How to use the seml.database.get_collection function in seml

To help you get started, we’ve selected a few seml examples, based on popular ways it is used in public projects.

Secure your code as it's written. Use Snyk Code to scan source code in minutes - no build needed - and fix issues immediately.

github TUM-DAML / seml / seml / View on Github external
def delete_experiments(db_collection_name, sacred_id, filter_states, batch_id, filter_dict):
    collection = get_collection(db_collection_name)
    if sacred_id is None:
        if len({'PENDING', 'RUNNING', 'KILLED'} & set(filter_states)) > 0:
            detect_killed(db_collection_name, print_detected=False)

        filter_dict = build_filter_dict(filter_states, batch_id, filter_dict)
        ndelete = collection.count_documents(filter_dict)
        batch_ids = collection.find(filter_dict, {'batch_id'})
        batch_ids_in_del = set([x['batch_id'] for x in batch_ids])

        if ndelete >= 10:
            if input(f"Deleting {ndelete} configuration{s_if(ndelete)} from database collection. "
                     f"Are you sure? (y/n) ").lower() != "y":
  "Deleting {ndelete} configuration{s_if(ndelete)} from database collection.")
github TUM-DAML / seml / seml / View on Github external
filter_states: list of strings or None
        List of statuses to filter for. Will cancel all jobs from the database collection
        with one of the given statuses.
    batch_id: int or None
        The ID of the batch of experiments to cancel. All experiments that are queued together (i.e. within the same
        command line call) have the same batch ID.
    filter_dict: dict or None
        Arbitrary filter dictionary to use for cancelling experiments. Any experiments whose database entries match all
        keys/values of the dictionary will be cancelled.


    collection = get_collection(db_collection_name)
    if sacred_id is None:
        # no ID is provided: we check whether there are slurm jobs for which after this action no
        # RUNNING experiment remains. These slurm jobs can be killed altogether.
        # However, it is NOT possible right now to cancel a single experiment in a Slurm job with multiple
        # running experiments.
            if len({'PENDING', 'RUNNING', 'KILLED'} & set(filter_states)) > 0:
                detect_killed(db_collection_name, print_detected=False)

            filter_dict = build_filter_dict(filter_states, batch_id, filter_dict)

            ncancel = collection.count_documents(filter_dict)
            if ncancel >= 10:
                if input(f"Cancelling {ncancel} experiment{s_if(ncancel)}. "
                         f"Are you sure? (y/n) ").lower() != "y":
github TUM-DAML / seml / seml / View on Github external
def detect_killed(db_collection_name, print_detected=True):
    collection = get_collection(db_collection_name)
    exps = collection.find({'status': {'$in': ['PENDING', 'RUNNING']},
                            '$or': [{'slurm.array_id': {'$exists': True}}, {'': {'$exists': True}}]})
    running_jobs = get_slurm_arrays_tasks()
    old_running_jobs = get_slurm_jobs()  # Backwards compatibility
    nkilled = 0
    for exp in exps:
        exp_running = ('array_id' in exp['slurm'] and exp['slurm']['array_id'] in running_jobs
                       and (any(exp['slurm']['task_id'] in r for r in running_jobs[exp['slurm']['array_id']][0])
                            or exp['slurm']['task_id'] in running_jobs[exp['slurm']['array_id']][1]))
        exp_running |= ('id' in exp['slurm'] and exp['slurm']['id'] in old_running_jobs)
        if not exp_running:
            if 'stop_time' in exp:
                collection.update_one({'_id': exp['_id']}, {'$set': {'status': 'INTERRUPTED'}})
                nkilled += 1
                collection.update_one({'_id': exp['_id']}, {'$set': {'status': 'KILLED'}})
github TUM-DAML / seml / seml / View on Github external
def get_results(db_collection_name, fields=['config', 'result'],
                to_data_frame=False, mongodb_config=None, suffix=None,
                states=['COMPLETED'], filter_dict={}, parallel=False):
    import pandas as pd

    collection = get_collection(db_collection_name, mongodb_config=mongodb_config, suffix=suffix)

    if len(states) > 0:
        if 'status' in filter_dict:
            logging.warning("'states' argument is not empty and will overwrite 'filter_dict['status']'.")
        filter_dict['status'] = {'$in': states}

    cursor = collection.find(filter_dict, fields)
    results = [x for x in tqdm(cursor, total=collection.count_documents(filter_dict))]

    if parallel:
        from multiprocessing import Pool
        with Pool() as p:
            parsed = list(tqdm(p.imap(parse_jsonpickle, results),
        parsed = [parse_jsonpickle(entry) for entry in tqdm(results)]
github TUM-DAML / seml / seml / View on Github external
            seml_config['conda_environment'] = None

    # Set Slurm config with default parameters as fall-back option
    if slurm_config is None:
        slurm_config = {'sbatch_options': {}}
    for k, v in SETTINGS.SLURM_DEFAULT['sbatch_options'].items():
        if k not in slurm_config['sbatch_options']:
            slurm_config['sbatch_options'][k] = v
    del SETTINGS.SLURM_DEFAULT['sbatch_options']
    for k, v in SETTINGS.SLURM_DEFAULT.items():
        if k not in slurm_config:
            slurm_config[k] = v

    slurm_config['sbatch_options'] = remove_prepended_dashes(slurm_config['sbatch_options'])
    collection = get_collection(db_collection_name)
    configs = generate_configs(experiment_config)

    batch_id = get_max_in_collection(collection, "batch_id")
    if batch_id is None:
        batch_id = 1
        batch_id = batch_id + 1

    if seml_config['use_uploaded_sources']:
        uploaded_files = upload_sources(seml_config, collection, batch_id)
        uploaded_files = None

    if not no_config_check:
        check_config(seml_config['executable'], seml_config['conda_environment'], configs)
github TUM-DAML / seml / seml / View on Github external
parser.add_argument("--experiment_id", type=int, help="The experiment ID.")
    parser.add_argument("--db_collection_name", type=str, help="The collection in the database to use.")
    parser.add_argument("--verbose", default=False, type=lambda x: (str(x).lower() == 'true'),
                        help="Display more log messages.")
    parser.add_argument("--unobserved", default=False, type=lambda x: (str(x).lower() == 'true'),
                        help="Run the experiments without Sacred observers.")
    parser.add_argument("--post-mortem", default=False, type=lambda x: (str(x).lower() == 'true'),
                        help="Activate post-mortem debugging with pdb.")
    parser.add_argument("--stored-sources-dir", default=None, type=str,
                        help="Load source files into this directory before starting.")
    args = parser.parse_args()

    exp_id = args.experiment_id
    db_collection_name = args.db_collection_name

    collection = get_collection(db_collection_name)

    exp = collection.find_one({'_id': exp_id})
    use_stored_sources = args.stored_sources_dir is not None
    if use_stored_sources and not os.listdir(args.stored_sources_dir):
        assert "source_files" in exp['seml'],\
               "--stored-sources-dir was supplied but queued experiment does not contain stored source files."
        load_sources_from_db(exp, collection, to_directory=args.stored_sources_dir)

    exe, config = get_command_from_exp(exp, db_collection_name, verbose=args.verbose,
                                       unobserved=args.unobserved, post_mortem=args.post_mortem)
    config_args = ' '.join(config)

    cmd = f"python {exe} with {config_args}"
    if use_stored_sources:
        # add command without the temp_dir prefix
        # also add the temp dir for debugging purposes
github TUM-DAML / seml / seml / View on Github external
If >0, will only submit the specified number of experiments to the cluster.
        This is useful when you only want to test your setup.
    filter_dict: dict
        Dictionary for filtering the entries in the collection.
    dry_run: bool
        Just return the executables and configurations instead of running them.
    output_to_file: bool
        Pipe all output (stdout and stderr) to an output file.
        Can only be False if slurm is False.


    collection = get_collection(db_collection_name)

    if unobserved and not slurm and '_id' in filter_dict:
        query_dict = {}
        query_dict = {'status': {"$in": ['QUEUED']}}

    if collection.count_documents(query_dict) <= 0:
        logging.error("No queued experiments.")

    exps_full = list(collection.find(query_dict))

    nexps = num_exps if num_exps > 0 else len(exps_full)
    exps_list = exps_full[:nexps]