def buildDataset(cls, all_results: Dataset, testie: Testie) -> List[tuple]:
dtype = testie.variables.dtype()
y = OrderedDict()
dataset = []
for i, (run, results_types) in enumerate(all_results.items()):
vars = list(run.variables.values())
if not results_types is None and len(results_types) > 0:
dataset.append(vars)
for result_type, results in results_types.items():
r = np.mean(results)
y.setdefault(result_type, []).append(r)
dtype['values'] = [None] * len(dtype['formats'])
for i, f in enumerate(dtype['formats']):
if f is str:
dtype['formats'][i] = int
values = OrderedSet()
for row in dataset:
values.add(row[i])
row[i] = values.index(row[i])
dtype['values'][i] = list(values)
X = np.array(dataset, ndmin=2)
lset = []
for result_type, v in y.items():
lset.append((result_type, X, np.array(v),dtype))
return lset
def writeversion(self,
testie,
all_results: Dataset,
allow_overwrite: bool = False,
kind=False,
reload=True):
if not reload and all_results:
prev = self.load_results(testie=testie, kind=kind, cache=True)
if prev and len(all_results) < len(prev):
print(
"ERROR ! Have less results than before. Forcing update write !"
)
reload = True
return
if reload:
results = self.load_results(testie=testie, kind=kind, cache=False)
if results:
results.update(all_results)
all_results = results
if kind:
for kind, kresult in all_results.items():
filename = self.__resultFilename(testie) + '-' + kind
self._writeversion(filename, kresult, allow_overwrite)
else:
filename = self.__resultFilename(testie)
self._writeversion(filename, all_results, allow_overwrite)
def writeversion(self, script, all_results: Dataset):
filename = self.__resultFilename(script)
try:
if not os.path.exists(os.path.dirname(filename)):
os.makedirs(os.path.dirname(filename))
except OSError:
print("Error : could not create %s" % os.path.dirname(filename))
f = open(filename, 'w+')
f.seek(0)
for run, results in all_results.items():
v = []
for key, val in sorted(run.variables.items()):
if type(val) is tuple:
val = val[1]
v.append(key + ":" + str(val).replace(':','\:'))
type_results = []
for t,r in results.items():
str_results = []
if r is None:
pass
else:
for val in r:
str_results.append(str(val))
type_results.append(t+':'+(','.join(str_results)))
f.write(','.join(v) + "={" + '},{'.join(type_results) + "}\n")
f.close()
def buildDataset(cls, all_results: Dataset, testie: Testie):
dtype = testie.variables.dtype()
y = []
dataset = []
for i, (run, results) in enumerate(all_results.items()):
vars = list(run.variables.values())
if not results is None:
dataset.append(vars)
y.append(np.mean(results))
dtype['formats'] = dtype['formats']
dtype['names'] = dtype['names']
for i, f in enumerate(dtype['formats']):
if f is str:
dtype['formats'][i] = int
values = OrderedSet()
for row in dataset:
values.add(row[i])
row[i] = values.index(row[i])
X = np.array(dataset, ndmin=2)
return X, np.array(y, dtype=[('result', float)])
def run(build: Build,
all_results: Dataset,
testie: Testie,
max_depth=3,
filename=None):
print("Building dataset...")
X, y = Statistics.buildDataset(all_results, testie)
print("Learning dataset built with %d samples and %d features..." %
(X.shape[0], X.shape[1]))
clf = tree.DecisionTreeRegressor(max_depth=max_depth)
clf = clf.fit(X, y)
if max_depth is None or max_depth > 8:
print("No tree graph when maxdepth is > 8")
else:
dot_data = tree.export_graphviz(
clf,
out_file=None,
filled=True,
rounded=True,
special_characters=True,
feature_names=testie.variables.dtype()['names'])
graph = pydotplus.graph_from_dot_data(dot_data)
if filename:
f = filename
else:
f = build.result_path(testie.filename, 'pdf', suffix='_clf')
graph.write(f, format=os.path.splitext(f)[1][1:])
print("Decision tree visualization written to %s" % f)
print("")
print("Feature importances :")
# noinspection PyUnresolvedReferences
for key, f in zip(testie.variables.dtype()['names'],
clf.feature_importances_):
print(" %s : %0.2f" % (key, f))
vars_values = {}
for run, results in all_results.items():
for k, v in run.variables.items():
vars_values.setdefault(k, set()).add(v)
print('')
print("Better :")
best = X[y['result'].argmax()]
print(" ", end='')
f = next(iter(all_results.items()))
for i, (k, v) in enumerate(f[0].variables.items()):
print("%s = %s, " % (k, best[i]), end='')
print(' : %.02f' % y['result'].max())
print('')
print("Means and std/mean per variables :")
for k, vals in vars_values.items():
if len(vals) is 1:
continue
print("%s :" % k)
for v in sorted(vals):
tot = 0
std = 0
n = 0
for run, results in all_results.items():
if run.variables[k] == v:
if not results is None:
tot += np.mean(results)
std += np.std(results)
n += 1
if n == 0:
print(" %s : None" % v)
else:
print(" %s : (%.02f,%.02f), " % (v, tot / n, std / n /
(tot / n)))
print("")