def test_first(self):
# lists
arr = [1, 2, 3, 4]
got = first(arr)
expected = 1
self.assertEqual(got, expected)
# iterators
arr = ['a', 'b', 'c']
it = arr.__iter__()
got = first(it)
expected = 'a'
self.assertEqual(got, expected)
def getBest(results):
best = first(results)
for r in results:
a = r.load()[0]
b = best.load()[0]
am = np.mean(a)
bm = np.mean(b)
if am > bm:
best = r
return best
def getBest(results: ResultList, reducer: Callable = np.mean):
best = first(results)
for r in results:
a = r.mean()
b = best.mean()
am = reducer(a)
bm = reducer(b)
if am > bm:
best = r
return best
def getBest(results: ResultList, steps: Optional[int] = None, percent: float = 1.0, comparator=lambda a, b: a < b):
low = first(results)
if steps is None:
steps = low.mean().shape[0]
steps = int(steps * percent)
for r in results:
a = r.mean()
b = low.mean()
am = np.mean(a[0 - steps:])
bm = np.mean(b[0 - steps:])
if np.isnan(bm) or comparator(am, bm):
low = r
return low
# ----------------
for path in experiment_paths:
print(path)
# load the experiment json file
exp = Experiment.load(path)
# load the slurm config file
slurm = Slurm.fromFile(slurm_path)
if exp.agent in SLOW_ALGS:
slurm.sequential = 1
# figure out how many indices to use
size = exp.numPermutations() * runs
paths = listResultsPaths(exp, runs)
res_path = first(paths)
data = []
data_path = f'{res_path}/returns.csv'
if os.path.exists(data_path):
f = open(data_path, 'r')
data = f.readlines()
f.close()
indices = listIndices(exp, runs)
# get all of the indices corresponding to missing results
indices = generateMissing(exp, indices, data)
indices = printProgress(size, indices)
# compute how many "tasks" to clump into each job
groupSize = slurm.cores * slurm.sequential
