Code for the Optimisation of ID's using Python and Opt-AI

Opt-ID is run is by providing:

• a main configuration file in YAML format which contains all the various parameters for the sort/shim job
• an existing directory in which output data will be written to

There are two main flags, --sort and --shim, to run sort and shim jobs. The idea is that using either of these flags in conjunction with the YAML config file will go through and run all the scripts that are used to produce intermediate files and pass them around appropriately, so then there's only one command needed to be executed to run a sort or shim job, and the YAML config file is the single source of all the parameter information used for that particular job.

There are several other processes that Opt-ID provides that are desired to be done after a sort/shim but don't require the sequence of scripts that a sort/shim job does (for example, the use of compare.py to compare a shimmed genome to the original genome), so the --sort and --shim flags aren't able to provide these sorts of processes. To do so, there are several shell scripts that are autogenerated when a sort or shim job is run that can be executed. These scripts run Opt-ID in the particular way that is needed to perform the process, without the user needing to worry about extra configuration on top of the YAML file.

Taking the compare.py example previously mentioned, a script would be autogenerated after a shim job called compare_shim.sh that can be passed any shimmed genome file in the data directory, and it will take care of calling Opt-ID in the particular way it needs to in order to run the compare.py script with the appropriate parameters. More details on how to use these autogenerated shell scripts are below in the "Using the autogenerated shell scripts" section.

The data outputted by Opt-ID is split roughly into two categories:

• large files such as .h5 files
• smaller files such as .json, .mag, .sh files

The smaller files get written to the directory passed as the second parameter to Opt-ID, so if OptID was passed /home/FedID/my_dir then the smaller files would get written to /home/FedID/my_dir.

The larger files get written to a directory within /dls/tmp/FedID whose path is based on the user's FedID and also the name of the data directory passed to Opt-ID. The name of the directory created in /dls/tmp/FedID will be the name of the very last directory in the path passed to Opt-ID. For example, if the path /home/FedID/my_dir is passed to Opt-ID, then the directory /dls/tmp/FedID/my_dir will be created. Symlinks are then created in /home/FedID/my_dir to point to the larger files inside /dls/tmp/FedID/my_dir.

One reason behind having two separate directories containing different data files is due to the large size of the .h5 files produced by Opt-ID and not having the space to put them just anywhere in the filesystem (/dls/tmp has much more available space than, for example, the home directory associated to a FedID). Another reason is that the automatic deletion of files in /dls/tmp can be used to do some automatic periodic cleanup of old, large files.

The intended usage of this dual-directory structure is that the smaller files are written to somewhere away from /dls/tmp so then they're not deleted periodically and can be referred to later if needed, whilst the larger files are written to the user's directory in /dls/tmp so then they are deleted periodically. Therefore, it is advised that the directory provided to Opt-ID is not a directory in /dls/tmp/FedID; this is not only because of potential deletion of the smaller files, but also because passing a directory in /dls/tmp/FedID can cause some confusion regarding the directory that is subsequently created by Opt-ID in /dls/tmp/FedID.

In particular, it is advised that the directory passed to Opt-ID is one within /dls/technical/id, as this is where output data from other Opt-ID jobs has typically been placed.

For example, if the directory /dls/technical/id/test/ is passed to Opt-ID, the expected directory structures right after having run a sort job on a cluster is given below:

/dls/technical/id/test/:

• test_sort.json (file)
• test_sort.mag (file)
• test_sort.h5 -> /dls/tmp/FedID/test/test_sort.h5 (symlink to a file)
• generate_report.sh (file)
• restart_sort.sh (file)
• logfiles/ (directory)
• genomes -> /dls/tmp/FedID/test/genomes/ (symlink to a directory)
• process_genome_output -> /dls/tmp/FedID/test/process_genome_output/ (symlink to a directory)

/dls/tmp/FedID/test/:

• test_sort.h5 (file, the symlink /dls/technical/id/test/test_sort.h5 points to this file)
• genomes/ (directory, the symlink /dls/technical/id/test/genomes points to this directory)
• process_genome_output/ (directory, the symlink /dls/technical/id/test/process_genome_output points to this directory)

As another example, for the same directory being passed but instead a shim job being run on a cluster, the expected directory structures right after the job are:

/dls/technical/id/test/:

• test_shim.json (file)
• test_shim.mag (file)
• test_shim.h5 -> /dls/tmp/FedID/test/test_shim.h5 (symlink to a file)
• generate_report.sh (file)
• compare_shim.sh (file)
• logfiles/ (directory)
• shimmed_genomes -> /dls/tmp/FedID/test/shimmed_genomes/ (symlink to a directory)
• process_genome_output -> /dls/tmp/FedID/test/process_genome_output/ (symlink to a directory)

/dls/tmp/FedID/test/:

• test_shim.h5 (file, the symlink /dls/technical/id/test/test_shim.h5 points to this file)
• shimmed_genomes/ (directory, the symlink /dls/technical/id/test/shimmed_genomes points to this directory)
• process_genome_output/ (directory, the symlink /dls/technical/id/test/process_genome_output points to this directory)

Note that the filenames test_sort.* and test_shim.* are just placeholders and have been chosen only for illustrative purposes, these files can be named as desired in the YAML config file.

A process that is not done by Opt-ID is the transfer of magnet information in the Excel files provided by the supplier to .sim files. To do so, from the Excel files supplied by the supplier, create tab delimited .sim files of magnetisation. This is a manual procedure done only on Windows. Note that, currently, Opt-ID requires the magnet names in the .sim files to have leading zeros that pad out the name to 3 digits. For example, instead of '1' it should be '001'.

To get the code, clone the Opt-ID repo to the desired place in the filesystem. To set up the environment for running Opt-ID on a Linux machine, in a terminal run the following commands:

module load python/3
module load global/cluster
export PYTHONPATH=$PYTHONPATH:/path/to/Opt-ID

where /path/to/Opt-ID is the path to the root directory of the cloned repo. (There is a change to how python is used to run the code which is detailed in the next section, and so the third command is to enable python to find the code in the repo).

The main script that is used for running Opt-ID is IDSort/src/optid.py. It should be run using the syntax python -m IDSort.src.optid as opposed to python /path/to/Opt-ID/IDSort/src/optid.py.

There are two sets of flags from which one flag from each set is mandatory to be passed to Opt-ID, and the rest are optional and have sensible default values if they are not provided.

The mandatory sets of flags are

• --sort vs --shim
• --cluster-on vs --cluster-off

where only one flag from each bullet point should be provided.

Examples of running Opt-ID with the bare mininum flags and parameters it needs are:

python -m IDSort.src.optid --sort --cluster-on /path/to/yaml /path/to/data/dir
python -m IDSort.src.optid --shim --cluster-off /path/to/yaml /path/to/data/dir

These are used for specifying what type of job is desired.

These are used for specifying whether the job is run on the local machine or submitted to run on a cluster.

These are used in conjunction with --cluster-on. Some examples of using these flags would be

python -m IDSort.src.optid --sort --cluster-on --node-os rhel7 /path/to/yaml /path/to/data/dir
python -m IDSort.src.optid --shim --cluster-on --queue low.q /path/to/yaml /path/to/data/dir

These are used in conjunction with --cluster-off. --seed is used to specify that the random number generator (RNG) should be seeded and thus produce the same output across multiple runs with the same parameters. --seed-value is specified if a particular value to seed the RNG is desired (by default its value is 1). Some examples of using these flags would be

python -m IDSort.src.optid --sort --cluster-off --seed /path/to/yaml /path/to/data/dir
python -m IDSort.src.optid --shim --cluster-off --seed --seed-value 30 /path/to/yaml /path/to/data/dir

The YAML config files contain the parameters used by the various scripts that Opt-ID runs. The top-level sections of the YAML config files are the script names minus the .py and the subsections are the different parameters passed to that particular script. For the most part, the subsection names are exactly the same as the script parameters they're associated to, for example, the id_setup.py script has a --periods flag, and the YAML subsection corresponding to that parameter is id_setup.periods.

A few exceptions exist to try and be more descriptive with what the parameter is, for example, process_genome.py refers to the files it's given as elements of the args list, but in the YAML the corresponding subsection for a shim job is process_genome.readable_genome_file which is hopefully a more useful description.

Examples of YAML config files can be found in the IDSort/example_configs directory. There are some placeholder values in these config files that aren't valid values for their associated section in the YAML, and the following sections detail the changes that need to be made to the example config files to get them in a state ready to run a job.

There are three values that need to be changed:

• magnets.hmags
• magnets.hemags
• magnets.htmags

Their values should be absolute paths to any .sim files of the relevant type.

There are five values that need to be changed:

• magnets.hmags
• magnets.hemags
• magnets.htmags
• process_genome.readable_genome_file
• mpi_runner_for_shim_opt.bfield_filename

The first three are the same as in the sort config example. The value of process_genome.readable_genome_file should be an absolute path to the .inp file that is used to start the shim job from. The value of mpi_runner_for_shim_opt.bfield_filename should be an absolute path to the .h5 file that is converted from .bfield files that are produced by igor.

Note that, currently, the use of the igor2h5.py script hasn't yet been integrated into the YAML configuration file for Opt-ID, so the process of converting .bfield data into .h5 data is one that needs to be done by manually executing the igor2h5.py script (or by any other means) prior to running a shim job with Opt-ID.

All the autogenerated scripts can be executed from anywhere in the filesystem, it's not necessary for the current working directory to be the same directory that the script is in.

Due to the facts that

• these scripts are generated on a job-by-job basis and are only meant to be run for the particular data within the directory the scripts are in
• the structure of the data directories are fixed and known in advance

when it comes to passing parameters to these scripts they are aware of the specific directories that the files they're expecting should be in, so only filenames need to be given to them and not absolute or even relative filepaths. Concrete examples are given below in the generate_report.sh and compare_shim.sh sections that hopefully explain in more detail how to pass parameters to these scripts.

This script is used to create a report with some useful data visualisation in a PDF file. For a sort job it can be passed multiple .genome and .inp files, and for a shim job it can be passed multiple .h5 files that are associated to the "full genomes" (as opposed to the smaller-sized "compare genomes") in the shim output.

For a sort job, Opt-ID will look in both the genomes/ and process_genome_output/ directories for the given .genome and .inp files, and for a shim job Opt-ID will look in the shimmed_genomes/ directory for the given .h5 files. Therefore, the parameters passed to generate_report.sh should only be the filenames and not filepaths.

For example, for a sort job, the correct way to pass a genome and a .inp file to the script would be

/path/to/generate_report.sh foo.genome bar.inp

as opposed to

/path/to/generate_report.sh genomes/foo.genome process_genome_output/bar.inp

Another example: for a shim job, the correct way to pass .h5 files to the script would be

/path/to/generate_report.sh foo.h5 bar.h5

as opposed to

/path/to/generate_report.sh shimmed_genomes/foo.h5 shimmed_genomes/bar.h5

An optional --report-filename flag can be passed before the files to specify the name of the PDF file, and genome reports are stored in the genome_reports/ directory within the directory passed to Opt-ID. Report filenames should have a .pdf extension to enable a simple check between the report filename parameter and .genome/.inp file parameters that follow it. The --report-filename option can be omitted and in that case the report filename will be a concatenation of all the filenames passed with an underscore character "_" as the separator between the filenames.

An example of using the --report-filename flag is

/path/to/generate_report --report-filename report.pdf foo.genome bar.inp

This script requires no parameters and can be run simply as /path/to/restart_sort.sh, Opt-ID will take care of loading the YAML config of the previous sort job and will use all the same flags and paramters as the original sort job. One example is that if the original sort job was run on a cluster, so will the restart-sort job, and another example is that the same .json, .mag and .h5 (lookup table) files from the original sort job will be reused in the restart-sort job instead of being regenerated.

This can be passed a single .genome file that is in the shimmed_genomes/ directory and it will generate a human readable diff between the original and shimmed genomes that will be written to the shim_diffs/ directory. It's not necessary to pass the original genome to this script, Opt-ID will take care of finding it so only the shimmed genome needs to be given as a parameter.

Similarly to what generate_report.sh does, compare_shim.sh will look in the shimmed_genomes/ directory so only filenames should be passed to it and not filepaths. An example of using this script would be:

/path/to/compare_shim.sh foo.genome

An optional --diff-filename flag can be passed before the shimmed genome file to specify the filename of the human readable diff. Currently Opt-ID appends a .txt extension to the filename so it's not necessary to put that in the parameter. Again, similarly to what generate_report.sh does, if this flag is omitted then the diff filename is a concatenation of the original genome and shimmed genome filenames with an underscore character as the separator, and then also prepended with shim_. For example, if the original genome is foo.genome and the shimmed genome is bar.genome, then if the --diff-filename flag is omitted then the diff filename would be shim_foo.genome_bar.genome.txt. An example of using the --diff-filename flag is

/path/to/compare_shim.sh --diff-filename my_shim foo.genome

There are several options that Opt-ID has but are only meant to be used by the autogenerated shell scripts and not intended to be invoked directly by a user; therefore, these options aren't of much interest to users and only of potential interest to developers. The following are just some useful notes to any developers viewing this document:

• these options are related to those kinds of processes that a user would want to do that aren't full sort/shim jobs that were referred to in the "Overview of how to use Opt-ID" section of this document
• these options are all used by the autogenerated shell scripts that were also referred to in the "Overview of how to use Opt-ID" section, hence why the users need not directly use them, the autogenerated scripts should take care of using these "hidden options" where necessary
• these are also processes that are done after a sort/shim, so they assume the existence of a YAML config that has already been used for the sort/shim job, as well as any output data from a sort/shim job

This option starts off the process of using the IDSort/src/genome_report_template.ipynb file to generate a Jupyter notebook file, and then running it to produce a PDF report.

This option starts off the process of reusing the same YAML config file that was used for the sort job to get all the parameters used for the original sort job, and then running Opt-ID to generate genomes from an initial population as opposed to generating genomes from scratch.

This option starts off the process of comparing the given shimmed genome to the original genome that was used to start the shim job.

Navigate to the root directory of the Opt-ID repo:

cd /path/to/Opt-ID

To run all the tests:

python -m pytest IDSort/test/

To run a particular test in the test/ directory, it can be specified in the path in the above command. For example, to run IDSort/test/magnets_test.py:

python -m pytest IDSort/test/magnets_test.py

Get more details at codescene.io.