def job_execution_wrapper(data):
redis_conn = redis.Redis(connection_pool=POOL)
job = get_current_job()
_context = {}
_context['redis_conn'] = redis_conn
_context['response_channel'] = data['broker_response_channel']
_context['job_id'] = job.id
_context['data_sequence_no'] = data['data_sequence_no']
_context['api_key'] = data['extra_params']['api_key']
# Register Job Running event
_update_job_event(
_context,
job_running_template(_context['data_sequence_no'], job.id)
)
result = {}
try:
if data["function_name"] == "grade_submission":
# Run the job
answer_file_path = config.answer_file_path
result = _submit(data["data"], answer_file_path, _context)
# Register Job Complete event
if data["data"]["round"] == 2:
# Hide scores in case of round 2
del result["score"]
del result["score_secondary"]
_update_job_event(
_context,
job_info_template(
_context, "Scores Submitted Successfully ! Please remember to upload your code and a description of your approach (as a short paper) to https://gitlab.crowdai.org , as a private repository, tto be eligible for the final round-2 rankings.")
)
_update_job_event(
_context,
job_complete_template(_context, result)
)
else:
_error_object = job_error_template(
job.id,
"Function not implemented error"
)
_update_job_event(
_context,
job_error_template(job.id, result)
)
result = _error_object
except Exception as e:
_error_object = job_error_template(
_context['data_sequence_no'],
job.id,
str(e)
)
_update_job_event(_context, _error_object)
return result
def grade_submission(data, _context):
file_key = data["file_key"]
_update_job_event(
_context, job_info_template(_context,
"Post Processing MIDI file...."))
pruned_filekey, converted_filekeys = post_process_midi(
_context, POOL, file_key)
processed_filekeys = converted_filekeys
print("Making submission on crowdAI")
api = CROWDAI_API(config.CROWDAI_TOKEN)
api.authenticate_participant(_context["api_key"])
meta = {}
meta["file_key"] = file_key
#meta["processed_filekeys"] = processed_filekeys
submission = api.create_submission(config.challenge_id)
submission.score = config.SCORE_DEFAULT
submission.score_secondary = config.SCORE_SECONDARY_DEFAULT
submission.grading_status = "graded"
submission.meta = meta
submission.update()
submission_id = submission.id
print("Submitted : ", submission)
register_submission_on_redis(_context, POOL, submission_id, pruned_filekey,
processed_filekeys)
message_for_participants = """
Score (mu) : {}
Secondary_score (sigma) : {}
Please note that these scores are only the initial scores,
and they will change over time as your submission is
evaluated by the human volunteers.
""".format(submission.score, submission.score_secondary)
_update_job_event(_context,
job_info_template(_context, message_for_participants))
result = {'result': 'submission recorded'}
result['score_mu'] = submission.score
result['score_sigma'] = submission.score_secondary
_update_job_event(_context, job_complete_template(_context, result))
def _submit(client_payload, answer_file_path, context):
"""
takes a list of predicted heights and actual heights
and computes the score
and prepares the plots for submission to the leaderboard
"""
file_key = client_payload["file_key"]
_update_job_event(context,
job_info_template(context, "Grading Submission...."))
_payload = {}
_meta = {}
_meta['file_key'] = file_key
_payload["meta"] = _meta
submission_id = report_to_crowdai(context,
_payload,
submission_id=False,
status='submitted')
print("Submission id : ", submission_id)
try:
localfilepath = download_file(context, file_key)
_client_payload = {}
_client_payload["submission_file_path"] = localfilepath
_result_object = config.evaluator._evaluate(_client_payload, context)
print _result_object
_payload = _result_object
report_to_crowdai(context,
_payload,
submission_id=submission_id,
status='graded')
# Clean up file if possible
os.remove(localfilepath)
return _result_object
except Exception as e:
# Report to crowdAI
if "meta" in _payload.keys():
del _payload["meta"]
report_to_crowdai(context,
_payload,
submission_id=submission_id,
status='failed',
message=str(e))
# raise the exception again
# so that it can be handled further down the chain
raise e
def job_execution_wrapper(data):
redis_conn = redis.Redis(connection_pool=POOL)
job = get_current_job()
_context = {}
_context['redis_conn'] = redis_conn
_context['response_channel'] = data['broker_response_channel']
_context['job_id'] = job.id
_context['data_sequence_no'] = data['data_sequence_no']
_context['api_key'] = data['extra_params']['api_key']
# Register Job Running event
_update_job_event(
_context, job_running_template(_context['data_sequence_no'], job.id))
result = {}
try:
if data["function_name"] == "grade_submission":
# Run the job
answer_file_path = config.answer_file_path
result = _submit(data["data"], answer_file_path, _context)
# Register Job Complete event
_update_job_event(
_context,
job_info_template(
_context,
"Scores Submitted Successfully ! Mean Absolute Percentage Error (MAPE) : %s"
% (str(result['score']))))
_update_job_event(_context,
job_complete_template(_context, result))
else:
_error_object = job_error_template(
job.id, "Function not implemented error")
_update_job_event(_context, job_error_template(job.id, result))
result = _error_object
except Exception as e:
_error_object = job_error_template(_context['data_sequence_no'],
job.id, str(e))
_update_job_event(_context, _error_object)
return result
def split_midi_into_chunks(_context, midifile, track_length, target_directory):
cumulative_ticks = 0
cumulative_time = 0
relative_time = 0
track_index = 0
SPLITS = []
for _ in range(config.MIDI_NUM_SPLITS):
SPLITS.append(mido.MidiTrack())
for _message in midifile.tracks[0]:
cumulative_ticks += _message.time
delta_s = mido.tick2second(
_message.time,
midifile.ticks_per_beat,
DEFAULT_MIDI_TEMPO
)
if random.randint(0, 100) < 5:
"""
Report progress with a probability of 5%
"""
progress_step_offset = 0
progress_step_weight = 0.33
percent_complete = (cumulative_time * 1.0 / track_length) * 100
update_progress(
_context,
(progress_step_offset + (progress_step_weight * percent_complete))
)
cumulative_time += delta_s
relative_time += delta_s
if not _message.is_meta:
SPLITS[track_index].append(_message.copy())
else:
"""
Ignore all meta messages
"""
pass
if relative_time >= (track_length*1.0/config.MIDI_NUM_SPLITS):
relative_time = 0
track_index += 1
_update_job_event(
_context,
job_info_template(
_context, "Saving split chunks..."))
split_file_paths = []
split_length_sum = 0
# Write split files into target_directory
for _idx, _track in enumerate(SPLITS):
_m = mido.MidiFile()
_m.ticks_per_beat = midifile.ticks_per_beat
_m.tracks.append(_track)
split_length_sum += _m.length
target_file_path = "{}/{}.midi".format(
target_directory,
str(uuid.uuid4())
)
_m.save(target_file_path)
split_file_paths.append(target_file_path)
progress_step_offset = 33
progress_step_weight = 0.25
percent_complete = (_idx * 1.0 / len(SPLITS)) * 100
update_progress(
_context,
(progress_step_offset + (progress_step_weight * percent_complete))
)
return split_file_paths
def post_process_midi(_context, redis_pool, filekey):
"""
Helper file to post process midi file
"""
local_directory_path, local_filepath = download_midi(_context, filekey)
"""
# Load and validate midi file
"""
_update_job_event(
_context,
job_info_template(
_context, "Validating MIDI file..."))
midi, track_length = load_and_validate_midi(_context, local_filepath)
_update_job_event(
_context,
job_info_template(
_context, "MIDI file validated..."))
"""
# Split midi file into NUMBER_OF_PARTS (180)
"""
_update_job_event(
_context,
job_info_template(
_context, "Splitting file into 120 chunks of ~30 seconds each..."))
split_file_paths = split_midi_into_chunks(
_context,
midi,
track_length,
local_directory_path
)
"""
# Convert midi files to dataURI
"""
_update_job_event(
_context,
job_info_template(
_context, "Encoding individual chunks..."))
pruned_filekey = filekey.split("/")[-1]
converted_filekeys = convert_midi_files_to_json(
_context,
[local_filepath]+split_file_paths,
pruned_filekey
)
"""
# Upload to target directory on S3
"""
_update_job_event(
_context,
job_info_template(
_context, "Saving encoded chunks..."))
upload_processed_files_to_s3(_context, local_directory_path, pruned_filekey)
"""
# Clean up
"""
_update_job_event(
_context,
job_info_template(
_context, "Cleaning up..."))
shutil.rmtree(local_directory_path)
# Add relevant entries in redis queues
return pruned_filekey, converted_filekeys
def _submit(client_payload, answer_file_path, context):
"""
takes a list of predicted heights and actual heights
and computes the score
and prepares the plots for submission to the leaderboard
"""
file_key = client_payload["file_key"]
_update_job_event(
context,
job_info_template(
context, "Grading Submission....")
)
if "round" not in client_payload.keys():
raise Exception("""
The round parameter has not been specified. Please upgrade your
crowdai client to atleast version 1.0.21 by :
pip install -U crowdai
and then update your submission code by following the latest instructions
from :
https://github.com/crowdAI/ieee_investment_ranking_challenge-starter-kit#submission-of-predicted-file-to-crowdai
""")
round_id = client_payload["round"]
assert round_id in config.crowdai_round_id_map.keys(), \
"Unknown Round ID Passed. Allowed values : {}".format(
str(config.crowdai_round_id_map.keys())
)
crowdai_round_id = config.crowdai_round_id_map[round_id]
_payload = {}
_meta = {}
_meta['file_key'] = file_key
_payload["meta"] = json.dumps(_meta)
_payload["challenge_round_id"] = crowdai_round_id
submission_id = report_to_crowdai(
context,
_payload,
submission_id=False,
status='submitted')
print("Submission id : ", submission_id)
try:
localfilepath = download_file(context, file_key)
_client_payload = {}
_client_payload["submission_file_path"] = localfilepath
_result_object = config.evaluator._evaluate(
client_payload=_client_payload,
round_indicator=round_id,
_context=context)
print _result_object
_payload = _result_object
report_to_crowdai(
context,
_payload,
submission_id=submission_id,
message = "graded successfully",
status='graded')
# Clean up file if possible
os.remove(localfilepath)
return _result_object
except Exception as e:
# Report to crowdAI
if "meta" in _payload.keys():
del _payload["meta"]
report_to_crowdai(
context,
_payload,
submission_id=submission_id,
status='failed',
message=str(e)
)
# raise the exception again
# so that it can be handled further down the chain
raise e