminescence of epipelagic organisms from the Sargasso Sea." Marine Biology 98, 441-446 (1988) https://doi.org/10.1007/BF00391120.

Code Example

A basic running example to interface with the package

# Importing the package
from fourth_day import Fourth_Day, config
# Creating fourth day object
fd = Fourth_Day()
# Running the simulation
fd.sim()
# The time array
t = fd.time
# The produced light
data = np.array([np.sum(fd.statistics[i].loc[:, 'photons'].values)
                 for i in range(len(fd.t))])
# Measured light
measured_detector = np.array([fd.measured["Detector 1"].values])

The last line produces results of the form

Depending on the detector specifications. In general, organism properties and emissions are stored in fd.statistics, while the expected measured time-series by the detectors is stored in fd.measured. For a more in-depth example, use the python notebook example_basics.ipynb in the examples folder. There you can find additional examples covering most use cases for the package.

Calibration mode

Besides the typical bioluminescence simulation, the code also offers a calibration mode. In this mode, standardized flashers (as defined by the user) are modeled and placed. The resulting measurements (time series) can then be extracted, allowing for quick and dirty calibration measurements in water. By defining possible errors in the different aspects of the measurement realistic data sets for calibration analysis can be generated. An example of such a simulation run is shown here

Code structure

The code is structed as

BETA

All projects listed here are currently in devolpment. We provide in the hopes they may help future development or advanced users. The installation requirements are not designed to accomodate these new modules and the user needs to install them themselves.

Probabilistic Modeling

Here examples are given how to construct emission pdfs (depending on location). These can in turn be used to construct simplified models for bioluminescence and when analyzing data.

VEGAN

A rudimentary GAN network, testing the waters if data generation can be replaced by using neural networks. One thing that needs improvement is the measure. We suggest introducing a Wasserstein Loss function. Here we give an example of the output of the NN after a few generations (black), compared to an example set from the MC sim (red).