def test_negative_selection_false():
from EvoDAG import EvoDAG
from EvoDAG.population import SteadyState
import numpy as np
class P(SteadyState):
def random_selection(self, negative=False):
if negative:
self._llamo = True
return np.random.randint(self.popsize)
Xt = X.copy()
y = cl.copy()
m = EvoDAG.init(seed=11,
popsize=10,
orthogonal_selection=True,
negative_selection=False,
population_class=P,
classifier=False,
early_stopping_rounds=10).fit(Xt, y)
assert not m._negative_selection
assert not m._p._negative_selection
assert m._p._llamo
m = EvoDAG.init(seed=11,
popsize=10,
orthogonal_selection=True,
negative_selection=True,
population_class=P,
classifier=False,
early_stopping_rounds=10).fit(Xt, y)
try:
m._p._llamo
except AttributeError:
return
assert False
def test_init_evodag():
from EvoDAG.model import EvoDAG
m = EvoDAG().fit(X, cl)
hy = m.predict(X)
print((cl == hy).mean(), cl, hy)
assert (cl == hy).mean() > 0.9
default_nargs()
def store_model(self, kw):
if self.data.ensemble_size == 1:
if self.data.seed >= 0:
kw['seed'] = self.data.seed
self.evo = EvoDAG(**kw).fit(self.X, self.y, test_set=self.Xtest)
self.model = self.evo.model()
else:
min_size = self.data.min_size
esize = self.data.ensemble_size
init = self.data.seed
end = init + esize
evo = []
while len(evo) < esize:
args = [(x, kw, self.X, self.y, self.Xtest)
for x in range(init, end)]
if self.data.cpu_cores == 1:
_ = [init_evodag(x) for x in tqdm(args, total=len(args))]
else:
p = Pool(self.data.cpu_cores, maxtasksperchild=1)
_ = [
x for x in tqdm(p.imap_unordered(init_evodag, args),
total=len(args))
]
p.close()
[evo.append(x) for x in _ if x.size >= min_size]
init = end
end = init + (esize - len(evo))
self.model = Ensemble(evo)
model_file = self.get_model_file()
with gzip.open(model_file, 'w') as fpt:
pickle.dump(self.model, fpt)
pickle.dump(self.word2id, fpt)
pickle.dump(self.label2id, fpt)
def test_model_hist():
from EvoDAG import EvoDAG
from EvoDAG.base import Model
y = cl.copy()
gp = EvoDAG(generations=np.inf,
multiple_outputs=True,
tournament_size=2,
early_stopping_rounds=-1,
seed=1,
popsize=30).fit(X[:-10], y[:-10], test_set=X[-10:])
hist = gp.population.hist
trace = gp.trace(gp.population.estopping)
a = hist[trace[-1]].variable
if not isinstance(a, list):
a = [a]
m = Model(trace, hist)
b = m._hist[-1].variable
if not isinstance(b, list):
b = [b]
print([(x, x.height) for x in m._hist])
print((m._map, a, b))
for v1, v2 in zip(a, b):
if v1 not in m._map:
assert v1 == v2
else:
assert m._map[v1] == v2
def test_share_inputs():
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(classifier=True,
multiple_outputs=True,
popsize=5,
share_inputs=True)
gp.fit(X, y)
assert gp._share_inputs
def test_multiple_outputs_error_rate_ts():
from EvoDAG import EvoDAG
from EvoDAG.node import Add, Min, Max
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
function_set=[Add, Min, Max],
early_stopping_rounds=100,
time_limit=0.9,
multiple_outputs=True,
fitness_function='ER',
seed=0,
popsize=100)
gp.X = X[:-1]
gp.nclasses(y[:-1])
gp.y = y[:-1]
gp.create_population()
a = gp.random_offspring()
hys = SparseArray.argmax(a.hy)
hy = np.array(hys.full_array())
# print(((hys - gp._y_klass).sign().fabs() * gp._mask_ts).sum())
mask = np.array(gp._mask_ts.full_array()).astype(np.bool)
# print((y[:-1][mask] != hy[mask]).mean())
print(-a.fitness, (y[:-1][mask] != hy[mask]).mean())
assert_almost_equals(-a.fitness, (y[:-1][mask] != hy[mask]).mean())
def rs_evodag(args_X_y):
args, X, y = args_X_y
rs = RandomParameterSearch
fit = []
init = time.time()
for seed in range(3):
evo = EvoDAG(seed=seed,
**rs.process_params(args)).fit(X, y)
fit.append(evo.model().fitness_vs)
args['_time'] = time.time() - init
gc.collect()
return fit, args
def test_min_class():
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
early_stopping_rounds=100,
time_limit=0.9,
multiple_outputs=True,
seed=0,
popsize=100)
gp.y = y[:-1]
gp.X = X[:-1]
assert gp._bagging_fitness.min_class == 2
def test_transform_to_mo():
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
early_stopping_rounds=100,
time_limit=0.9,
multiple_outputs=True,
seed=0,
popsize=10000)
gp.nclasses(y)
k = np.unique(y)
y = gp._bagging_fitness.transform_to_mo(y)
assert k.shape[0] == y.shape[1]
def test_inputs_func_argument_regression():
from EvoDAG import EvoDAG
class Error:
nargs = 2
min_nargs = 2
classification = True
regression = True
def __init__(self, *args, **kwargs):
raise RuntimeError('aqui')
y = cl.copy()
y[y == 0] = -1
y[y > -1] = 1
gp = EvoDAG(classifier=False,
multiple_outputs=False,
pr_variable=0,
input_functions=[Error],
popsize=5,
share_inputs=True)
gp.X = X
gp.nclasses(y)
gp.y = y
try:
gp.create_population()
assert False
except RuntimeError:
pass
def test_two_instances():
from EvoDAG import EvoDAG
y = cl.copy()
y[:-2] = -1
y[-2:] = 1
function_set = [x for x in EvoDAG()._function_set if x.regression and x.nargs]
gp = EvoDAG(generations=np.inf,
tournament_size=2,
classifier=False,
function_set=function_set,
early_stopping_rounds=-1,
seed=0,
popsize=10).fit(X, y, test_set=X)
assert gp
def test_classification_mo2():
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
early_stopping_rounds=10,
time_limit=0.9,
multiple_outputs=True,
all_inputs=True,
remove_raw_inputs=False,
seed=0,
popsize=10000)
gp.X = X
gp.nclasses(y)
y = gp._bagging_fitness.transform_to_mo(y)
y = [SparseArray.fromlist(x) for x in y.T]
gp = EvoDAG(generations=np.inf,
tournament_size=2,
early_stopping_rounds=10,
time_limit=0.9,
multiple_outputs=True,
all_inputs=True,
seed=0,
remove_raw_inputs=False,
popsize=10000).fit(X, y)
m = gp.model()
print([(x, x._variable, x.height) for x in m._hist])
# assert False
assert len(m.decision_function(gp.X)) == 3
def test_g_recall():
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
early_stopping_rounds=100,
time_limit=0.9,
multiple_outputs=True,
seed=0,
popsize=500)
gp.y = y
gp.X = X
gp.create_population()
off = gp.random_offspring()
hy = SparseArray.argmax(off.hy)
index = np.array(gp._mask_ts.index)
y = np.array(gp._y_klass.full_array())[index]
hy = np.array(hy.full_array())[index]
nclasses = gp._bagging_fitness.nclasses
recall = np.array([(hy[y == k] == k).mean() for k in range(nclasses)])
score = np.prod(recall) - 1
gp._fitness_function = 'g_recall'
gp._bagging_fitness.set_fitness(off)
assert_almost_equals(score, off.fitness)
index = np.array(gp._mask_ts.full_array()) == 0
y = np.array(gp._y_klass.full_array())[index]
hy = SparseArray.argmax(off.hy)
hy = np.array(hy.full_array())[index]
recall = np.array([(hy[y == k] == k).mean() for k in range(nclasses)])
score = np.prod(recall) - 1
assert_almost_equals(score, off.fitness_vs)
def test_process_params():
from EvoDAG.utils import RandomParameterSearch
from EvoDAG import EvoDAG
rs = RandomParameterSearch(npoints=1)
args = [x for x in rs][0]
evo = EvoDAG(**rs.process_params(args))
params = evo.get_params()
for k, v in args.items():
if k in params:
print(v, params[k])
if hasattr(params[k], '__name__'):
assert v == params[k].__name__
else:
assert v == params[k]
def test_multiple_outputs2():
from EvoDAG import EvoDAG
from EvoDAG.model import Model
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
early_stopping_rounds=100,
time_limit=0.9,
multiple_outputs=True,
seed=0,
popsize=10000).fit(X, y, test_set=X)
m = gp.model()
assert isinstance(m, Model)
assert len(gp.y) == 3
def rs_evodag(args_X_y):
args, X, y = args_X_y
rs = RandomParameterSearch
fit = []
init = time.time()
for seed in range(3):
try:
evo = EvoDAG(seed=seed, **rs.process_params(args)).fit(X, y)
fit.append(evo.model().fitness_vs)
except RuntimeError:
fit.append(-np.inf)
args['_time'] = time.time() - init
gc.collect()
return fit, args
def test_a_precision():
from EvoDAG.cython_utils import Score
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
early_stopping_rounds=100,
time_limit=0.9,
multiple_outputs=True,
seed=0,
popsize=500)
gp.y = y
gp.X = X
gp.create_population()
off = gp.random_offspring()
hy = SparseArray.argmax(off.hy)
index = np.array(gp._mask_ts.index)
y = np.array(gp._y_klass.full_array())[index]
hy = np.array(hy.full_array())[index]
nclasses = gp._bagging_fitness.nclasses
precision = np.array([(y[hy == k] == k).mean() for k in range(nclasses)])
f1 = Score(nclasses)
mf1, mf1_v = f1.a_precision(gp._y_klass, SparseArray.argmax(off.hy),
gp._mask_ts.index)
assert_almost_equals(np.mean(precision), mf1)
gp._fitness_function = 'a_precision'
gp._bagging_fitness.set_fitness(off)
assert_almost_equals(mf1 - 1, off.fitness)
index = np.array(gp._mask_ts.full_array()) == 0
y = np.array(gp._y_klass.full_array())[index]
hy = SparseArray.argmax(off.hy)
hy = np.array(hy.full_array())[index]
precision = np.array([(y[hy == k] == k).mean() for k in range(nclasses)])
assert_almost_equals(np.mean(precision) - 1, off.fitness_vs)
def test_multiple_outputs_predict():
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
multiple_outputs=True,
early_stopping_rounds=-1,
seed=0,
popsize=10).fit(X[:-10], y[:-10], test_set=X[-10:])
m = gp.model()
assert m.multiple_outputs
hy = m.predict(X)
u = np.unique(y)
for i in np.unique(hy):
assert i in u
def test_multiple_outputs_decision_function():
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
multiple_outputs=True,
early_stopping_rounds=-1,
seed=0,
popsize=10).fit(X[:-10], y[:-10], test_set=X[-10:])
m = gp.model()
assert m.multiple_outputs
hy = m.decision_function(X)
assert len(hy) == 3
for i in hy:
assert i.isfinite()
def test_gp_population_full():
Add.nargs = 2
Mul.nargs = 2
from EvoDAG.gp import Population
from EvoDAG import EvoDAG
fs = EvoDAG()._function_set
class Population2(Population):
def __init__(self, *args, **kwargs):
super(Population2, self).__init__(*args, **kwargs)
self._funcs = [Add, Sin]
self._terms = [2, 0]
def random_function(self):
func = self._funcs.pop()
if func.nargs == 1:
return func(0, weight=1)
return func(range(func.nargs), weight=np.ones(func.nargs))
def random_terminal(self):
return Variable(self._terms.pop(), 1)
pop = Population2(fs, nterminals=3)
ind = pop.create_random_ind_full(depth=2)
assert len(pop._funcs) == 0 and len(pop._terms) == 0
assert isinstance(ind[0], Sin) and isinstance(ind[1], Add)
assert ind[2].variable == 0 and ind[3].variable == 2
ind = Individual(ind)
print(X.shape, ind.individual)
hy = ind.decision_function(X)
assert hy.isfinite()
default_nargs()
def test_init_evodag_extras():
from EvoDAG import EvoDAG
from test_command_line import default_nargs
m = EvoDAG.init(seed=10, popsize=10,
early_stopping_rounds=10).fit(X, cl)
assert m.popsize == 10
default_nargs()
def test_finite():
from EvoDAG import EvoDAG
evo = EvoDAG.init()
evo._finite = False
evo.fit(X, cl)
hy = evo.predict(X)
assert (hy == cl).mean() > 0.9
def init_evodag(seed_args_X_y_test):
seed, args, X, y, test, dirname = seed_args_X_y_test
if dirname is not None:
output = os.path.join(dirname, '%s.evodag' % seed)
if os.path.isfile(output):
with gzip.open(output) as fpt:
try:
return pickle.load(fpt)
except Exception:
pass
m = EvoDAG(seed=seed, **args).fit(X, y, test_set=test)
m = m.model()
gc.collect()
if dirname is not None:
with gzip.open(output, 'w') as fpt:
pickle.dump(m, fpt)
return m
def test_popsize_nvar():
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG.init(popsize='nvar', time_limit=5)
print(X.shape)
gp.fit(X, y)
default_nargs()
assert gp.population._popsize == (X.shape[1] + len(gp._input_functions))
def test_model_nvar():
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(classifier=True,
multiple_outputs=True,
popsize=5,
share_inputs=True)
gp.fit(X, y)
assert gp._share_inputs
m = gp.model()
print(X.shape)
assert m.nvar == X.shape[1]
try:
m.predict(X[:, :3])
assert False
except RuntimeError:
pass
def test_add_repeated_args():
from EvoDAG import EvoDAG
from EvoDAG.node import Add, Min, Max
y = cl.copy()
for ff in [Add, Min, Max]:
ff.nargs = 10
gp = EvoDAG(
generations=np.inf,
tournament_size=2,
early_stopping_rounds=100,
time_limit=0.9,
# multiple_outputs=True,
classifier=False,
all_inputs=True,
function_set=[ff],
pr_variable=1,
seed=0,
popsize=10000)
gp.X = X
# gp.nclasses(y)
gp.y = y
gp.create_population()
print(gp.population.population)
node = gp.random_offspring()
print(node, node._variable, X.shape)
assert len(node._variable) <= X.shape[1]
ff.nargs = 2
def test_finite():
from EvoDAG import EvoDAG
evo = EvoDAG.init()
evo._finite = False
evo.fit(X, cl)
m = evo.model()
hy = m.predict(X)
print((hy == cl).mean(), [x.full_array() for x in m.decision_function(np.array(X))])
assert (hy == cl).mean() > 0.9
def test_X_list():
from EvoDAG import EvoDAG
from test_command_line import default_nargs
m = EvoDAG.init(seed=10, popsize=10,
early_stopping_rounds=10).fit(X.tolist(), cl)
assert m.popsize == 10
default_nargs()
print(X.shape, len(m.X))
assert len(m.X) == 4
def test_SteadyState_generation():
from EvoDAG import EvoDAG
y = cl.copy()
y[y != 1] = -1
gp = EvoDAG(population_class='SteadyState',
all_inputs=True,
classifier=False,
early_stopping_rounds=1,
popsize=2)
gp.X = X
gp.y = y
gp.create_population()
for i in range(3):
gp.replace(gp.random_offspring())
assert gp.population.generation == 2
def process_params(a):
from EvoDAG import EvoDAG
fs_class = {}
function_set = []
for x in EvoDAG()._function_set:
fs_class[x.__name__] = x
args = {}
for k, v in a.items():
if k in fs_class:
if not isinstance(v, bool):
fs_class[k].nargs = v
if v:
function_set.append(fs_class[k])
else:
args[k] = v
fs_evo = EvoDAG()._function_set
fs_evo = filter(lambda x: x in function_set, fs_evo)
args['function_set'] = [x for x in fs_evo]
return args
def test_macro_F1():
from EvoDAG.cython_utils import Score
from EvoDAG import EvoDAG
y = cl.copy()
gp = EvoDAG(generations=np.inf,
tournament_size=2,
early_stopping_rounds=100,
time_limit=0.9,
multiple_outputs=True,
seed=2,
popsize=1000)
gp.y = y
gp.X = X
gp.create_population()
off = gp.random_offspring()
hy = SparseArray.argmax(off.hy)
index = np.array(gp._mask_ts.index)
y = np.array(gp._y_klass.full_array())[index]
hy = np.array(hy.full_array())[index]
nclasses = gp._bagging_fitness.nclasses
precision = np.array([(y[hy == k] == k).mean() for k in range(nclasses)])
recall = np.array([(hy[y == k] == k).mean() for k in range(nclasses)])
print(precision, recall)
f1 = Score(nclasses)
mf1, mf1_v = f1.a_F1(gp._y_klass, SparseArray.argmax(off.hy),
gp._mask_ts.index)
for x, y in zip(precision, f1.precision):
if not np.isfinite(x):
continue
assert_almost_equals(x, y)
for x, y in zip(recall, f1.recall):
if not np.isfinite(x):
continue
assert_almost_equals(x, y)
_ = (2 * precision * recall) / (precision + recall)
m = ~np.isfinite(_)
_[m] = 0
assert_almost_equals(np.mean(_), mf1)
print(f1.precision, f1.recall, mf1, mf1_v)
gp._fitness_function = 'macro-F1'
gp._bagging_fitness.set_fitness(off)
assert_almost_equals(off.fitness, mf1 - 1)
assert_almost_equals(off.fitness_vs, mf1_v - 1)
index = np.array(gp._mask_ts.full_array()) == 0
y = np.array(gp._y_klass.full_array())[index]
hy = SparseArray.argmax(off.hy)
hy = np.array(hy.full_array())[index]
precision = np.array([(y[hy == k] == k).mean() for k in range(nclasses)])
recall = np.array([(hy[y == k] == k).mean() for k in range(nclasses)])
_ = (2 * precision * recall) / (precision + recall)
m = ~np.isfinite(_)
_[m] = 0
assert_almost_equals(np.mean(_) - 1, off.fitness_vs)
def test_process_params():
from EvoDAG.utils import RandomParameterSearch
from EvoDAG import EvoDAG
rs = RandomParameterSearch(npoints=1)
args = [x for x in rs][0]
evo = EvoDAG(**rs.process_params(args))
params = evo.get_params()
for k, v in args.items():
if k in params:
if k == 'generations':
v = np.inf
print(k, v, params[k])
if isinstance(v, list):
for a, b in zip(v, params[k]):
assert a == b.__name__
elif hasattr(params[k], '__name__'):
assert v == params[k].__name__
else:
assert v == params[k]
def test_all_init_popsize():
from EvoDAG import EvoDAG
y = cl.copy()
y[y != 1] = -1
gp = EvoDAG(population_class='Generational',
all_inputs=True,
early_stopping_rounds=1,
popsize=2)
gp.X = X
gp.y = y
gp.create_population()
assert gp.init_popsize == len(gp.X)
gp = EvoDAG(population_class='Generational',
# all_inputs=True,
early_stopping_rounds=1,
popsize=2)
gp.X = X
gp.y = y
gp.create_population()
assert gp.init_popsize == gp.popsize
def test_SteadyState_generation():
from EvoDAG import EvoDAG
y = cl.copy()
y[y != 1] = -1
gp = EvoDAG(population_class='SteadyState',
all_inputs=True,
early_stopping_rounds=1,
popsize=2)
gp.X = X
gp.y = y
gp.create_population()
for i in range(3):
gp.replace(gp.random_offspring())
assert gp.population.generation == 2
def test_all_inputs2():
from EvoDAG import EvoDAG
y = cl.copy()
y[y != 1] = -1
gp = EvoDAG(population_class='Generational',
all_inputs=True,
popsize=3)
gp.X = X
gp.y = y
gp.create_population()
print(len(gp.population.population), len(gp.X))
assert len(gp.population.population) == len(gp.X)
for i in range(gp.popsize):
a = gp.random_offspring()
gp.replace(a)
assert len(gp.population.population) == gp.popsize
def test_all_inputs():
from EvoDAG import EvoDAG
y = cl.copy()
y[y != 1] = -1
for pc in ['Generational', 'SteadyState']:
gp = EvoDAG(population_class=pc,
all_inputs=True,
popsize=10)
gp.X = X
gp.y = y
gp.create_population()
assert len(gp.population.population) < 10
for i in range(gp.population.popsize,
gp.population._popsize):
a = gp.random_offspring()
gp.replace(a)
assert len(gp.population.population) == 10
def test_clean():
from EvoDAG import EvoDAG
y = cl.copy()
y[y != 1] = -1
for pc in ['Generational', 'SteadyState']:
gp = EvoDAG(population_class=pc,
popsize=5)
gp.X = X
gp.y = y
gp.create_population()
for i in range(10):
v = gp.random_offspring()
gp.replace(v)
pop = gp.population.population
esi = gp.population.estopping
for i in gp.population._hist:
print(i == esi, i in pop, i, '-'*10, i.fitness)
if i == esi:
assert i.hy is not None
elif i in pop:
assert i.hy is not None
assert gp.population.estopping.hy is not None
def store_model(self, kw):
if self.data.ensemble_size == 1:
self.evo = EvoDAG(**kw).fit(self.X, self.y, test_set=self.Xtest)
self.model = self.evo.model()
else:
seed = self.data.seed
esize = self.data.ensemble_size
args = [(x, kw, self.X, self.y, self.Xtest)
for x in range(seed, seed+esize)]
if self.data.cpu_cores == 1:
evo = [init_evodag(x) for x in tqdm(args, total=len(args))]
else:
p = Pool(self.data.cpu_cores, maxtasksperchild=1)
evo = [x for x in tqdm(p.imap_unordered(init_evodag, args),
total=len(args))]
p.close()
self.model = Ensemble(evo)
model_file = self.get_model_file()
with gzip.open(model_file, 'w') as fpt:
pickle.dump(self.model, fpt)
pickle.dump(self.word2id, fpt)
pickle.dump(self.label2id, fpt)
def test_generational_generation():
from EvoDAG.population import Generational
from EvoDAG import EvoDAG
gp = EvoDAG(population_class='Generational',
popsize=10)
gp.X = X
y = cl.copy()
y[y != 1] = -1
gp.y = y
gp.create_population()
assert isinstance(gp.population, Generational)
p = []
for i in range(gp.popsize-1):
a = gp.random_offspring()
p.append(a)
gp.replace(a)
assert len(gp.population._inner) == (gp.popsize - 1)
a = gp.random_offspring()
p.append(a)
gp.replace(a)
assert len(gp.population._inner) == 0
for a, b in zip(gp.population.population, p):
assert a == b
def init_evodag(seed_args_X_y_test):
seed, args, X, y, test = seed_args_X_y_test
m = EvoDAG(seed=seed, **args).fit(X, y, test_set=test)
m = m.model()
gc.collect()
return m
def test_models_fitness_vs():
from EvoDAG import EvoDAG
evo = EvoDAG(popsize=10, early_stopping_rounds=2).fit(X, cl)
l_fs = [x.fitness_vs for x in evo.model().models]
assert evo.model().fitness_vs == np.median(l_fs)
def test_random_generations():
from EvoDAG import EvoDAG
from EvoDAG.population import SteadyState
class P(SteadyState):
def random_selection(self, negative=False):
raise RuntimeError('!')
y = cl.copy()
y[y != 1] = -1
for pop in ['SteadyState', 'Generational', P]:
gp = EvoDAG(population_class=pop,
all_inputs=True, random_generations=1,
early_stopping_rounds=1, popsize=2)
gp.X = X
gp.y = y
gp.create_population()
print(gp.population._random_generations)
assert gp.population._random_generations == 1
if pop == P:
try:
ind = gp.random_offspring()
gp.replace(ind)
assert False
except RuntimeError:
pass
else:
for i in range(3):
gp.replace(gp.random_offspring())
assert gp.population.generation == 2
class CommandLine(object):
def version(self):
pa = self.parser.add_argument
pa('--version',
action='version', version='EvoDAG %s' % evodag.__version__)
def output_file(self):
self.parser.add_argument('-o', '--output-file',
help='File to store the test set',
dest='output_file',
default=None,
type=str)
def ensemble(self):
self.parser.add_argument('-n', '--ensemble-size',
help='Ensemble size',
dest='ensemble_size',
default=1,
type=int)
def cores(self):
self.parser.add_argument('-u', '--cpu-cores',
help='Number of cores',
dest='cpu_cores',
default=1,
type=int)
def test_set(self):
cdn = 'File containing the test set on csv.'
self.parser.add_argument('-t', '--test_set',
default=None, type=str,
help=cdn)
def init_params(self):
pa = self.parser.add_argument
g = self.parser.add_mutually_exclusive_group(required=True)
g.add_argument('-C', '--classifier', dest='classifier',
help='The task is classification (default)',
default=True,
action="store_true")
g.add_argument('-R', '--regressor', dest='regressor',
help='The task is regression',
action="store_true")
pa('-e', '--early_stopping_rounds', dest='early_stopping_rounds',
type=int,
help='Early stopping rounds')
pa('-p', '--popsize', dest='popsize',
type=int, help='Population size')
pa('-s', '--seed', dest='seed',
default=0,
type=int, help='Seed')
pa('-j', '--json', dest='json',
action="store_true",
help='Whether the inputs are in json format',
default=False)
pa('--evolution', dest='population_class',
help="Type of evolution (SteadyState|Generational)",
type=str)
pa('--all-inputs', dest='all_inputs',
help="The initial population has all the available inputs ",
action="store_true")
pa('--time-limit', dest='time_limit',
help='Time limit in seconds', type=int)
pa('--random-generations', dest='random_generations',
help='Number of random generations', type=int)
def training_set(self):
cdn = 'File containing the training set on csv.'
self.parser.add_argument('training_set',
nargs='?',
default=None,
help=cdn)
def parse_args(self):
self.data = self.parser.parse_args()
if hasattr(self.data, 'regressor') and self.data.regressor:
self.data.classifier = False
self.main()
def convert(self, x):
try:
return float(x)
except ValueError:
if x not in self.word2id:
self.word2id[x] = len(self.word2id)
return self.word2id[x]
def convert_label(self, x):
try:
return float(x)
except ValueError:
if x not in self.label2id:
self.label2id[x] = len(self.label2id)
return self.label2id[x]
def read_data(self, fname):
with open(fname, 'r') as fpt:
l = fpt.readlines()
X = []
for i in l:
x = i.rstrip().lstrip()
if len(x):
X.append([i for i in x.split(',')])
return X
@staticmethod
def _num_terms(a):
if 'num_terms' in a:
num_terms = a['num_terms']
else:
num_terms = len(a)
if 'klass' in a:
num_terms -= 1
return num_terms
def read_data_json(self, fname):
import json
X = None
y = []
if fname.endswith('.gz'):
with gzip.open(fname, 'rb') as fpt:
l = fpt.readlines()
else:
with open(fname, 'r') as fpt:
l = fpt.readlines()
for row, d in enumerate(l):
try:
a = json.loads(str(d, encoding='utf-8'))
except TypeError:
a = json.loads(d)
if X is None:
X = [list() for i in range(self._num_terms(a))]
for k, v in a.items():
try:
k = int(k)
X[k].append((row, self.convert(v)))
except ValueError:
if k == 'klass' or k == 'y':
y.append(self.convert_label(v))
num_rows = len(l)
X = [SparseArray.init_index_data([i[0] for i in x],
[i[1] for i in x],
num_rows) for x in X]
if len(y) == 0:
y = None
else:
y = np.array(y)
return X, y
def read_training_set(self):
if self.data.training_set is None:
return
if not self.data.json:
d = self.read_data(self.data.training_set)
X = []
y = []
for x in d:
X.append([self.convert(i) for i in x[:-1]])
y.append(self.convert_label(x[-1]))
self.X = np.array(X)
self.y = np.array(y)
return True
else:
X, y = self.read_data_json(self.data.training_set)
self.X = X
self.y = y
return True
def read_test_set(self):
if self.data.test_set is None:
return False
if not self.data.json:
X = self.read_data(self.data.test_set)
self.Xtest = np.array([[self.convert(i) for i in x] for x in X])
return True
else:
X, _ = self.read_data_json(self.data.test_set)
self.Xtest = X
return True
def get_model_file(self):
if self.data.model_file is None:
a = self.data.training_set.split('.')[0]
self.data.model_file = a + '.evodag.gz'
return self.data.model_file
def store_model(self, kw):
if self.data.ensemble_size == 1:
self.evo = EvoDAG(**kw).fit(self.X, self.y, test_set=self.Xtest)
self.model = self.evo.model()
else:
seed = self.data.seed
esize = self.data.ensemble_size
args = [(x, kw, self.X, self.y, self.Xtest)
for x in range(seed, seed+esize)]
if self.data.cpu_cores == 1:
evo = [init_evodag(x) for x in tqdm(args, total=len(args))]
else:
p = Pool(self.data.cpu_cores, maxtasksperchild=1)
evo = [x for x in tqdm(p.imap_unordered(init_evodag, args),
total=len(args))]
p.close()
self.model = Ensemble(evo)
model_file = self.get_model_file()
with gzip.open(model_file, 'w') as fpt:
pickle.dump(self.model, fpt)
pickle.dump(self.word2id, fpt)
pickle.dump(self.label2id, fpt)
def get_output_file(self):
if self.data.output_file is None:
self.data.output_file = self.data.test_set + '.evodag.csv'
# if self.data.json:
# self.data.output_file += '.json'
# else:
# self.data.output_file += '.csv'
return self.data.output_file
def id2label(self, x):
if not self.data.classifier:
return x
if len(self.label2id) == 0:
return x
i2w = dict([(i[1], i[0]) for i in self.label2id.items()])
return [i2w[int(i)] for i in x]
def main(self):
pass
