import plasma.global_vars as g

g.init_MPI()

import os.path

# TODO(KGF): replace this workaround for the "from plasma.conf import conf"

def is_valid_file(parser, arg):

if not (os.path.exists(arg) and os.path.isfile(arg)):

parser.error("The file %s does not exist!" % arg)

else:

return arg

import argparse

parser = argparse.ArgumentParser(prog='mpi_learn', description='FusionDL TensorFlow 1.x + mpi4py')

parser.add_argument("--input_file", "-i", # type=str,

required=False, dest="conf_file",

help="input YAML file for configuration", metavar="YAML_FILE",

type=lambda x: is_valid_file(parser, x))

args = parser.parse_args()

g.conf_file = args.conf_file

from plasma.conf import conf

from plasma.models.mpi_runner import (

mpi_train, mpi_make_predictions_and_evaluate

)

from plasma.preprocessor.preprocess import guarantee_preprocessed

from plasma.models.loader import Loader

'''

#########################################################

This file trains a deep learning model to predict

disruptions on time series data from plasma discharges.

Must run guarantee_preprocessed.py in order for this to work.

Dependencies:

conf.py: configuration of model,training,paths, and data

model_builder.py: logic to construct the ML architecture

data_processing.py: classes to handle data processing

Author: Julian Kates-Harbeck, [email protected]

This work was supported by the DOE CSGF program.

#########################################################

'''

import os

import sys

import datetime

import random

import numpy as np

import matplotlib

matplotlib.use('Agg')

sys.setrecursionlimit(10000)

if conf['model']['shallow']:

print("Shallow learning using MPI is not supported yet. ",

"Set conf['model']['shallow'] to False.")

exit(1)

if conf['data']['normalizer'] == 'minmax':

from plasma.preprocessor.normalize import MinMaxNormalizer as Normalizer

elif conf['data']['normalizer'] == 'meanvar':

from plasma.preprocessor.normalize import MeanVarNormalizer as Normalizer

elif conf['data']['normalizer'] == 'var':

# performs !much better than minmaxnormalizer

from plasma.preprocessor.normalize import VarNormalizer as Normalizer

elif conf['data']['normalizer'] == 'averagevar':

# performs !much better than minmaxnormalizer

from plasma.preprocessor.normalize import (

AveragingVarNormalizer as Normalizer

)

else:

print('unkown normalizer. exiting')

exit(1)

# set PRNG seed, unique for each worker, based on MPI task index for

# reproducible shuffling in guranteed_preprocessed() and training steps

np.random.seed(g.task_index)

random.seed(g.task_index)

only_predict = len(sys.argv) > 1

custom_path = None

if only_predict:

custom_path = sys.argv[1]

g.print_unique("predicting using path {}".format(custom_path))

#####################################################

# NORMALIZATION #

#####################################################

normalizer = Normalizer(conf)

if g.task_index == 0:

# make sure preprocessing has been run, and results are saved to files

# if not, only master MPI rank spawns thread pool to perform preprocessing

(shot_list_train, shot_list_validate,

shot_list_test) = guarantee_preprocessed(conf)

# similarly, train normalizer (if necessary) w/ master MPI rank only

normalizer.train() # verbose=False only suppresses if purely loading

g.comm.Barrier()

g.print_unique("begin preprocessor+normalization (all MPI ranks)...")

# second call has ALL MPI ranks load preprocessed shots from .npz files

(shot_list_train, shot_list_validate,

shot_list_test) = guarantee_preprocessed(conf, verbose=True)

# second call to normalizer training

normalizer.conf['data']['recompute_normalization'] = False

normalizer.train(verbose=True)

# KGF: may want to set it back...

# normalizer.conf['data']['recompute_normalization'] = conf['data']['recompute_normalization'] # noqa

loader = Loader(conf, normalizer)

g.print_unique("...done")

# TODO(KGF): both preprocess.py and normalize.py are littered with print()

# calls that should probably be replaced with print_unique() when they are not

# purely loading previously-computed quantities from file

# (or we can continue to ensure that they are only ever executed by 1 rank)

#####################################################

# TRAINING #

#####################################################

# Prevent Keras TF backend deprecation messages from mpi_train() from

# appearing jumbled with stdout, stderr msgs from above steps

g.comm.Barrier()

g.flush_all_inorder()

# reminder: ensure training has a separate random seed for every worker

if not only_predict:

mpi_train(conf, shot_list_train, shot_list_validate, loader,

shot_list_test=shot_list_test)

g.flush_all_inorder()

if conf['training']['no_validation'] and conf['training']['step_limit'] > 0:

sys.stdout.flush()

g.print_unique('SHORT TRAINING ONLY. conf.yaml (step_limit) finished without VALIDATION')

quit()

#####################################################

# TESTING #

#####################################################

# load last model for testing

loader.set_inference_mode(True)

g.print_unique('saving results')

y_prime = []

y_gold = []

disruptive = []

# TODO(KGF): check tuple unpack

(y_prime_train, y_gold_train, disruptive_train, roc_train,

loss_train) = mpi_make_predictions_and_evaluate(conf, shot_list_train,

loader, custom_path)

(y_prime_test, y_gold_test, disruptive_test, roc_test,

loss_test) = mpi_make_predictions_and_evaluate(conf, shot_list_test,

loader, custom_path)

g.print_unique('=========Summary========')

g.print_unique('Train Loss: {:.3e}'.format(loss_train))

g.print_unique('Train ROC: {:.4f}'.format(roc_train))

g.print_unique('Test Loss: {:.3e}'.format(loss_test))

g.print_unique('Test ROC: {:.4f}'.format(roc_test))

if g.task_index == 0:

disruptive_train = np.array(disruptive_train)

disruptive_test = np.array(disruptive_test)

y_gold = y_gold_train + y_gold_test

y_prime = y_prime_train + y_prime_test

disruptive = np.concatenate((disruptive_train, disruptive_test))

shot_list_test.make_light()

shot_list_train.make_light()

save_str = 'results_' + datetime.datetime.now().strftime(

"%Y-%m-%d-%H-%M-%S")

result_base_path = conf['paths']['results_prepath']

if not os.path.exists(result_base_path):

os.makedirs(result_base_path)

np.savez(result_base_path+save_str, y_gold=y_gold,

y_gold_train=y_gold_train, y_gold_test=y_gold_test,

y_prime=y_prime, y_prime_train=y_prime_train,

y_prime_test=y_prime_test, disruptive=disruptive,

disruptive_train=disruptive_train,

disruptive_test=disruptive_test, shot_list_train=shot_list_train,

shot_list_test=shot_list_test, conf=conf)

# TODO(KGF): Intel NumPy fork

# https://conda.anaconda.org/intel/linux-64/numpy-1.16.2-py36h7b7c402_0.tar.bz2

# applies cve_2019_6446_fix.patch, which unlike main NumPy, adds

# requirement for "allow_pickle=True" to savez() calls

sys.stdout.flush()

g.print_unique('finished.')