1. dtk-tools
Benoit Raybaud, Research Software Manager

2. dtk-tools Experiments and analysis
IDM Weather and Demographics service
Models
On premise cluster Cloud environments
dtk-tools
To carry out experiments and analysis we rely on models.
Those models need to be parametrized and files usually coming from the local machine or some sort of services
They also need to run somewhere: Either the local machine or some high performance computing services
Dtk-tools is the suite of tools allowing to put it all together
Local files
Local computer

3. Overview
Provide a shared way for the different teams to create and execute workflows involving experiments, analysis, and calibrations
Easy to reproduce, share and reuse those workflows
Abstracts a lot of technical minutia
Lower the technical expertise requirement to leverage available resources
Accelerate the creation of large experiments
Provide a central repository for utilities
Before this framework was created each teams had their own way of running workflows and using the IDM ecosystem.
By providing a common way to leverage the available resources we enabled users to easily share reuse and reproduce the workflows and to abstract a lot of technical minutia.
This led to lowering the technical expertise required and allowed our users to focus more on the science and less on the software.
Now that dtk-tools is widely adopted it also provides a centralized repository for all sorts of utilities.

4. Technical details Written in Python 3.6
Free, open source, and easily extensible
Supports EMOD and CMS
Disease specific packages
Malaria
HIV TB
Typhoid
Disease specific packages
: The different diseases have custom reports for each of them and analyzers to consume them

5. Main features – Input files
EMOD needs input files to represent experiments’ environment
Heterogeneous sources
Local files
IDM Weather and Demographics Input service
National services (NOAA, WorldPop, etc.)
Not all models need weather

6. IDM Weather and Demographics Service
Input files workflow
IDM Weather and Demographics Service
Climate for 2014 – 2016
Resolution: 2.5’
Custom climate
Local files
Custom temperature for a given location
A usual workflow would be to query the IDM service for the climate in a given place, for a given timestamp.
You can query the service directly within the tools and retrieve the set of files.
Then Let’s imagine I want to override the temperature for a given location with a CSV file that I have created locally.
Well, the tools let you easily use both the base weather and the custom temperature file to create a custom weather.
We also provide utilities to transform this custom weather into model readable files. You can then use those files with your simulations.
It is interesting to note that we also provide utilities to go the other way and make model files readable.
Model consumable set of files

7. Main features – Model parameters
Models need parameters and scenarios
EMOD Model needs parameters and interventions definition files
The tools provide
Starting points with default files for different types of simulations
Convenient ways of editing model parameters
Shortcut functions allowing to accelerate the creation of interventions

8. Main features – Asset collections
When running simulations, collects all the necessary files
Packaging handled transparently by the tools
Easy to share, reproduce, and reuse
Namawala input files
EMOD 2.13 executable
New collection
Local custom files
Creation of a new asset collection from different collections

9. Main features – Sweeps
Convenient and easy way of exploring parameters space
Accelerate the creation and tagging of simulations
Coverage
Multiplier
20%
0.5 1 2 …
75% 5
Base configuration files
Simulation
Simulation
Simulation
Sweep Definition
Coverage: 20% to 75% (5% step)
Rainfall multiplier: 0.5, 1, 2, 5
12*4 = 48
Simulation
Simulation
Creation of a sweep

10. Main features – Templating
Templating system to modify specific blocks within the model inputs
Build complicated scenarios easily
Base campaign
Scenario definition
10% A
0.5
20% 1
30% 2 B 5
Simulation
Intervention 1
Simulation
Intervention 2
Simulation
Simulation
Intervention 3
Simulation
Simulation
Creation of an experiment based on templates

11. Main features – Analyzers
Simulations can provide a lot of results but usually only a subset of those results are relevant
Analyzers are extracting needed per simulation data and then combining results to support investigations
Simulations outputs
Analysis
Simulations can provide a lot of results but usually only a subset of those results are relevant.
Analyzers are extracting the relevant data from each simulations and then combine those together to support your investigations.
This analyzer process works as follow:
In one hand, we have a set of simulations outputs that can come from multiple experiments
In the other hand we have our analysis that can include one or more analyzers (here 3)
Each analyzers will automatically select the data it needs and produce the output.
Analyzers can span a wide range of outputs, from charts to CSV files or heatmaps but they are flexible enough to basically be able to generate anything that Python can handle.
Also those analyzers are not tied to a specific experiment or users and can be shared.
Another major advantage is that you are not required to know where those outputs files are located, the tools are taking care of all the technical details for you to let you focus on how you want to process the data.
Analysis process

12. Main features – Calibration
Enables exploration of a multidimensional parameter space
Fit the model outputs based on reference data
Several algorithms available to optimize search in the parameter space
Set model parameters that cannot be measured
Reasonable parameter set when data is sparse

13. Calibration – Overview
Study Site
Base simulation
Reference data
Model
Run the scenarios
Produce outputs
Comparison
Compare with reference
Assign a likelihood
New parameters
Likelihoods
Search state
Monitoring
Next point algorithm
Define next best set of parameters to try

14. How to get the tools? GitHub repository:
Private repository
Contact us:

15. Thank you !