def write_setup_entries(network: metawards.Network, design_index: int,
run_ident: str, **kwargs):
global _sql_file_name
global _run_index
database: Union[sql.Connection, None] = None
try:
database = sql.connect(_sql_file_name)
# Prepare database entries
c: sql.Cursor = database.cursor()
# Write the hypercube point, ignoring repeats
# Repeat runs can create duplicate keys, ideally this would be done before entering here
# So we use 'insert or ignore' instead of 'insert'
hypercube_keys = [
var[1:] for var in get_input_header_names()[:-1] if var[0] == '.'
]
hypercube_values = tuple(
[network.params.user_params[x] for x in hypercube_keys])
i_str = f"insert or ignore into design_table({','.join(hypercube_keys)}) values ({','.join(['?'] * len(hypercube_keys))})"
c.execute(i_str, hypercube_values)
database.commit()
# Write this run
i_str = f"insert into run_table(design_index,end_day,mw_folder) values (?,?,?)"
vals = (design_index, -1, run_ident)
c.execute(i_str, vals)
_run_index = c.lastrowid
Console.print("This is run: " + str(_run_index))
database.commit()
finally:
database.close()
def read_time_file(filename):
global seed_file_cache
if filename not in seed_file_cache:
with open(filename, "r") as FILE:
time_seeds = [[num for num in line.split(',')] for line in FILE]
Console.debug("Times of seeding:", variables=[time_seeds])
# convert to dates and counts
time_seeds = np.array(time_seeds)
time_seeds_trust = time_seeds[:, 1].astype(int)
time_seeds_count = time_seeds[:, 2].astype(int)
time_seeds_date = [[i.split('-')] for i in time_seeds[:, 0]]
time_seeds_date = [[int(i[0][0]),
int(i[0][1]),
int(i[0][2])] for i in time_seeds_date]
time_seeds_date = [
datetime(i[0], i[1], i[2]).date() for i in time_seeds_date
]
# store outputs
seed_file_cache[filename] = "STORED"
seed_file_cache["time_seeds_date"] = time_seeds_date
seed_file_cache["time_seeds_trust"] = time_seeds_trust
seed_file_cache["time_seeds_count"] = time_seeds_count
return seed_file_cache["time_seeds_date"], seed_file_cache[
"time_seeds_trust"], seed_file_cache["time_seeds_count"]
def print_version_string():
from metawards.utils import Console
Console.panel(get_version_string(),
markdown=True,
style="header",
width=72,
expand=False)
def go_stage_test(infections, **kwargs):
old_num_in_red = sum(infections.subinfs[0].work[2]) + \
sum(infections.subinfs[0].play[2])
old_num_in_blue = sum(infections.subinfs[1].work[1]) + \
sum(infections.subinfs[1].play[1])
go_stage(go_from="red",
go_to="blue",
from_stage=2,
to_stage=1,
infections=infections,
**kwargs)
new_num_in_red = sum(infections.subinfs[0].work[2]) + \
sum(infections.subinfs[0].play[2])
new_num_in_blue = sum(infections.subinfs[1].work[1]) + \
sum(infections.subinfs[1].play[1])
from metawards.utils import Console
Console.print(f"{old_num_in_blue} -> {new_num_in_blue}, "
f"{old_num_in_red} -> {new_num_in_red}")
assert new_num_in_red == 0
assert old_num_in_red - new_num_in_red == new_num_in_blue - old_num_in_blue
def output_db(population: Population, network: Networks,
workspace: Workspace, output_dir: OutputFiles, **kwargs):
Console.print(f"Calling output_db for a {network.__class__} object")
# open a database to hold the data - call the 'create_tables'
# function on this database when it is first opened
conn = output_dir.open_db("stages.db", initialise=create_tables(network))
conn2 = output_dir.open_db("stages2.db", initialise=create_tables_2(network))
conn3 = output_dir.open_db("stages3.db", initialise=create_tables_3(network))
c = conn.cursor()
c2 = conn2.cursor()
c3 = conn3.cursor()
# get each demographics data
for i, subnet in enumerate(network.subnets):
name = subnet.name
ward_inf_tot = workspace.subspaces[i].ward_inf_tot
N_INF_CLASSES = workspace.subspaces[i].n_inf_classes
col_names = ["day", "ward"] + [f"{subnet.name}_{i}" for i in range(0, N_INF_CLASSES)]
col2_names = ["day", "ward"] + [f"{subnet.name}_{i}" for i in _out_channels[subnet.name]]
for k in range(1, workspace.subspaces[i].nnodes+1):
vals = [population.day, k]
for j in range(0, N_INF_CLASSES):
# TODO: What is this?? Why are some classes deltas?
if j == 1 or j == 3:
vals.append(ward_inf_tot[j - 1][k] + ward_inf_tot[j][k])
else:
vals.append(ward_inf_tot[j][k])
vals_str = ",".join([str(v) for v in vals])
# Technically this is open to SQL injection, perhaps let CW know?
c.execute(f"insert into {name}_totals VALUES ({vals_str})")
col_str = ','.join(col_names)
update_cols = col_names[2:]
update_str = ','.join([f"{c} = {v}" for c, v in zip(update_cols, vals[2:])])
qstring = f"insert into results ({col_str}) values ({vals_str}) on conflict(day, ward) do update set {update_str}"
c2.execute(qstring)
col2_str = ','.join(col2_names)
#Console.print(col2_names)
update_cols = col2_names[2:]
#Console.print(update_cols)
keeps = [vals[x + 2] for x in range(N_INF_CLASSES) if x in _out_channels[subnet.name]]
keeps_str = ",".join([str(v) for v in [population.day, k] + keeps])
#Console.print(keeps_str)
update_str = ','.join([f"{c} = {v}" for c, v in zip(update_cols, keeps)])
#Console.print(update_str)
qstring = f"insert into compact ({col2_str}) values ({keeps_str}) on conflict(day, ward) do update set {update_str}"
#Console.print(f"SQL: {qstring}")
c3.execute(qstring)
conn.commit()
conn2.commit()
conn3.commit()
def output_skip(workspace, **kwargs):
totals = [
inf + pinf for inf, pinf in zip(workspace.inf_tot, workspace.pinf_tot)
]
Console.print(str(totals))
# should have moved from stage[1] to stage[3], so totals[2] is 0
assert totals[2] == 0
def extractor_setup(network: metawards.Network, **kwargs):
# Globals
global _created_file_flag
global _sql_file_name
global _run_ident
# Get the unique output directory for this run
out_object: metawards.OutputFiles = kwargs["output_dir"]
run_ident = os.path.basename(out_object.get_path())
out_folder = os.path.abspath(
os.path.join(out_object.get_path(), os.path.pardir))
# Identify the design point
design_index = int(network.params.user_params["design_index"])
# Create a valid URI to use the SQLite access rights
data_base_file_name: str = os.path.join(out_folder, "sql_wards.dat")
_sql_file_name = data_base_file_name
fixed_path = os.path.abspath(data_base_file_name).replace("\\", "/")
db_uri = f"file:{fixed_path}?mode=rw"
# Concurrency guard: multiple processes can connect, but only one should create
# NOTE: Connections only fail if the file doesn't exist with mode "rw"
test_connection: Union[sql.Connection, None] = None
create_connection: Union[sql.Connection, None] = None
# One at a time here
mutex = multiprocessing.Lock()
with mutex:
try:
test_connection = sql.connect(db_uri, uri=True)
# TODO: What do we do if there is an old database in there?
# FIXME: For re-running partial ensembles this needs to be handled
except sql.OperationalError:
# This process is the first one in to create the database
try:
# Append "c" to make the connection create a blank database
create_connection = sql.connect(db_uri + "c", uri=True)
_created_file_flag = True
except sql.OperationalError as err:
# An actual problem happened if we get here
Console.print("SQL Error: " + str(err))
raise
# Drop repeats
u_vars = get_input_header_names()[:-1]
design_vars = [var[1:] for var in u_vars if var[0] == '.']
make_sql_template(design_vars, create_connection)
finally:
if test_connection:
test_connection.close()
if create_connection:
create_connection.close()
# Do first commits for setting up the run
write_setup_entries(network, design_index, run_ident, **kwargs)
def initialise(conn, network):
c = conn.cursor()
values: List[str] = []
for i, subnet in enumerate(network.subnets):
values += [f"{subnet.name}_{i} int" for i in _out_channels[subnet.name]]
qstring = f"CREATE TABLE IF NOT EXISTS {table_name}(design INT NOT NULL, repeat INT NOT NULL, " \
f"day INT NOT NULL, ward INT NOT NULL, {','.join(values)}, " \
f"PRIMARY KEY (design, repeat, day, ward));"
Console.print(f"POSTGRESQL Exec: \n{qstring}")
c.execute(qstring)
conn.commit()
def iterate_custom(network, population, **kwargs):
from metawards.iterators import iterate_working_week
state, rate, can_work = get_lock_down_vars(network=network, population=population)
Console.print(f"state {state}: scale_rate = {rate}, can_work = {can_work}")
if state > 0:
return [advance_lock_down]
else:
return iterate_working_week(network=network, population=population, **kwargs)
def write_setup_entries(database: Connection, design_index: int, run_ident: str):
global _run_index
c: Cursor = database.cursor()
# Write this run
i_str = f"insert into run_table(design_index,end_day,mw_folder) values (?,?,?)"
vals = (design_index, -1, run_ident)
c.execute(i_str, vals)
_run_index = c.lastrowid
Console.print("This is run: " + str(_run_index))
database.commit()
def iterate_lockdown(network, population, **kwargs):
from metawards.iterators import iterate_default
advance_funcs = iterate_default(network=network,
population=population,
**kwargs)
# read parameters from custom user parameters
params = network.params
nstages = int(params.user_params["nstages"])
start_day = int(params.user_params["start_day"])
R0 = params.user_params["R0"]
duration = params.user_params["duration"]
scale_uv = params.user_params["scale_uv"]
cutoff = params.user_params["cutoff"]
day = population.day
if day >= start_day:
# in lockdown
# which stage are we in?
stage = 0
scl = scale_uv[0] / R0
cut = cutoff[0]
end_day = start_day + int(duration[0])
if day <= end_day:
stage = 0
else:
for i in range(1, nstages):
end_day += int(duration[i])
scl = scale_uv[i] / R0
cut = cutoff[i]
if day <= end_day:
stage = i
break
if day > end_day:
print("Lockdown ended")
scl = 1.0
cut = 1000 # 1000 km
population.scale_uv = scl
params.dyn_dist_cutoff = cut
Console.debug("LOCKDOWN", variables=[stage, scl, cut])
return advance_funcs
def output_wards_serial(network: metawards.Network,
population: metawards.Population,
workspace: metawards.Workspace,
out_dir: metawards.OutputFiles, **kwargs):
global _sql_file_name
global _run_index
connection = out_dir.open_db("rundata.dat",
auto_bzip=False,
initialise=None)
# Potential problem: what if we accidentally hit a real attribute?
if not hasattr(network.params, "_uq4covid_setup"):
network.params._uq4covid_setup = True
extractor_setup(network, **kwargs)
# Prepare database entries
database = connect(_sql_file_name)
c: Cursor = database.cursor()
# Write the current day, some may be longer / shorter, so ignore duplicate entries
c.execute(f"insert or ignore into day_table(day,date) values (?,?)",
(int(population.day), str(population.date)))
database.commit()
# Write results for infections and removed
# NOTE: Don't re-use time_index as if there is a duplicate then the rowid will be zero
# TODO: List comprehension is fine, but consider numpy (or equivalent) for more speed
output_src = 0
mode_str = "normal write"
# The index has already been sent to the database, so reuse it here safely
for channel_index, channel_name in enumerate(_output_channels_list.keys()):
Console.print(
f"Writing channel {channel_name} to database src = "
f"{_output_channels_list[channel_name][output_src]}: {mode_str}")
data_source = eval(_output_channels_list[channel_name][output_src])
values = [(i, channel_index, x, int(population.day), _run_index)
for i, x in enumerate(data_source) if i != 0]
c.executemany(
f"insert into results_table(ward_id,output_channel,sim_out,sim_time,run_id) "
f"values (?,?,?,?,?)", values)
# Write last day into run table
c.execute("update run_table set end_day = ? where run_index = ?",
(int(population.day), _run_index))
database.commit()
database.close()
def read_file(filename, nage):
global file_cache
if filename not in file_cache:
with open(filename, "r") as FILE:
contact_matrix = [[num for num in line.split(',')]
for line in FILE]
# set up interaction matrix
matrix = [[0.0 for i in range(nage)] for j in range(nage)]
for i in range(nage):
for j in range(nage):
matrix[i][j] = float(contact_matrix[i][j])
Console.debug("Interaction matrix", variables=[matrix])
file_cache[filename] = matrix
return file_cache[filename]
def output_db(population: Population, network: Networks,
workspace: Workspace, output_dir: OutputFiles, **kwargs):
Console.print(f"Calling output_db for a {network.__class__} object")
postgres_connection = psg.connect(host="localhost", user="******", password="******", database=database_name)
# Make sure init is only called once, it actually doesn't matter, but saves time
if not hasattr(network, "init_guard"):
initialise(postgres_connection, network)
setattr(network, "init_guard", False)
cur = postgres_connection.cursor()
run_ident = path.basename(output_dir.get_path())
repeat_ident = int(run_ident[-3:])
design_index = int(network.params.user_params["design_index"])
# get each demographics data
for i, subnet in enumerate(network.subnets):
ward_inf_tot = workspace.subspaces[i].ward_inf_tot
N_INF_CLASSES = workspace.subspaces[i].n_inf_classes
col2_names = ["design", "repeat", "day", "ward"] + \
[f"{subnet.name}_{i}" for i in _out_channels[subnet.name]]
for k in range(1, workspace.subspaces[i].nnodes + 1):
if k not in _zero_crossings:
_zero_crossings[k] = False
vals = [design_index, repeat_ident, population.day, k]
for j in range(0, N_INF_CLASSES):
# Try to fudge a marker for first infections
if ward_inf_tot[0][k] != 0 and _zero_crossings[k] is False:
_zero_crossings[k] = True
Console.print(f"Got first infection in ward {k}")
# TODO: What is this?? Why are some classes deltas?
if j == 1 or j == 3:
vals.append(ward_inf_tot[j - 1][k] + ward_inf_tot[j][k])
else:
vals.append(ward_inf_tot[j][k])
col2_str = ','.join(col2_names)
update_cols = col2_names[4:]
keeps = [vals[x + 4] for x in range(N_INF_CLASSES) if x in _out_channels[subnet.name]]
keeps_str = ",".join([str(v) for v in [design_index, repeat_ident, population.day, k] + keeps])
update_str = ','.join([f"{c} = {v}" for c, v in zip(update_cols, keeps)])
qstring = f"INSERT INTO {table_name} ({col2_str}) VALUES ({keeps_str}) " \
f"ON CONFLICT (design, repeat, day, ward) DO UPDATE SET {update_str};"
if _zero_crossings[k] is True:
cur.execute(qstring)
postgres_connection.commit()
postgres_connection.close()
def test_progress_error():
with pytest.raises(KeyError):
with Console.progress() as progress:
task = progress.add_task("Something", total=10)
for i in range(1, 11):
progress.update(task, completed=i)
raise KeyError()
def test_skip_stage(nthreads=1, prompt=None):
seed = None
Console.set_debugging_enabled(True)
# load all of the parameters
try:
params = Parameters.load(parameters="march29")
except Exception as e:
print(f"Unable to load parameter files. Make sure that you have "
f"cloned the MetaWardsData repository and have set the "
f"environment variable METAWARDSDATA to point to the "
f"local directory containing the repository, e.g. the "
f"default is $HOME/GitHub/MetaWardsData")
raise e
# load the disease and starting-point input files
params.set_disease(os.path.join(script_dir, "data", "ncov.json"))
params.set_input_files("2011Data")
params.add_seeds("ExtraSeedsBrighton.dat")
# the size of the starting population
population = Population(initial=57104043)
nsteps = 30
print("Building the network...")
network = Network.build(params=params)
outdir = os.path.join(script_dir, "test_go_to")
with OutputFiles(outdir, force_empty=True, prompt=prompt) as output_dir:
trajectory = network.copy().run(population=population,
seed=seed,
output_dir=output_dir,
nsteps=nsteps,
mixer=mix_evenly,
mover=move_skip,
extractor=extract_skip,
nthreads=nthreads)
OutputFiles.remove(outdir, prompt=None)
# red demographic was seeded, but all moved to blue, so should have
# no outbreak
pop = trajectory[-1]
print(pop)
def output_db(population: Population, network: Networks, workspace: Workspace,
output_dir: OutputFiles, **kwargs):
Console.print(f"Calling output_db for a {network.__class__} object")
conn3 = output_dir.open_db("stages3.db",
initialise=create_tables_3(network))
c3 = conn3.cursor()
# get each demographics data
for i, subnet in enumerate(network.subnets):
ward_inf_tot = workspace.subspaces[i].ward_inf_tot
N_INF_CLASSES = workspace.subspaces[i].n_inf_classes
col2_names = ["day", "ward"] + [
f"{subnet.name}_{i}" for i in _out_channels[subnet.name]
]
for k in range(1, workspace.subspaces[i].nnodes + 1):
if k not in _zero_crossings:
_zero_crossings[k] = False
vals = [population.day, k]
for j in range(0, N_INF_CLASSES):
# Try to fudge a marker for first infections
if ward_inf_tot[0][k] != 0 and _zero_crossings[k] is False:
_zero_crossings[k] = True
Console.print(f"Got first infection in ward {k}")
# TODO: What is this?? Why are some classes deltas?
if j == 1 or j == 3:
vals.append(ward_inf_tot[j - 1][k] + ward_inf_tot[j][k])
else:
vals.append(ward_inf_tot[j][k])
col2_str = ','.join(col2_names)
update_cols = col2_names[2:]
keeps = [
vals[x + 2] for x in range(N_INF_CLASSES)
if x in _out_channels[subnet.name]
]
keeps_str = ",".join([str(v) for v in [population.day, k] + keeps])
update_str = ','.join(
[f"{c} = {v}" for c, v in zip(update_cols, keeps)])
qstring = f"insert into compact ({col2_str}) values ({keeps_str}) " \
f"on conflict(day, ward) do update set {update_str}"
if _zero_crossings[k] is True:
c3.execute(qstring)
conn3.commit()
def read_age_file(filename):
global seed_file_cache
if filename not in seed_file_cache:
with open(filename, "r") as FILE:
age_probs = [[num for num in line.split(',')] for line in FILE]
Console.debug("Age-level seeding probabilities:",
variables=[age_probs])
# convert to correct format
age_probs = np.array(age_probs)
age_probs_ind = age_probs[:, 0].astype(int)
age_probs = age_probs[:, 1].astype(float)
# save in cache
seed_file_cache[filename] = "STORED"
seed_file_cache["age_probs_ind"] = age_probs_ind
seed_file_cache["age_probs"] = age_probs
return seed_file_cache["age_probs_ind"], seed_file_cache["age_probs"]
def read_seed_file(filename):
global seed_file_cache
if filename not in seed_file_cache:
with open(filename, "r") as FILE:
ward_probs = [[num for num in line.split(',')] for line in FILE]
Console.debug("Ward-level seeding probabilities:",
variables=[ward_probs])
# convert to correct format
ward_probs = np.array(ward_probs)
ward_probs_ind = ward_probs[:, 0].astype(int)
ward_probs_trust = ward_probs[:, 1].astype(int)
ward_probs = ward_probs[:, 2].astype(float)
# save in cache
seed_file_cache[filename] = "STORED"
seed_file_cache["ward_probs_ind"] = ward_probs_ind
seed_file_cache["ward_probs_trust"] = ward_probs_trust
seed_file_cache["ward_probs"] = ward_probs
return seed_file_cache["ward_probs_ind"], seed_file_cache[
"ward_probs_trust"], seed_file_cache["ward_probs"]
def output_wards_ir_serial(network: metawards.Network,
population: metawards.Population,
workspace: metawards.Workspace, **kwargs):
global _sql_file_name
global _run_index
# Potential problem: what if we accidentally hit a real attribute?
if not hasattr(network.params, "_uq4covid_setup"):
network.params._uq4covid_setup = True
Console.print("First output")
extractor_setup(network, **kwargs)
# Prepare database entries
database = sql.connect(_sql_file_name)
c: sql.Cursor = database.cursor()
# Write the current day, some may be longer / shorter, so ignore duplicate entries
c.execute(f"insert or ignore into day_table(day,date) values (?,?)",
(int(population.day), str(population.date)))
time_index = c.lastrowid
database.commit()
# Write results for infections and removed
v = []
for i, x in enumerate(workspace.I_in_wards):
if i == 0:
continue
v.append((i, _output_channels["infected"], x, time_index, _run_index))
for i, x in enumerate(workspace.R_in_wards):
if i == 0:
continue
v.append((i, _output_channels["removed"], x, time_index, _run_index))
c.executemany(
f"insert into results_table(ward_id,output_channel,sim_out,sim_time,run_id) values (?,?,?,?,?)",
v)
database.commit()
database.close()
def test_output():
try:
params = Parameters.load(parameters="march29")
except Exception as e:
print(f"Unable to load parameter files. Make sure that you have "
f"cloned the MetaWardsData repository and have set the "
f"environment variable METAWARDSDATA to point to the "
f"local directory containing the repository, e.g. the "
f"default is $HOME/GitHub/MetaWardsData")
raise e
# load the disease and starting-point input files
params.set_disease("ncov")
params.set_input_files("2011Data")
params.add_seeds("ExtraSeedsBrighton.dat")
# the size of the starting population
population = Population(initial=57104043)
print("Building the network...")
network = Network.build(params=params)
from metawards.iterators import build_custom_iterator
iterator = build_custom_iterator(iterate_debug, __name__)
print("Run the model...")
outdir = os.path.join(script_dir, "test_output")
with OutputFiles(outdir, force_empty=True, prompt=None) as output_dir:
with Console.redirect_output(outdir, auto_bzip=True):
trajectory = network.run(population=population,
output_dir=output_dir,
iterator=iterator,
nthreads=1)
OutputFiles.remove(outdir, prompt=None)
print(trajectory[-1])
print("End of the run")
def get_frames(self, width: int = 80):
"""Return the frames used to animate a spinner in a console
of specified width
This returns the list of frames plus the timeout between
the list
"""
if width in self.frames:
return self.frames[width]
frames = []
frames.append("")
if Console.supports_emojis():
bar = "👉 👉 👉 😷 😷 😷 👌 👍 👏 👏 👏 👏 👏 "
else:
bar = "-> -> -> #WearAMask :-) :-) :-) "
for i in range(1, len(bar), 1):
frames.append(bar[0:i])
self.frames[width] = (frames, 50)
return self.frames[width]
def extract(network: metawards.Network,
**kwargs) -> List[metawards.utils.MetaFunction]:
Console.print(f"Sending I and R per ward to the output stream")
return [output_wards_i]
def test_integration_pox(prompt=None):
"""This test repeats main_RepeatsNcov.c and validates that the
various stages report the same results as the original C code
for the POX
"""
# user input parameters
seed = 15324
inputfile = ncovparams_csv
line_num = 0
UV = 1.0
# load all of the parameters
try:
params = Parameters.load(parameters="march29")
except Exception as e:
print(f"Unable to load parameter files. Make sure that you have "
f"cloned the MetaWardsData repository and have set the "
f"environment variable METAWARDSDATA to point to the "
f"local directory containing the repository, e.g. the "
f"default is $HOME/GitHub/MetaWardsData")
raise e
# load the disease and starting-point input files
params.set_disease("pox")
params.set_input_files("2011Data")
params.add_seeds("ExtraSeedsBrighton.dat")
# start from the parameters in the specified line number of the
# provided input file
variables = params.read_variables(inputfile, line_num)
# extra parameters that are set
params.UV = UV
params.static_play_at_home = 0
params.play_to_work = 0
params.work_to_play = 0
params.daily_imports = 0.0
# the size of the starting population
population = Population(initial=57104043)
profiler = Profiler()
print("Building the network...")
network = Network.build(params=params,
profiler=profiler)
params = params.set_variables(variables[0])
network.update(params, profiler=profiler)
outdir = os.path.join(script_dir, "test_integration_output")
with OutputFiles(outdir, force_empty=True, prompt=prompt) as output_dir:
print("Run the model...")
trajectory = network.run(population=population, seed=seed,
output_dir=output_dir,
nsteps=31, profiler=profiler,
nthreads=1)
OutputFiles.remove(outdir, prompt=None)
print("End of the run")
Console.print_profiler(profiler)
Console.rule("Model output")
Console.print_population(trajectory[-1])
# The original C code has this expected population after 47 steps
expected = Population(initial=57104043,
susceptibles=56080780,
latent=374,
total=370,
recovereds=553,
n_inf_wards=289,
day=31)
Console.rule("Expected output")
Console.print_population(expected)
assert trajectory[-1] == expected
def cli():
"""Main function for the command line interface. This does one of three
things:
1. If this is the main process, then it parses the arguments and
runs and manages the jobs
2. If this is a worker process, then it starts up and waits for work
3. If this is a supervisor process, then it query the job scheduling
system for information about the compute nodes to use, and will then
set up and run a manager (main) process that will use those
nodes to run the jobs
"""
from metawards.utils import Console
# get the parallel scheme now before we import any other modules
# so that it is clear if mpi4py or scoop (or another parallel module)
# has been imported via the required "-m module" syntax
parallel_scheme = get_parallel_scheme()
if parallel_scheme == "mpi4py":
from mpi4py import MPI
comm = MPI.COMM_WORLD
nprocs = comm.Get_size()
rank = comm.Get_rank()
if rank != 0:
# this is a worker process, so should not do anything
# more until it is given work in the pool
Console.print(f"Starting worker process {rank+1} of {nprocs-1}...")
return
else:
Console.print("Starting main process...")
elif parallel_scheme == "scoop":
Console.print("STARTING SCOOP PROCESS")
import sys
args, parser = parse_args()
if not args.already_supervised:
hostfile = get_hostfile(args)
if hostfile:
# The user has asked to run a parallel job - this means that this
# process is the parallel supervisor
if args.mpi:
mpi_supervisor(hostfile, args)
return
elif args.scoop:
scoop_supervisor(hostfile, args)
return
# neither is preferred - if scoop is installed then use that
try:
import scoop # noqa - disable unused warning
have_scoop = True
except Exception:
have_scoop = False
if have_scoop:
scoop_supervisor(hostfile, args)
return
# do we have MPI?
try:
import mpi4py # noqa - disable unused warning
have_mpi4py = True
except Exception:
have_mpi4py = False
if have_mpi4py:
mpi_supervisor(hostfile, args)
return
# we don't have any other option, just keep going and
# use multiprocessing - in this case we don't need a
# supervisor and this is the main process
# This is now the code for the main process
# WE NEED ONE OF these listed options;
should_run = False
for arg in [
args.input, args.repeats, args.disease, args.additional,
args.model, args.iterator, args.extractor, args.demographics,
args.mixer, args.mover
]:
if arg is not None:
should_run = True
break
if not should_run:
parser.print_help(sys.stdout)
sys.exit(0)
if args.repeats is None:
args.repeats = [1]
# import the parameters here to speed up the display of help
from metawards import Parameters, Network, Population, print_version_string
# print the version information first, so that there is enough
# information to enable someone to reproduce this run
print_version_string()
Console.rule("Initialise")
if args.input:
# get the line numbers of the input file to read
if args.line is None or len(args.line) == 0:
linenums = None
Console.print(f"* Using parameters from all lines of {args.input}",
markdown=True)
else:
from metawards.utils import string_to_ints
linenums = string_to_ints(args.line)
if len(linenums) == 0:
Console.error(f"You cannot read no lines from {args.input}?")
sys.exit(-1)
elif len(linenums) == 1:
Console.print(
f"* Using parameters from line {linenums[0]} of "
f"{args.input}",
markdown=True)
else:
Console.print(
f"* Using parameters from lines {linenums} of "
f"{args.input}",
markdown=True)
from metawards import VariableSets, VariableSet
variables = VariableSets.read(filename=args.input,
line_numbers=linenums)
else:
from metawards import VariableSets, VariableSet
# create a VariableSets with one null VariableSet
variables = VariableSets()
variables.append(VariableSet())
nrepeats = args.repeats
if nrepeats is None or len(nrepeats) < 1:
nrepeats = [1]
if len(nrepeats) > 1 and len(variables) != len(nrepeats):
Console.error(f"The number of repeats {len(nrepeats)} must equal the "
f"number of adjustable variable lines {len(variables)}")
raise ValueError("Disagreement in the number of repeats and "
"adjustable variables")
# ensure that all repeats are >= 0
nrepeats = [0 if int(x) < 0 else int(x) for x in nrepeats]
if sum(nrepeats) == 0:
Console.error(f"The number of the number of repeats is 0. Are you "
f"sure that you don't want to run anything?")
raise ValueError("Cannot run nothing")
if len(nrepeats) == 1 and nrepeats[0] == 1:
Console.print("* Performing a single run of each set of parameters",
markdown=True)
elif len(nrepeats) == 1:
Console.print(
f"* Performing {nrepeats[0]} runs of each set of parameters",
markdown=True)
else:
Console.print(
f"* Performing {nrepeats} runs applied to the parameters",
markdown=True)
variables = variables.repeat(nrepeats)
# working out the number of processes and threads...
from metawards.utils import guess_num_threads_and_procs
(nthreads,
nprocs) = guess_num_threads_and_procs(njobs=len(variables),
nthreads=args.nthreads,
nprocs=args.nprocs,
parallel_scheme=parallel_scheme)
Console.print(
f"\n* Number of threads to use for each model run is {nthreads}",
markdown=True)
if nprocs > 1:
Console.print(
f"* Number of processes used to parallelise model "
f"runs is {nprocs}",
markdown=True)
Console.print(
f"* Parallelisation will be achieved using {parallel_scheme}",
markdown=True)
# sort out the random number seed
seed = args.seed
if seed is None:
import random
seed = random.randint(10000, 99999999)
if seed == 0:
# this is a special mode that a developer can use to force
# all jobs to use the same random number seed (15324) that
# is used for comparing outputs. This should NEVER be used
# for production code
Console.warning("Using special mode to fix all random number"
"seeds to 15324. DO NOT USE IN PRODUCTION!!!")
else:
Console.print(f"* Using random number seed {seed}", markdown=True)
# get the starting day and date
start_day = args.start_day
if start_day < 0:
raise ValueError(f"You cannot use a start day {start_day} that is "
f"less than zero!")
start_date = None
if args.start_date:
try:
from dateparser import parse
start_date = parse(args.start_date).date()
except Exception:
pass
if start_date is None:
from datetime import date
try:
start_date = date.fromisoformat(args.start_date)
except Exception as e:
raise ValueError(f"Cannot interpret a valid date from "
f"'{args.start_date}'. Error is "
f"{e.__class__} {e}")
if start_date is None:
from datetime import date
start_date = date.today()
Console.print(f"* Day zero is {start_date.strftime('%A %B %d %Y')}",
markdown=True)
if start_day != 0:
from datetime import timedelta
start_day_date = start_date + timedelta(days=start_day)
Console.print(f"Starting on day {start_day}, which is "
f"{start_day_date.strftime('%A %B %d %Y')}")
else:
start_day_date = start_date
# now find the MetaWardsData repository as this will be needed
# for the repeat command line too
(repository,
repository_version) = Parameters.get_repository(args.repository)
Console.print(f"* Using MetaWardsData at {repository}", markdown=True)
if repository_version["is_dirty"]:
Console.warning("This repository is dirty, meaning that the data"
"has not been committed to git. This may make "
"this calculation very difficult to reproduce")
# now work out the minimum command line needed to repeat this job
args.seed = seed
args.nprocs = nprocs
args.nthreads = nthreads
args.start_date = start_date.isoformat()
args.repository = repository
# also print the source of all inputs
import configargparse
Console.rule("Souce of inputs")
p = configargparse.get_argument_parser("main")
Console.print(p.format_values())
# print out the command used to repeat this job
repeat_cmd = "metawards"
for key, value in vars(args).items():
if value is not None:
k = key.replace("_", "-")
if isinstance(value, bool):
if value:
repeat_cmd += f" --{k}"
elif isinstance(value, list):
repeat_cmd += f" --{k}"
for val in value:
v = str(val)
if " " in v:
repeat_cmd += f" '{v}''"
else:
repeat_cmd += f" {v}"
else:
v = str(value)
if " " in v:
repeat_cmd += f" --{k} '{v}''"
else:
repeat_cmd += f" --{k} {v}"
Console.rule("Repeating this run")
Console.print("To repeat this job use the command;")
Console.command(repeat_cmd)
Console.print("Or alternatively use the config.yaml file that will be "
"written to the output directory and use the command;")
Console.command("metawards -c config.yaml")
# load all of the parameters
try:
params = Parameters.load(parameters=args.parameters)
except Exception as e:
Console.warning(
f"Unable to load parameter files. Make sure that you have "
f"cloned the MetaWardsData repository and have set the "
f"environment variable METAWARDSDATA to point to the "
f"local directory containing the repository, e.g. the "
f"default is $HOME/GitHub/MetaWardsData")
raise e
# should we profile the code? (default no as it prints a lot)
profiler = None
if args.no_profile:
profiler = None
elif args.profile:
from metawards.utils import Profiler
profiler = Profiler()
# load the disease and starting-point input files
Console.rule("Disease")
if args.disease:
params.set_disease(args.disease)
else:
params.set_disease("ncov")
Console.rule("Model data")
if args.model:
params.set_input_files(args.model)
else:
params.set_input_files("2011Data")
# load the user-defined custom parameters
Console.rule("Custom parameters and seeds")
if args.user_variables:
custom = VariableSet.read(args.user_variables)
Console.print(f"Adjusting variables to {custom}")
custom.adjust(params)
else:
Console.print("Not adjusting any parameters...")
# read the additional seeds
if args.additional is None or len(args.additional) == 0:
Console.print("Not using any additional seeds...")
else:
for additional in args.additional:
Console.print(f"Loading additional seeds from {additional}")
params.add_seeds(additional)
# what to do with the 0 state?
stage_0 = "R"
if args.disable_star:
Console.print("Disabling the * state. Stage 0 is the one and "
"only E state.")
stage_0 = "disable"
elif args.star_is_E:
Console.print("Setting the * state as an additional E state.")
stage_0 = "E"
else:
Console.print("Setting the * state as an additional R state.")
stage_0 = "R"
params.stage_0 = stage_0
# extra parameters that are set
params.UV = args.UV
# set these extra parameters to 0
params.static_play_at_home = 0
params.play_to_work = 0
params.work_to_play = 0
params.daily_imports = 0.0
Console.rule("Parameters")
Console.print(params, markdown=True)
# the size of the starting population
population = Population(initial=args.population,
date=start_day_date,
day=start_day)
Console.rule("Building the network")
network = Network.build(params=params,
population=population,
max_nodes=args.max_nodes,
max_links=args.max_links,
profiler=profiler)
if args.demographics:
from metawards import Demographics
Console.rule("Specialising into demographics")
demographics = Demographics.load(args.demographics)
Console.print(demographics)
network = network.specialise(demographics,
profiler=profiler,
nthreads=nthreads)
Console.rule("Preparing to run")
from metawards import OutputFiles
from metawards.utils import run_models
outdir = args.output
if outdir is None:
outdir = "output"
if args.force_overwrite_output:
prompt = None
else:
from metawards import input
def prompt(x):
return input(x, default="y")
auto_bzip = True
if args.auto_bzip:
auto_bzip = True
elif args.no_auto_bzip:
auto_bzip = False
if args.iterator:
iterator = args.iterator
else:
iterator = None
if args.extractor:
extractor = args.extractor
else:
extractor = None
if args.mixer:
mixer = args.mixer
else:
mixer = None
if args.mover:
mover = args.mover
else:
mover = None
with OutputFiles(outdir,
force_empty=args.force_overwrite_output,
auto_bzip=auto_bzip,
prompt=prompt) as output_dir:
# write the config file for this job to output/config.yaml
Console.rule("Running the model")
CONSOLE = output_dir.open("console.log")
Console.save(CONSOLE)
lines = []
max_keysize = None
for key, value in vars(args).items():
if max_keysize is None:
max_keysize = len(key)
elif len(key) > max_keysize:
max_keysize = len(key)
for key, value in vars(args).items():
if value is not None:
key = key.replace("_", "-")
spaces = " " * (max_keysize - len(key))
if isinstance(value, bool):
if value:
lines.append(f"{key}:{spaces} true")
else:
lines.append(f"{key}:{spaces} false")
elif isinstance(value, list):
s_value = [str(x) for x in value]
lines.append(f"{key}:{spaces} [ {', '.join(s_value)} ]")
else:
lines.append(f"{key}:{spaces} {value}")
CONFIG = output_dir.open("config.yaml", auto_bzip=False)
lines.sort(key=str.swapcase)
CONFIG.write("\n".join(lines))
CONFIG.write("\n")
CONFIG.flush()
CONFIG.close()
lines = None
result = run_models(network=network,
variables=variables,
population=population,
nprocs=nprocs,
nthreads=nthreads,
seed=seed,
nsteps=args.nsteps,
output_dir=output_dir,
iterator=iterator,
extractor=extractor,
mixer=mixer,
mover=mover,
profiler=profiler,
parallel_scheme=parallel_scheme)
if result is None or len(result) == 0:
Console.print("No output - end of run")
return 0
Console.rule("End of the run", style="finish")
Console.save(CONSOLE)
return 0
def mpi_supervisor(hostfile, args):
"""Function used by the MPI supervisor to get the information needed to
form the mpiexec call to run an MPI version of the program
"""
import os
import stat
import sys
from metawards.utils import Console
Console.print("RUNNING AN MPI PROGRAM")
cores_per_node = get_cores_per_node(args)
Console.print(
f"Will run jobs assuming {cores_per_node} cores per compute node")
# based on the number of threads requested and the number of cores
# per node, we can work out the number of mpi processes to start,
# and can write a hostfile that will create the right layout
nthreads = get_threads_per_task(args)
Console.print(f"Will use {nthreads} OpenMP threads per model run...")
tasks_per_node = int(cores_per_node / nthreads)
Console.print(f"...meaning that the number of model runs per node will be "
f"{tasks_per_node}")
# Next, read the hostfile to get a unique list of hostnames
hostnames = {}
with open(hostfile, "r") as FILE:
line = FILE.readline()
while line:
hostname = line.strip()
if len(hostname) > 0:
hostnames[hostname] = 1
line = FILE.readline()
hostnames = list(hostnames.keys())
hostnames.sort()
Console.print(f"Number of compute nodes equals {len(hostnames)}")
Console.print(", ".join(hostnames))
# how many tasks can we perform in parallel?
nprocs = tasks_per_node * len(hostnames)
if args.nprocs:
if nprocs != args.nprocs:
Console.print(f"WARNING: You are using an unrecommended number of "
f"processes {args.nprocs} for the cluster {nprocs}.")
nprocs = args.nprocs
Console.print(
f"Total number of parallel processes to run will be {nprocs}")
Console.print(f"Total number of cores in use will be {nprocs*nthreads}")
# Now write a new hostfile that round-robins the MPI tasks over
# the nodes for 'tasks_per_node' runs
hostfile = f"_metawards_hostfile_{os.getpid()}"
Console.print(f"Writing hostfile to {hostfile}")
with open(hostfile, "w") as FILE:
i = 0
while i < nprocs:
for hostname in hostnames:
FILE.write(hostname + "\n")
i += 1
if i == nprocs:
break
# now craft the mpiexec command that will use this hostfile to
# run the job - remember to pass the option to stop the main process
# attempt to become a supervisor itself...
mpiexec = os.getenv("MPIEXEC")
if mpiexec is None:
mpiexec = "mpiexec"
# check for weird mpiexecs...
import subprocess
import shlex
try:
args = shlex.split(f"{mpiexec} -v")
p = subprocess.run(args,
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT)
v = p.stdout.decode("utf-8").strip()
Console.print(f"{mpiexec} -v => {v}")
if v.find("HPE HMPT") != -1:
raise ValueError(
"metawards needs a more modern MPI library than HPE's, "
"so please compile to another MPI and use that.")
except Exception as e:
Console.error(f"[ERROR] {e.__class__} {e}")
pyexe = sys.executable
script = os.path.abspath(sys.argv[0])
args = " ".join(sys.argv[1:])
cmd = f"{mpiexec} -np {nprocs} -hostfile {hostfile} " \
f"{pyexe} -m mpi4py {script} --already-supervised {args} " \
f"--nprocs {nprocs}"
Console.print("Executing MPI job using")
Console.command(cmd)
try:
args = shlex.split(cmd)
subprocess.run(args).check_returncode()
except Exception as e:
Console.error("ERROR: Something went wrong!")
Console.error(f"{e.__class__}: {e}")
sys.exit(-1)
# clean up the hostfile afterwards... (we leave it if something
# went wrong as it may help debugging)
os.unlink(hostfile)
Console.print("MPI processes completed successfully")
def scoop_supervisor(hostfile, args):
"""Function used by the scoop supervisor to get the information needed to
form the scoop call to run a scoop version of the program
"""
import os
import stat
import sys
from metawards.utils import Console
Console.print("RUNNING A SCOOP PROGRAM")
cores_per_node = get_cores_per_node(args)
Console.print(
f"Will run jobs assuming {cores_per_node} cores per compute node")
# based on the number of threads requested and the number of cores
# per node, we can work out the number of scoop processes to start,
# and can write a hostfile that will create the right layout
nthreads = get_threads_per_task(args)
Console.print(f"Will use {nthreads} OpenMP threads per model run...")
tasks_per_node = int(cores_per_node / nthreads)
Console.print(f"...meaning that the number of model runs per node will be "
f"{tasks_per_node}")
# Next, read the hostfile to get a unique list of hostnames
hostnames = {}
with open(hostfile, "r") as FILE:
line = FILE.readline()
while line:
hostname = line.strip()
if len(hostname) > 0:
hostnames[hostname] = 1
line = FILE.readline()
hostnames = list(hostnames.keys())
hostnames.sort()
Console.print(f"Number of compute nodes equals {len(hostnames)}")
Console.print(", ".join(hostnames))
# how many tasks can we perform in parallel?
nprocs = tasks_per_node * len(hostnames)
if args.nprocs:
if nprocs != args.nprocs:
Console.warning(
f"You are using a not-recommended number of "
f"processes {args.nprocs} for the cluster {nprocs}.")
nprocs = args.nprocs
Console.print(
f"Total number of parallel processes to run will be {nprocs}")
Console.print(f"Total number of cores in use will be {nprocs*nthreads}")
# Now write a new hostfile that round-robins the MPI tasks over
# the nodes for 'tasks_per_node' runs
hostfile = f"_metawards_hostfile_{os.getpid()}"
Console.print(f"Writing hostfile to {hostfile}")
with open(hostfile, "w") as FILE:
i = 0
while i < nprocs:
for hostname in hostnames:
FILE.write(hostname + "\n")
i += 1
if i == nprocs:
break
# now craft the scoop command that will use this hostfile to
# run the job - remember to pass the option to stop the main process
# attempting to become a supervisor itself...
import subprocess
import shlex
pyexe = sys.executable
script = os.path.abspath(sys.argv[0])
args = " ".join(sys.argv[1:])
# also need to tell the main program the number of processes
# as it can't work it out itself
cmd = f"{pyexe} -m scoop --hostfile {hostfile} -n {nprocs} " \
f"{script} --already-supervised {args} --nprocs {nprocs}"
Console.print("Executing scoop job using")
Console.command(cmd)
try:
args = shlex.split(cmd)
subprocess.run(args).check_returncode()
except Exception as e:
Console.error("ERROR: Something went wrong!")
Console.error(f"{e.__class__}: {e}")
sys.exit(-1)
# clean up the hostfile afterwards... (we leave it if something
# went wrong as it may help debugging)
os.unlink(hostfile)
Console.print("Scoop processes completed successfully")
return 0
Console.rule("End of the run", style="finish")
Console.save(CONSOLE)
return 0
if __name__ == "__main__":
import multiprocessing
multiprocessing.freeze_support() # needed to stop fork bombs
try:
# needed to stop OpenMP hang on Linux with libgomp
multiprocessing.set_start_method("spawn")
except Exception:
pass
method = multiprocessing.get_start_method()
if method != "spawn":
from ..utils._console import Console
error = f"We need to run with multiprocessing in 'spawn' mode, " \
f"else this will cause deadlocks with OpenMP. The mode " \
f"'{method}' is thus not supported!"
Console.error(error)
raise AssertionError(error)
cli()
def extract_only_i_per_ward(**kwargs):
Console.print("Sending I per ward to the output stream")
return [output_wards_i]
def extract_only_i_per_ward(**kwargs):
from metawards.utils import Console
Console.print("Sending I per ward to the output stream")
return [output_wards_i]
