def hp_categorical(self, memo, node, label, tid, val):
"""
Parameters: See `hp_uniform`
"""
size = memo[node.arg['size']]
if size == 0:
return []
val1 = np.atleast_1d(val)
p = p_orig = np.asarray(memo[node.arg['p']])
if p.ndim == 2:
if len(p) not in (1, len(val1)):
print node
print p
print np.asarray(p).shape
assert len(p) in (1, len(val1))
else:
assert p.ndim == 1
p = p[np.newaxis, :]
upper = memo[node.arg['upper']]
if val1.size:
counts = np.bincount(val1, minlength=upper) / float(val1.size)
prior = self.shrinking(label)
else:
counts = np.zeros(upper)
prior = 1.0
new_p = (1 - prior) * counts + prior * p
assert new_p.ndim == 2
rval = categorical(p=new_p, rng=self.rng, size=size)
if p_orig.ndim == 1:
assert len(rval) == 1
return rval[0]
else:
return rval
def hp_categorical(self, memo, node, label, tid, val):
"""
Parameters: See `hp_uniform`
"""
size = memo[node.arg['size']]
if size == 0:
return []
val1 = np.atleast_1d(val)
p = p_orig = np.asarray(memo[node.arg['p']])
if p.ndim == 2:
if len(p) not in (1, len(val1)):
print node
print p
print np.asarray(p).shape
assert len(p) in (1, len(val1))
else:
assert p.ndim == 1
p = p[np.newaxis, :]
upper = memo[node.arg['upper']]
if val1.size:
counts = np.bincount(val1, minlength=upper) / float(val1.size)
prior = self.shrinking(label)
else:
counts = np.zeros(upper)
prior = 1.0
new_p = (1 - prior) * counts + prior * p
assert new_p.ndim == 2
rval = categorical(p=new_p, rng=self.rng, size=size)
if p_orig.ndim == 1:
assert len(rval) == 1
return rval[0]
else:
return rval
def hp_categorical(self, memo, node, label, tid, val):
"""
Parameters: See `hp_uniform`
"""
p = p_orig = np.asarray(memo[node.arg['p']])
if p.ndim == 2:
assert len(p) == 1
p = p[0]
counts = np.zeros_like(p)
counts[val] += 1
prior = self.shrinking(label)
new_p = (1 - prior) * counts + prior * p
if p_orig.ndim == 2:
rval = categorical(p=[new_p], rng=self.rng,
size=memo[node.arg['size']])
else:
rval = categorical(p=new_p, rng=self.rng,
size=memo[node.arg['size']])
return rval
def hp_randint(self, memo, node, label, tid, val):
"""
Parameters: See `hp_uniform`
"""
upper = memo[node.arg['upper']]
counts = np.zeros(upper)
counts[val] += 1
prior = self.shrinking(label)
p = (1 - prior) * counts + prior * (1.0 / upper)
rval = categorical(p=p, upper=upper, rng=self.rng,
size=memo[node.arg['size']])
return rval
def hp_categorical(self, memo, node, label, tid, val):
"""
Parameters: See `hp_uniform`
"""
p = p_orig = np.asarray(memo[node.arg['p']])
if p.ndim == 2:
assert len(p) == 1
p = p[0]
counts = np.zeros_like(p)
counts[val] += 1
prior = self.shrinking(label)
new_p = (1 - prior) * counts + prior * p
if p_orig.ndim == 2:
rval = categorical(p=[new_p],
rng=self.rng,
size=memo[node.arg['size']])
else:
rval = categorical(p=new_p,
rng=self.rng,
size=memo[node.arg['size']])
return rval
def hp_randint(self, memo, node, label, tid, val):
"""
Parameters: See `hp_uniform`
"""
upper = memo[node.arg['upper']]
counts = np.zeros(upper)
counts[val] += 1
prior = self.shrinking(label)
p = (1 - prior) * counts + prior * (1.0 / upper)
rval = categorical(p=p,
upper=upper,
rng=self.rng,
size=memo[node.arg['size']])
return rval
def hp_randint(self, memo, node, label, tid, val):
"""
Parameters: See `hp_uniform`
"""
upper = memo[node.arg['upper']]
val1 = np.atleast_1d(val)
if val1.size:
counts = np.bincount(val1, minlength=upper) / float(val1.size)
else:
counts = np.zeros(upper)
prior = 1.0
prior = self.shrinking(label)
p = (1 - prior) * counts + prior * (1.0 / upper)
rval = categorical(p=p, upper=upper, rng=self.rng,
size=memo[node.arg['size']])
return rval
def hp_randint(self, memo, node, label, tid, val):
"""
Parameters: See `hp_uniform`
"""
upper = memo[node.arg['upper']]
val1 = np.atleast_1d(val)
if val1.size:
counts = np.bincount(val1, minlength=upper) / float(val1.size)
else:
counts = np.zeros(upper)
prior = 1.0
prior = self.shrinking(label)
p = (1 - prior) * counts + prior * (1.0 / upper)
rval = categorical(p=p,
upper=upper,
rng=self.rng,
size=memo[node.arg['size']])
return rval
