ENVS.notebook_tutorial

##
# The following variables set the environment, _HOME and _CONDA should be set to
# a default location to save you having to continually set these in your shell.
# _ENVIRONMENT can usually be left alone, it helps when running via symlinked
# pipeline directories which is common when running several distinct
# environments
#
export ICENET_ENVIRONMENT=${ICENET_ENVIRONMENT:-"$( basename `realpath .` )"}
export ICENET_HOME=${ICENET_HOME:-${HOME}/icenet/${ICENET_ENVIRONMENT}}
export ICENET_CONDA=${ICENET_CONDA:-/data/hpcdata/users/${USER}/miniconda3/envs/icenet}

##
# Setting the data download arguments for variables and pressures/depths
#
export DATA_ARGS_ERA5="--vars uas,vas,tas,zg --levels ',,,500|250'"
export DATA_ARGS_ORAS5=""

##
# Linking the preprocessing arguments to the above downloads, as well as the
# SIC ground truth for the model
#
export PROC_ARGS_ERA5="--abs uas,vas --anom tas,zg500,zg250"
export PROC_ARGS_ORAS5=""
export PROC_ARGS_SIC="--abs siconca --trends siconca --trend-lead 7"

##
# Training and prediction parameters
#
export TRAIN_STATIC_ARGS=""
export PREDICT_STATIC_ARGS=""

export FILTER_FACTOR=0.4
export FORECAST_DAYS=7
export LAG=1
export PREP_SCRIPT=local.sh

export BATCH_SIZE=4
export WORKERS=16

##
# HPC specifics - configure as appropriate for data (CPU) and training / prediction (GPU)
#
#    --partition=medium
#    --account=medium
#    --time=12:00:00
#    --nodes=1
#
export ICENET_SLURM_ARGS="--chdir=$ICENET_HOME \
    --output=${ICENET_HOME}/logs/icenet.%j.%N.out"
export ICENET_SLURM_DATA_PART="--partition cpu-example --time 48:00:00"
export ICENET_SLURM_RUN_PART="--partition gpu-example --time 24:00:00"

##
# Training date ranges
#
TUTORIAL_PIPELINE_TRAIN_START="2020-1-1,2020-3-1"
TUTORIAL_PIPELINE_TRAIN_END="2020-2-29,2020-3-31"
TUTORIAL_PIPELINE_VAL_START="2020-4-3"
TUTORIAL_PIPELINE_VAL_END="2020-4-23"
TUTORIAL_PIPELINE_TEST_START="2020-4-1"
TUTORIAL_PIPELINE_TEST_END="2020-4-2"

##
# Training & Prediction ensemble run seeds
#
TUTORIAL_PIPELINE_ENSEMBLE_TRAIN_SEEDS="42,46"
TUTORIAL_PIPELINE_ENSEMBLE_PREDICT_SEEDS="42,46"

##
# The prefix to use for training date ranges
#
PREFIX="TUTORIAL_PIPELINE"

##
# USER NOTE: you shouldn't need to configure anything further after this point!
#

##
# Mapping variables, these shouldn't need changing but link to the usage of the
# variables in scripts, meaning you can alter the scripts simply by creating
# extra ranges and changing PREFIX
#
TRAIN_DATA_NAME=${PREFIX,,}
TRAIN_START_NAME="${PREFIX}_TRAIN_START"
TRAIN_END_NAME="${PREFIX}_TRAIN_END"
VAL_START_NAME="${PREFIX}_VAL_START"
VAL_END_NAME="${PREFIX}_VAL_END"
TEST_START_NAME="${PREFIX}_TEST_START"
TEST_END_NAME="${PREFIX}_TEST_END"

ENSEMBLE_TRAIN_SEEDS_NAME="${PREFIX}_ENSEMBLE_TRAIN_SEEDS"
ENSEMBLE_PREDICT_SEEDS_NAME="${PREFIX}_ENSEMBLE_PREDICT_SEEDS"

# What are we exporting

export TRAIN_START=${!TRAIN_START_NAME}
export TRAIN_END=${!TRAIN_END_NAME}
export VAL_START=${!VAL_START_NAME}
export VAL_END=${!VAL_END_NAME}
export TEST_START=${!TEST_START_NAME}
export TEST_END=${!TEST_END_NAME}

export ENSEMBLE_TRAIN_SEEDS=${!ENSEMBLE_TRAIN_SEEDS_NAME}
export ENSEMBLE_PREDICT_SEEDS=${!ENSEMBLE_PREDICT_SEEDS_NAME}