Skip to content

Latest commit

 

History

History
122 lines (92 loc) · 5.66 KB

README.md

File metadata and controls

122 lines (92 loc) · 5.66 KB
Raw

HH → bbWW

Analysis based on columnflow, law and order.

Quickstart

A couple test tasks are listed below. They might require a valid voms proxy for accessing input data.

# clone the project
git clone --recursive [email protected]:uhh-cms/hh2bbww.git
cd hh2bbww

# source the setup and store decisions in .setups/dev.sh (arbitrary name)
source setup.sh dev

# index existing tasks once to enable auto-completion for "law run"
law index --verbose

# run your first task
# (they are all shipped with columnflow and thus have the "cf." prefix)
law run cf.ReduceEvents \
    --version v1 \
    --dataset st_tchannel_t_powheg \
    --branch 0

# create some plots
law run cf.PlotVariables1D \
    --version v1 \
    --datasets st_tchannel_t_powheg \
    --producers features \
    --variables jet1_pt,jet2_pt \
    --categories 1e

# create a (test) datacard (CMS-style)
law run cf.CreateDatacards \
    --version v1 \
    --inference-model default \
    --workers 3

Most important files

(Note: please tell me, if a link or task does not work. I did not test all of them)

Config

Files relevant for the configuration of an analysis are mainly to be found in this folder. The main analysis object is defined here. The analysis object contains multiple configs (at least one config per campaign). Most of the configuration takes place here and defines meta-data like the used datasets, processes, shifts, categories, variables and much more. At the moment, there are only two configs, the default config config_2017 and a config with reduced event statistics (for test purposes) named config_2017_limited. Most tasks can use a --config parameter as an input, e.g.

law run cf.SelectEvents --version v1 --config config_2017

Selectors

Modules that are used to define event selections are usually named selectors and can be found in this folder. The main selector is called default and can be found here. You can call the SelectEvents task using this selector, e.g. via

law run cf.SelectEvents --version v1 --selector default

Producers

Modules that are used to produce new columns are usually named producers and can be found in this folder. A producer can be used as a part of another calibrator/selector/producer. You can also call a producer on it's own as part of the ProduceColumns task, e.g. via

law run cf.ProduceColumns --version v1 --producer example

Machine learning

Modules for machine learning are located in this folder. Our base ML Model ist defined here and parameters are defined as class parameters. From the base class, new ML Models can be built via class inheritance. Our default ML Model is the DenseClassifier, which uses the base model and some mixins for additional functionality and overwrites their class parameters. From each ML Model, new models can be derived with different parameters using the derive method:

DenseClassifier.derive("dense_default", cls_dict={"folds": 5})

All ML Models can be used as part of law tasks, e.g. with

law run cf.MLTraining --version v1 --ml-model dense_default --branches 0

NOTE: the DenseClasifier already defines, which config, selector and producers are required, so you don't need to add them on the command line.

Inference

Modules to prepare datacards are located in this folder. At the moment, there is only the 'default' inference model (here). Datacard are produced via calling

law run cf.CreateDatacards --version v1 --inference-model default

Similar to the ML Models, we can also derive additional inference models using the derive method.

NOTE: our inference model already defines, which ml_model is required and based on that, producer requirements are resolved automatically per default. It is therefore better to not add them to the task parameters since the dependencies are already a bit complicated.

Tasks

Analysis-specific tasks are defined in this folder.

Columnflow

The analysis uses many functionalities of columnflow. We rely on:

Law config

The law config is located here and takes care of information available to law when calling tasks. In here, we can for example:

  • set some defaults for parameters (e.g. when not setting the --dataset parameter in a task that needs this parameter, we use the default_dataset parameter instead)
  • define, in which file system to store outputs for each task
  • define, which tasks should be loaded for the analysis

Development