More about the Environment

In most use cases you will interact with the Envrionemnt to do your numerical simulations. The environment is your handyman for your numerical experiments, it sets up new trajectories, keeps log files and can be used to distribute your simulations onto several cpus.

Note in case you use the environment there is no need to call f_store() for data storage, this will always be called before the runs and at the end of a single run automatically.

You start your simulations by creating an environment object:

>>> env = Environment(trajectory='trajectory',\
                 comment='',\
                 dynamically_imported_classes=None,\
                 log_folder=None,\
                 multiproc=False,\
                 ncores=1,\
                 wrap_mode=pypetconstants.WRAP_MODE_LOCK,\
                 continuable=1,\
                 use_hdf5=True,\
                 filename=None,\
                 file_title=None,\
                 purge_duplicate_comments=True,\
                 small_overview_tables=True,\
                 large_overview_tables=True,\
                 results_per_run=0,\
                 derived_parameters_per_run=0,\
                 git_repository = None,\
                 git_message=''):

You can pass the following arguments:

  • trajectory

    The first argument trajectory can either be a string or a given trajectory object. In case of a string, a new trajectory with that name is created. You can access the new trajectory via v_trajectory property. If a new trajectory is created, the comment and dynamically imported classes are added to the trajectory.

  • dynamically_imported_classes

    The argument dynamically_imported_classes is important if you have written your own parameter or result classes, you can pass these either as class variables MyCustomParameterClass or as strings leading to the classes in your package: ‘mysim.myparameters.MyCustomParameterClass’. If you have several classes, just put them in a list dynamically_imported_classes=[MyCustomParameterClass,MyCustomResultClass]. The trajectory needs to know your custom classes in case you want to load a custom class from disk and the trajectory needs to know how they are built.

    It is VERY important, that every class name is UNIQUE. So you should not have two classes named ‘MyCustomParameterClass’ in two different python modules! The identification of the class is based only on its name and not its path in your packages.

  • multiproc

    multiproc specifies whether or not to use multiprocessing (take a look at Multiprocessing). Default is 0 (False). If you use multiprocessing, all your data and the tasks you compute must be pickable!

  • ncores

    If multiproc is 1 (True), this specifies the number of processes that will be spawned to run your experiment. Note if you use ‘QUEUE’ mode (see below) the queue process is not included in this number and will add another extra process for storing.

  • wrap_mode

    If multiproc is 1 (True), specifies how storage to disk is handled via the storage service. Since hdf5 is not thread safe, the hdf5 storage service needs to be wrapped with a helper class to allow the interaction with multiple processes.

    There are two options:

    pypet.pypetconstants.MULTIPROC_MODE_QUEUE: (‘QUEUE’)

    Another process for storing the trajectory is spawned. The sub processes running the individual single runs will add their results to a multiprocessing queue that is handled by an additional process

    pypet.pypetconstants.MULTIPROC_MODE_LOCK: (‘LOCK’)

    Each individual process takes care about storage by itself. Before carrying out the storage, a lock is placed to prevent the other processes to store data.

    If you don’t want wrapping at all use pypet.pypetconstants.MULTIPROC_MODE_NONE (‘NONE’)

    If you have no clue what I am talking about, you might want to take a look at multiprocessing in python to learn more about locks, queues and thread safety and so forth.

  • continuable

    Whether the environment should take special care to allow to resume or continue crashed trajectories. Default is 1 (True). Everything must be pickable in order to allow continuing of trajectories (take a look at Resuming an Experiment) In order to resume trajectories use f_continue_run().

  • log_folder

    The log_folder specifies where all log files will be stored. The environment will create a sub-folder with the name of the trajectory where all txt files will be put. The environment will create a major logfile (main.txt) incorporating all messages of the current log level and beyond and a log file that only contains warnings and errors warnings_and_errors.txt. Moreover, if you use multiprocessing, there will be a log file for every process named proces_XXXX.txt with XXXX the process id. If you don’t set a log level elsewhere before, the standard level will be INFO (if you have no clue what I am talking about, take a look at the logging module).

  • use_hdf5:

    If you want to use the standard hdf5 storage service provided with this package, set use_hdf5=True. You can specify the name of the hdf5 file and, if it has to be created new, the file title. If you want to use your own storage service (You don’t have an SQL one do you?), set use_hdf5=False and add your custom storage service to the trajectory:

    >>> env.v_trajectory.v_storage_service = MyCustomService(...)

  • purge_duplicate_comments

    If you add a result via pypet.trajectory.SingleRun.f_add_result() or a derived parameter pypet.trajectory.SingleRun.f_add_derived_parameter() and you set a comment, normally that comment would be attached to each and every instance. This can produce a lot of unnecessary overhead if the comment is the same for every result over all runs. If hdf5.purge_duplicate_comments=1 than only the comment of the first result or derived parameter instance created is stored or comments that differ from this first comment. You might want to take a look at Purging duplicate Comments.

  • small_overview_tables

    Whether the small overview tables should be created. Small tables are giving overview about ‘config’,’parameters’,’derived_parameters_trajectory’, ‘derived_parameters_runs_summary’, ‘results_trajectory’,’results_runs_summary’. You might want to check out Overview Tables.

  • large_overview_tables

    Whether to add large overview tables. This encompasses information about every derived parameter and result and the explored parameters in every single run. If you want small hdf5 files, this is the first option to set to False.

  • results_per_run

    Expected results you store per run. If you give a good/correct estimate storage to hdf5 file is much faster in case you store LARGE overview tables.

    Default is 0, i.e. the number of results is not estimated!

  • derived_parameters_per_run

    Analogous to the above.

  • git_repository

    If your code base is under git version control you can specify the path (relative or absolute) to the folder containing the .git directory. See also Git Integration.

  • git_message

    Message passed onto git command.

Config Data added by the Environment

The Environment will automatically add some config settings to your trajectory. Thus, you can always look up how your trajectory was run. This encompasses all above named parameters as well as some information about the environment. This additional information includes a timestamp and a SHA-1 hash code that uniquely identifies your environment. If you use git integration (Git Integration), the SHA-1 hash code will be the one from your git commit. Otherwise the code will be calculated from the trajectory name, the current time, and your current pypet version.

The environment will be named environment_XXXXXXX_XXXX_XX_XX_XXhXXmXXs. The first seven X are the first seven characters of the SHA-1 hash code followed by a human readable timestamp.

All information about the environment can be found in your trajectory under config.environment.environment_XXXXXXX_XXXX_XX_XX_XXhXXmXXs. Your trajectory could potentially be run by several environments due to merging or extending an existing trajectory. Thus, you will be able to track how your trajectory was build over time.

Overview Tables

Overview tables give you a nice summary about all parameters and results you needed and computed during your simulations. They will be placed under the subgroup overview at the top-level in your trajectory group in the hdf5 file. In addition, for every single run there will be a small overview table about the explored parameter values of that run (see also Storing).

However, if you have many runs and results and derived_parameters, I would advice you to switch of the result, derived parameter and explored parameter overview in each single run. You don’t have to do that by hand, simply use f_switch_off_large_overview() or f_switch_off_all_overview() to disable all tables. Or specify whether you want large and/or small tables on environment creation.

Purging duplicate Comments

If you added a result with the same name and same comment in every single run, this would create a lot of overhead. Since the very same comment would be stored in every node in the hdf5 file. For instance, during a single run you call traj.f_add_result(‘my_result’,42, comment=’Mostly harmless!’) and the result will be renamed to results.run_00000000.my_result. After storage in the node associated with this result in your hdf5 file, you will find the comment ‘Mostly harmless!’. If you call traj.f_add_result(‘my_result’,-55, comment=’Mostly harmless!’) in another run again, let’s say run_00000001, the name will be mapped to results.run_00000001.my_result. But this time the comment will not be saved to disk, since ‘Mostly harmless!’ is already part of the very first result with the name ‘my_result’. Note that comments will be compared and storage will only be discarded if the strings are exactly the same. Moreover, the comment will only be compared to the comment of the very first result, if all comments are equal except for the very first one, all of these equal comments will be stored!

In order to allow the purge of duplicate comments you need the summary overview tables.

Furthermore, consider if you reload your data, the result instance results.run_00000001.my_result won’t have a comment only the instance results.run_00000000.my_result.

If you do not want to purge duplicate comments, set the config parameter ‘purge_duplicate_comments’ to 0 or False.

Multiprocessing

For an example on multiprocessing see 04 Multiprocessing.

The following code snippet shows how to enable multiprocessing with 4 cpus and a queue.

env = Environment(self, trajectory='trajectory',
             comment='',
             dynamically_imported_classes=None,
             log_folder='../log/',
             use_hdf5=True,
             filename='../experiments.h5',
             file_title='experiment',
             multiproc=True,
             ncores=4,
             wrap_mode='QUEUE')

Note that hdf5 is not thread safe, so you cannot use the standard hdf5 storage service out of the box. However, if you want multiprocessing, the environment will automatically provide wrapper classes for the hdf5 storage service to allow safe data storage.

There are two different modes that are supported. You can choose between them via setting wrap_mode. You can choose between ‘QUEUE’ and ‘LOCK’. If you have your own service that is already thread safe you can also choose ‘NONE’ to skip wrapping.

If you chose the ‘QUEUE’ mode, there will be an additional process spawned that is the only one writing to the hdf5 file. Everything that is supposed to be stored is send over a queue to the process. This has the advantage that your worker processes are only busy with your simulation and are not bothered with writing data to a file. More important, they don’t spend time waiting for other processes to release a thread lock to allow file writing. The disadvantage is that this storage relies a lot on pickling of data, so often your entire trajectory is send over the queue.

If you chose the ‘LOCK’ mode, every process will pace a lock before it opens the hdf5 file for writing data. Thus, only one process at a time stores data. The advantage is that your data does not need to be send over a queue over and over again. Yet, your simulations might take longer since processes have to wait for each other to release locks quite often.

IMPORTANT: In order to allow multiprocessing, all your data and objects of your simulation need to be serialized with pickle. But don’t worry, most of the python stuff you use is automatically picklable.

Git Integration

The environment can make use of version control. If you manage your code with git you can trigger automatic commits with the environment to get a proper snapshot of the code you actually use. This ensures that your experiments are repeatable! In order to use the feature of git integration you additionally need GitPython.

To trigger an automatic commit simply pass the arguments git_repository and git_message to the Environment constructor. git_repository specifies the path to the folder containing the .git directory. git_message is optional and adds the corresponding message to the commit. Note that the message will always be augmented with some short information about the trajectory you are running.

The commit SHA-1 hash and some other information about the commit will be added to the config subtree of your trajectory, so you can easily recall that commit from git later on.

The automatic commit will only commit changes in files that are currently tracked by your git repository, it will NOT add new files. So make sure that if you create new files you put them into your repository before running an experiment.

The autocommit function is similar to calling $ git add -u and $ git commit -m ‘Some Message` in your linux console!

Running an Experiment

In order to run an experiment, you need to define a job or a top level function that specifies your simulation. This function gets as first positional argument the trajectory, or to be more precise a single run (SingleRun), and optionally other positional and keyword arguments of your choice.

def myjobfunc(traj,*args,**kwargs)
    #Do some sophisticated simulations with your trajectory
    ...

In order to run this simulation, you need to hand over the function to the environment, where you can also specify the additional arguments and keyword arguments using f_run():

env.f_run(myjobfunc,*args,**kwargs)

The argument list args and keyword dictionary kwargs are directly handed over to the myjobfunc during runtime.

Note that the first postional argument used by myjobfunc is not a full pypet.trajectory.Trajectory() but only a ~pypet.trajectory.SingleRun (also see Single Runs). There is not much difference to a full trajectory. You have slightly less functionality and usually no access to the fully explored parameters but only to a single parameter space point.

Resuming an Experiment

If all of your data is picklable, you can use the config parameter continuable=1. This will create a ‘.cnt’ file with the name of your trajectory in the folder where your final hdf5 file will be placed. The .cnt file is your safety net for data loss due to a computer crash. If for whatever reason your day or week-long lasting simulation was interrupted, you can resume it without recomputing already obtained results. Note that this works only if the hdf5 file is not corrupted and with interruptions due to computer crashes, like power failure etc. If your simulations crashed due to errors in your code, there is no way to restore that!

You can resume a crashed trajectory via f_continue_run() with the name of the corresponding ‘.cnt’ file.

env = Environment()


env.f_continue_run('./experiments/my_traj_2015_10_21_04h29m00s.cnt')