easyvvuq_restart_campaign_tutorial.py

import os
import easyvvuq as uq
import chaospy as cp
from pprint import pprint
from shutil import rmtree

# 0. Setup some variables describing app to be run
#
#    gauss.py is in current directory and takes one input file
#    and writes to 'output.csv'.
work_dir = os.path.dirname(os.path.abspath(__file__))


input_filename = 'gauss_in.json'
out_file = "output.csv"
template = "{}/gauss.template".format(work_dir)


# clear the target campaign dir
campaign_work_dir = os.path.join(
    work_dir, "easyvvuq_restart_campaign_tutorial"
)
if os.path.exists(campaign_work_dir):
    rmtree(campaign_work_dir)
os.makedirs(campaign_work_dir)


# 1. Set up a fresh campaign called "gauss"
db_location = "sqlite:///" + campaign_work_dir + "/campaign.db"
my_campaign = uq.Campaign(
    name="gauss",
    db_location=db_location,
    work_dir=campaign_work_dir
)


# 2. Parameter space definition
params = {
    "sigma": {
        "type": "float",
        "min": 0.0,
        "max": 100000.0,
        "default": 0.25
    },
    "mu": {
        "type": "float",
        "min": 0.0,
        "max": 100000.0,
        "default": 1
    },
    "num_steps": {
        "type": "integer",
        "min": 0,
        "max": 100000,
        "default": 10
    },
    "out_file": {
        "type": "string",
        "default": "output.csv"
    }
}

# 3. Wrap Application
#    - Define a new application (we'll call it 'gauss'),
#      and the encoding/decoding elements it needs
#    - Also requires a collation element - his will be responsible for
#      aggregating the results
encoder = uq.encoders.GenericEncoder(
    template_fname=template,
    target_filename=input_filename
)

decoder = uq.decoders.SimpleCSV(
    target_filename=out_file,
    output_columns=["Step", "Value"]
)

execute = uq.actions.ExecuteLocal(
    "python3 {}/gauss.py {}".format(work_dir, input_filename)
)

actions = uq.actions.Actions(
    uq.actions.CreateRunDirectory(root=campaign_work_dir, flatten=True),
    uq.actions.Encode(encoder),
    execute,
    uq.actions.Decode(decoder)
)

my_campaign.add_app(
    name="gauss",
    params=params,
    actions=actions
)


# 4. Specify Sampler
#    -  vary the `mu` parameter only
vary = {
    "mu": cp.Uniform(1.0, 100.0),
}

my_sampler = uq.sampling.RandomSampler(vary=vary)

my_campaign.set_sampler(my_sampler)

# 5. Run the cases and Collate output
my_campaign.execute(nsamples=2).collate()


# 9. Print the list of runs
print("=" * 40)
print("list of runs in the my_campaign")
print("=" * 40)
pprint(my_campaign.list_runs())


# 11. Load state in new campaign object
print("Reloading campaign...")

reloaded_campaign = uq.Campaign(
    name="gauss",
    db_location=db_location
)

sampler = reloaded_campaign.get_active_sampler()
reloaded_campaign.set_sampler(sampler, update=True)

# 12. Draw some more samples, execute the runs, and collate onto existing
# dataframe

reloaded_campaign.execute(nsamples=4).collate()


# 13. Print the list of runs again
print("=" * 40)
print("list of runs in the reloaded_campaign")
print("=" * 40)
pprint(reloaded_campaign.list_runs())

# 14. Run Analysis
#     - Calculate bootstrap statistics for collated data
# stats = uq.analysis.EnsembleBoot(groupby=["mu"], qoi_cols=["Value"])
# reloaded_campaign.apply_analysis(stats)
# print("stats:\n", my_campaign.get_last_analysis())