class Truncate(LibFcn):
name = prefix + "truncate"
sig = Sigs([Sig([{"x": P.Array(P.Array(P.Double()))}, {"keep": P.Int()}], P.Array(P.Array(P.Double()))),
Sig([{"x": P.Map(P.Map(P.Double()))}, {"keep": P.Array(P.String())}], P.Map(P.Map(P.Double())))])
errcodeBase = 24120
def __call__(self, state, scope, pos, paramTypes, x, keep):
if keep < 0:
keep = 0
if isinstance(x, (list, tuple)) and all(isinstance(xi, (list, tuple)) for xi in x):
rows = len(x)
if rows < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
cols = len(x[0])
if cols < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
if raggedArray(x):
raise PFARuntimeException("ragged columns", self.errcodeBase + 1, self.name, pos)
return x[:keep]
elif isinstance(x, dict) and all(isinstance(x[i], dict) for i in x.keys()):
rows = rowKeys(x)
cols = colKeys(x)
if len(rows) < 1 or len(cols) < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
return dict((k, x[k]) for k in rows if k in keep)
class Symmetric(LibFcn):
name = prefix + "symmetric"
sig = Sigs([Sig([{"x": P.Array(P.Array(P.Double()))}, {"tol": P.Double()}], P.Boolean()),
Sig([{"x": P.Map(P.Map(P.Double()))}, {"tol": P.Double()}], P.Boolean())])
errcodeBase = 24100
@staticmethod
def same(x, y, tol):
if math.isinf(x) and math.isinf(y) and ((x > 0.0 and y > 0.0) or (x < 0.0 and y < 0.0)):
return True
elif math.isnan(x) and math.isnan(y):
return True
elif not math.isinf(x) and not math.isnan(x) and not math.isinf(y) and not math.isnan(y):
return abs(x - y) < tol
else:
return False
def __call__(self, state, scope, pos, paramTypes, x, tol):
if isinstance(x, (list, tuple)) and all(isinstance(xi, (list, tuple)) for xi in x):
rows = len(x)
if rows < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
cols = len(x[0])
if cols < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
if raggedArray(x):
raise PFARuntimeException("ragged columns", self.errcodeBase + 1, self.name, pos)
if rows != cols:
raise PFARuntimeException("non-square matrix", self.errcodeBase + 2, self.name, pos)
return all(all(self.same(x[i][j], x[j][i], tol) for j in range(cols)) for i in range(rows))
elif isinstance(x, dict) and all(isinstance(x[i], dict) for i in x.keys()):
keys = list(rowKeys(x).union(colKeys(x)))
if len(keys) < 1 or all(len(row) == 0 for row in x.values()):
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
return all(all(self.same(x.get(i, {}).get(j, 0.0), x.get(j, {}).get(i, 0.0), tol) for j in keys) for i in keys)
class Bernoulli(LibFcn):
name = prefix + "bernoulli"
sig = Sigs([
Sig([{
"datum": P.Array(P.String())
}, {
"classModel": P.Map(P.Double())
}], P.Double()),
Sig([{
"datum": P.Array(P.String())
}, {
"classModel": P.WildRecord("C", {"values": P.Map(P.Double())})
}], P.Double())
])
errcodeBase = 10020
def __call__(self, state, scope, pos, paramTypes, datum, classModel):
if paramTypes[1]["type"] == "record":
classModel = classModel["values"]
ll = 0.0
for v in list(classModel.values()):
if (v <= 0.0) or (v >= 1.0):
raise PFARuntimeException(
"probability in classModel cannot be less than 0 or greater than 1",
self.errcodeBase + 0, self.name, pos)
ll += math.log(1.0 - v)
for item in datum:
p = classModel.get(item, None)
if p is not None:
ll += math.log(p) - math.log(1.0 - p)
return ll
class Det(LibFcn):
name = prefix + "det"
sig = Sigs([Sig([{"x": P.Array(P.Array(P.Double()))}], P.Double()),
Sig([{"x": P.Map(P.Map(P.Double()))}], P.Double())])
errcodeBase = 24090
def __call__(self, state, scope, pos, paramTypes, x):
if isinstance(x, (list, tuple)) and all(isinstance(xi, (list, tuple)) for xi in x):
rows = len(x)
if rows < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
cols = len(x[0])
if cols < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
if raggedArray(x):
raise PFARuntimeException("ragged columns", self.errcodeBase + 1, self.name, pos)
if rows != cols:
raise PFARuntimeException("non-square matrix", self.errcodeBase + 2, self.name, pos)
if any(any(math.isnan(z) or math.isinf(z) for z in row) for row in x):
return float("nan")
else:
return float(np().linalg.det(arraysToMatrix(x)))
elif isinstance(x, dict) and all(isinstance(x[i], dict) for i in x.keys()):
keys = list(rowKeys(x).union(colKeys(x)))
if len(keys) < 1 or all(len(row) == 0 for row in x.values()):
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
if any(any(math.isnan(z) or math.isinf(z) for z in row.values()) for row in x.values()):
return float("nan")
else:
return float(np().linalg.det(mapsToMatrix(x, keys, keys)))
class AbsDiff(LibFcn):
name = prefix + "absDiff"
sig = Sig([{"x": P.Double()}, {"y": P.Double()}], P.Double())
errcodeBase = 28010
def __call__(self, state, scope, pos, paramTypes, x, y):
return abs(x - y)
class Divide(LibFcn):
name = "/"
sig = Sig([{"x": P.Double()}, {"y": P.Double()}], P.Double())
errcodeBase = 18030
def __call__(self, state, scope, pos, paramTypes, x, y):
return div(x, y)
class KSTwoSample(LibFcn):
name = prefix + "kolmogorov"
sig = Sig([{
"x": P.Array(P.Double())
}, {
"y": P.Array(P.Double())
}], P.Double())
errcodeBase = 38000
def __call__(self, state, scope, pos, paramTypes, x, y):
x = sorted([xi for xi in x if not math.isnan(xi)])
y = sorted([yi for yi in y if not math.isnan(yi)])
n1, n2 = len(x), len(y)
if (x == y):
return 1.0
elif ((len(x) == 0) or (len(y) == 0)):
return 0.0
else:
j1 = j2 = 0
fn1 = fn2 = d = 0.0
while ((j1 < n1) and (j2 < n2)):
d1 = x[j1]
d2 = y[j2]
if d1 <= d2:
j1 += 1
fn1 = float(j1) / n1
if d2 <= d1:
j2 += 1
fn2 = float(j2) / n2
dt = abs(fn2 - fn1)
if dt > d:
d = dt
en = math.sqrt((n1 * n2) / float(n1 + n2))
stat = (en + 0.12 + 0.11 / en) * d
return 1.0 - kolomogorov_cdf(stat)
class Euclidean(MetricWithMissingValues):
name = prefix + "euclidean"
sig = Sigs([
Sig([{
"similarity":
P.Fcn([P.Wildcard("A"), P.Wildcard("B")], P.Double())
}, {
"x": P.Array(P.Union([P.Null(), P.Wildcard("A")]))
}, {
"y": P.Array(P.Union([P.Null(), P.Wildcard("B")]))
}], P.Double()),
Sig([{
"similarity":
P.Fcn([P.Wildcard("A"), P.Wildcard("B")], P.Double())
}, {
"x": P.Array(P.Union([P.Null(), P.Wildcard("A")]))
}, {
"y": P.Array(P.Union([P.Null(), P.Wildcard("B")]))
}, {
"missingWeight": P.Array(P.Double())
}], P.Double())
])
errcodeBase = 28030
def increment(self, tally, x):
return tally + x**2
def finalize(self, x):
return math.sqrt(x)
class Linear(LibFcn):
name = prefix + "linear"
sig = Sig([{"x": P.Array(P.Double())},
{"y": P.Array(P.Double())}], P.Double())
errcodeBase = 23000
def __call__(self, state, scope, pos, paramTypes, x, y):
return dot(x, y, self.errcodeBase + 0, self.name, pos)
class Gaussian(LibFcn):
name = prefix + "gaussian"
sig = Sig([{"mu": P.Double()}, {"sigma": P.Double()}], P.Double())
errcodeBase = 34120
def __call__(self, state, scope, pos, paramTypes, mu, sigma):
return state.rand.gauss(mu, sigma)
class SoftMax(LibFcn):
name = prefix + "softmax"
sig = Sigs([
Sig([{
"x": P.Array(P.Double())
}], P.Array(P.Double())),
Sig([{
"x": P.Map(P.Double())
}], P.Map(P.Double()))
])
errcodeBase = 25000
def __call__(self, state, scope, pos, paramTypes, x):
if len(x) == 0:
raise PFARuntimeException("empty input", self.errcodeBase + 0,
self.name, pos)
if paramTypes[0]["type"] == "map":
xx = x.copy()
tmp = map(abs, xx.values())
if xx.values()[tmp.index(max(tmp))] >= 0:
m = max(xx.values())
else:
m = min(xx.values())
denom = sum([math.exp(v - m) for v in x.values()])
for key in x.keys():
xx[key] = float(math.exp(xx[key] - m) / denom)
return xx
else:
tmp = map(abs, x)
if x[tmp.index(max(tmp))] >= 0:
m = max(x)
else:
m = min(x)
denom = sum([math.exp(v - m) for v in x])
return [float(math.exp(val - m) / denom) for val in x]
class Mahalanobis(LibFcn):
name = prefix + "mahalanobis"
sig = Sigs([Sig([{"observation": P.Array(P.Double())}, {"prediction": P.Array(P.Double())}, {"covariance": P.Array(P.Array(P.Double()))}], P.Double(), Lifespan(None, PFAVersion(0, 7, 2), PFAVersion(0, 9, 0), "use test.mahalanobis instead")),
Sig([{"observation": P.Map(P.Double())}, {"prediction": P.Map(P.Double())}, {"covariance": P.Map(P.Map(P.Double()))}], P.Double(), Lifespan(None, PFAVersion(0, 7, 2), PFAVersion(0, 9, 0), "use test.mahalanobis instead"))])
errcodeBase = 31040
def __call__(self, state, scope, pos, paramTypes, observation, prediction, covariance):
if isinstance(observation, (tuple, list)):
if (len(observation) < 1):
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
if (len(observation) != len(prediction)):
raise PFARuntimeException("misaligned prediction", self.errcodeBase + 1, self.name, pos)
if (not all(len(i)==len(covariance[0]) for i in covariance)) and (len(covariance) != len(covariance[0])):
raise PFARuntimeException("misaligned covariance", self.errcodeBase + 2, self.name, pos)
x = np().array([(o - p) for o, p in zip(observation, prediction)])
C = np().array(covariance)
else:
if (len(observation) < 1):
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
if (len(observation) != len(prediction)):
raise PFARuntimeException("misaligned prediction", self.errcodeBase + 1, self.name, pos)
# use observation keys throughout
keys = list(observation.keys())
try:
x = np().array([observation[key] - prediction[key] for key in keys])
except:
raise PFARuntimeException("misaligned prediction", self.errcodeBase + 1, self.name, pos)
C = np().empty((len(keys), len(keys)))
try:
for i,k1 in enumerate(keys):
for j,k2 in enumerate(keys):
C[i,j] = float(covariance[k1][k2])
except:
raise PFARuntimeException("misaligned covariance", self.errcodeBase + 2, self.name, pos)
return float(np().sqrt(np().linalg.solve(C, x).T.dot(x)))
class ErrorOnNonNum(LibFcn):
name = prefix + "errorOnNonNum"
sig = Sigs([
Sig([{
"x": P.Float()
}], P.Float()),
Sig([{
"x": P.Double()
}], P.Double())
])
errcodeBase = 21050
def __call__(self, state, scope, pos, paramTypes, x):
if math.isnan(x):
raise PFARuntimeException("encountered nan", self.errcodeBase + 0,
self.name, pos)
elif math.isinf(x):
if x > 0.0:
raise PFARuntimeException("encountered +inf",
self.errcodeBase + 1, self.name, pos)
else:
raise PFARuntimeException("encountered -inf",
self.errcodeBase + 2, self.name, pos)
else:
return x
class Transpose(LibFcn):
name = prefix + "transpose"
sig = Sigs([Sig([{"x": P.Array(P.Array(P.Double()))}], P.Array(P.Array(P.Double()))),
Sig([{"x": P.Map(P.Map(P.Double()))}], P.Map(P.Map(P.Double())))])
errcodeBase = 24060
def __call__(self, state, scope, pos, paramTypes, x):
if isinstance(x, (list, tuple)) and all(isinstance(xi, (list, tuple)) for xi in x):
rows = len(x)
if rows < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
cols = len(x[0])
if cols < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
if raggedArray(x):
raise PFARuntimeException("ragged columns", self.errcodeBase + 1, self.name, pos)
return [[x[r][c] for r in range(rows)] for c in range(cols)]
elif isinstance(x, dict) and all(isinstance(x[i], dict) for i in x.keys()):
rows = rowKeys(x)
cols = colKeys(x)
if len(rows) < 1 or len(cols) < 1:
raise PFARuntimeException("too few rows/cols", self.errcodeBase + 0, self.name, pos)
if raggedMap(x):
raise PFARuntimeException("ragged columns", self.errcodeBase + 1, self.name, pos)
return dict((c, dict((r, x[r][c]) for r in rows)) for c in cols)
class Chi2Prob(LibFcn):
name = prefix + "chi2Prob"
sig = Sig([{
"state": P.WildRecord("A", {
"chi2": P.Double(),
"dof": P.Int()
})
}], P.Double())
errcodeBase = 38060
def __call__(self, state, scope, pos, paramTypes, state_):
chi2 = state_["chi2"]
dof = state_["dof"]
if dof < 0:
raise PFARuntimeException("invalid parameterization",
self.errcodeBase + 0, self.name, pos)
elif dof == 0:
if chi2 > 0:
return 1.0
else:
return 0.0
elif math.isnan(chi2):
return float("nan")
elif math.isinf(chi2):
if chi2 > 0:
return 1.0
else:
return 0.0
else:
return float(
Chi2Distribution(dof, self.errcodeBase + 0, self.name,
pos).CDF(chi2))
class UpdateHoltWintersPeriodic(LibFcn):
name = prefix + "updateHoltWintersPeriodic"
sig = Sig([{"x": P.Double()}, {"alpha": P.Double()}, {"beta": P.Double()}, {"gamma": P.Double()}, {"state": P.WildRecord("A", {"level": P.Double(), "trend": P.Double(), "cycle": P.Array(P.Double()), "multiplicative": P.Boolean()})}], P.Wildcard("A"))
errcodeBase = 14050
def __call__(self, state, scope, pos, paramTypes, x, alpha, beta, gamma, theState):
if alpha < 0.0 or alpha > 1.0:
raise PFARuntimeException("alpha out of range", self.errcodeBase + 0, self.name, pos)
if beta < 0.0 or beta > 1.0:
raise PFARuntimeException("beta out of range", self.errcodeBase + 1, self.name, pos)
if gamma < 0.0 or gamma > 1.0:
raise PFARuntimeException("gamma out of range", self.errcodeBase + 2, self.name, pos)
level_prev = theState["level"]
trend_prev = theState["trend"]
cycle_unrotated = theState["cycle"]
if len(cycle_unrotated) == 0:
raise PFARuntimeException("empty cycle", self.errcodeBase + 3, self.name, pos)
cycle_rotated = cycle_unrotated[1:] + [cycle_unrotated[0]]
cycle_prev = cycle_rotated[0]
if theState["multiplicative"]:
level = div(alpha * x, cycle_prev) + (1.0 - alpha) * (level_prev + trend_prev)
trend = beta * (level - level_prev) + (1.0 - beta) * trend_prev
cycle = div(gamma * x, level) + (1.0 - gamma) * cycle_prev
else:
level = alpha * (x - cycle_prev) + (1.0 - alpha) * (level_prev + trend_prev)
trend = beta * (level - level_prev) + (1.0 - beta) * trend_prev
cycle = gamma * (x - level) + (1.0 - gamma) * cycle_prev
return dict(theState, level=level, trend=trend, cycle=([cycle] + cycle_rotated[1:]))
class Update(UpdateMeanVariance):
name = prefix + "update"
sig = Sig([{"x": P.Double()}, {"w": P.Double()}, {"state": P.WildRecord("A", {"count": P.Double()})}], P.Wildcard("A"))
errcodeBase = 14000
def _getRecord(self, paramType):
if isinstance(paramType, AvroUnion):
for t in paramType.types:
if not isinstance(t, AvroNull):
return t
else:
return paramType
def __call__(self, state, scope, pos, paramTypes, x, w, theState, level):
originalCount = theState["count"]
count = originalCount + w
if level == 0:
return dict(theState, count=count)
else:
mean = theState["mean"]
delta = x - mean
shift = div(delta * w, count)
mean += shift
if level == 1:
return dict(theState, count=count, mean=mean)
else:
varianceTimesCount = theState["variance"] * originalCount
varianceTimesCount += originalCount * delta * shift
return dict(theState, count=count, mean=mean, variance=div(varianceTimesCount, count))
class UpdateEWMA(LibFcn):
name = prefix + "updateEWMA"
sig = Sig([{"x": P.Double()}, {"alpha": P.Double()}, {"state": P.WildRecord("A", {"mean": P.Double()})}], P.Wildcard("A"))
errcodeBase = 14030
def _getRecord(self, paramType):
if isinstance(paramType, AvroUnion):
for t in paramType.types:
if not isinstance(t, AvroNull):
return t
else:
return paramType
def genpy(self, paramTypes, args, pos):
hasVariance = False
for x in self._getRecord(paramTypes[2]).fields:
if x.name == "variance":
if not x.avroType.accepts(AvroDouble()):
raise PFASemanticException(self.name + " is being given a state record type in which the \"variance\" field is not a double: " + str(x.avroType), None)
hasVariance = True
return "self.f[{0}]({1}, {2})".format(repr(self.name), ", ".join(["state", "scope", repr(pos), repr(paramTypes)] + args), hasVariance)
def __call__(self, state, scope, pos, paramTypes, x, alpha, theState, hasVariance):
if alpha < 0.0 or alpha > 1.0:
raise PFARuntimeException("alpha out of range", self.errcodeBase + 0, self.name, pos)
mean = theState["mean"]
diff = x - mean
incr = alpha * diff
if hasVariance:
variance = theState["variance"]
return dict(theState, mean=(mean + incr), variance=((1.0 - alpha) * (variance + diff * incr)))
else:
return dict(theState, mean=(mean + incr))
class Bin(LibFcn):
name = prefix + "bin"
sig = Sigs([Sig([{"x": P.Double()}, {"numbins": P.Int()}, {"low": P.Double()}, {"high": P.Double()}], P.Int()),
Sig([{"x": P.Double()}, {"origin": P.Double()}, {"width": P.Double()}], P.Int())])
errcodeBase = 22000
def __call__(self, state, scope, pos, paramTypes, x, *args):
if len(args) == 3:
numbins, low, high = args
if low >= high or math.isnan(low) or math.isnan(high):
raise PFARuntimeException("bad histogram range", self.errcodeBase + 0, self.name, pos)
if numbins < 1:
raise PFARuntimeException("bad histogram scale", self.errcodeBase + 1, self.name, pos)
if math.isnan(x) or x < low or x >= high:
raise PFARuntimeException("x out of range", self.errcodeBase + 2, self.name, pos)
out = int(math.floor(numbins * div((x - low), (high - low))))
if out < 0 or out >= numbins:
raise PFARuntimeException("x out of range", self.errcodeBase + 2, self.name, pos)
return out
else:
origin, width = args
if math.isnan(origin) or math.isinf(origin):
raise PFARuntimeException("bad histogram range", self.errcodeBase + 0, self.name, pos)
if width <= 0.0 or math.isnan(width):
raise PFARuntimeException("bad histogram scale", self.errcodeBase + 1, self.name, pos)
if math.isnan(x) or math.isinf(x):
raise PFARuntimeException("x out of range", self.errcodeBase + 2, self.name, pos)
else:
return int(math.floor(div((x - origin), width)))
class UpdateHoltWinters(LibFcn):
name = prefix + "updateHoltWinters"
sig = Sig(
[{
"x": P.Double()
}, {
"alpha": P.Double()
}, {
"beta": P.Double()
}, {
"state": P.WildRecord("A", {
"level": P.Double(),
"trend": P.Double()
})
}], P.Wildcard("A"))
errcodeBase = 14040
def __call__(self, state, scope, pos, paramTypes, x, alpha, beta,
theState):
if alpha < 0.0 or alpha > 1.0:
raise PFARuntimeException("alpha out of range",
self.errcodeBase + 0, self.name, pos)
if beta < 0.0 or beta > 1.0:
raise PFARuntimeException("beta out of range",
self.errcodeBase + 1, self.name, pos)
level_prev = theState["level"]
trend_prev = theState["trend"]
level = alpha * x + (1.0 - alpha) * (level_prev + trend_prev)
trend = beta * (level - level_prev) + (1.0 - beta) * trend_prev
return dict(theState, level=level, trend=trend)
class Trace(LibFcn):
name = prefix + "trace"
sig = Sigs([
Sig([{
"x": P.Array(P.Array(P.Double()))
}], P.Double()),
Sig([{
"x": P.Map(P.Map(P.Double()))
}], P.Double())
])
errcodeBase = 24080
def __call__(self, state, scope, pos, paramTypes, x):
if isinstance(x, (list, tuple)) and all(
isinstance(xi, (list, tuple)) for xi in x):
rows = len(x)
if rows == 0:
return 0.0
else:
cols = len(x[0])
if raggedArray(x):
raise PFARuntimeException("ragged columns",
self.errcodeBase + 0, self.name,
pos)
return sum(x[i][i] for i in xrange(min(rows, cols)))
elif isinstance(x, dict) and all(
isinstance(x[i], dict) for i in x.keys()):
keys = rowKeys(x).intersection(colKeys(x))
return sum(x[i][i] for i in keys)
class Taxicab(MetricWithMissingValues):
name = prefix + "taxicab"
sig = Sigs([
Sig([{
"similarity":
P.Fcn([P.Wildcard("A"), P.Wildcard("B")], P.Double())
}, {
"x": P.Array(P.Union([P.Null(), P.Wildcard("A")]))
}, {
"y": P.Array(P.Union([P.Null(), P.Wildcard("B")]))
}], P.Double()),
Sig([{
"similarity":
P.Fcn([P.Wildcard("A"), P.Wildcard("B")], P.Double())
}, {
"x": P.Array(P.Union([P.Null(), P.Wildcard("A")]))
}, {
"y": P.Array(P.Union([P.Null(), P.Wildcard("B")]))
}, {
"missingWeight": P.Array(P.Double())
}], P.Double())
])
errcodeBase = 28060
def increment(self, tally, x):
return tally + x
def finalize(self, x):
return x
class Forecast1HoltWinters(LibFcn):
name = prefix + "forecast1HoltWinters"
sig = Sig(
[{
"state": P.WildRecord("A", {
"level": P.Double(),
"trend": P.Double()
})
}], P.Double())
errcodeBase = 14060
def genpy(self, paramTypes, args, pos):
hasCycle = False
hasMultiplicative = False
for x in paramTypes[0].fields:
if x.name == "cycle":
if not x.avroType.accepts(AvroArray(AvroDouble())):
raise PFASemanticException(
self.name +
" is being given a state record type in which the \"cycle\" field is not an array of double: "
+ str(x.avroType), None)
hasCycle = True
elif x.name == "multiplicative":
if not x.avroType.accepts(AvroBoolean()):
raise PFASemanticException(
self.name +
" is being given a state record type in which the \"multiplicative\" field is not a boolean: "
+ str(x.avroType), None)
hasMultiplicative = True
if hasCycle ^ hasMultiplicative:
raise PFASemanticException(
self.name +
" is being given a state record type with a \"cycle\" but no \"multiplicative\" or vice-versa",
None)
return "self.f[{0}]({1}, {2})".format(
repr(self.name), ", ".join(
["state", "scope",
repr(pos), repr(paramTypes)] + args), hasCycle)
def __call__(self, state, scope, pos, paramTypes, theState, hasPeriodic):
level = theState["level"]
trend = theState["trend"]
if not hasPeriodic:
return level + trend
else:
cycle = theState["cycle"]
L = len(cycle)
if L == 0:
raise PFARuntimeException("empty cycle", self.errcodeBase + 0,
self.name, pos)
if theState["multiplicative"]:
return (level + trend) * cycle[1 % L]
else:
return level + trend + cycle[1 % L]
class GeoMean(LibFcn):
name = prefix + "geomean"
sig = Sig([{"a": P.Array(P.Double())}], P.Double())
errcodeBase = 15440
def __call__(self, state, scope, pos, paramTypes, a):
if len(a) == 0:
return float("nan")
else:
return reduce(lambda a, b: a * b, a, 1.0)**(1.0/len(a))
class Score(LibFcn):
name = prefix + "score"
sig = Sig([{
"datum": P.Array(P.Double())
}, {
"model":
P.WildRecord(
"L", {
"const":
P.Double(),
"posClass":
P.Array(
P.WildRecord("M", {
"supVec": P.Array(P.Double()),
"coeff": P.Double()
})),
"negClass":
P.Array(
P.WildRecord("N", {
"supVec": P.Array(P.Double()),
"coeff": P.Double()
}))
})
}, {
"kernel":
P.Fcn([P.Array(P.Double()), P.Array(P.Double())], P.Double())
}], P.Double())
errcodeBase = 12000
def __call__(self, state, scope, pos, paramTypes, datum, model, kernel):
const = model["const"]
negClass = model["negClass"]
posClass = model["posClass"]
if len(negClass) == 0 and len(posClass) == 0:
raise PFARuntimeException("no support vectors",
self.errcodeBase + 0, self.name, pos)
negClassScore = 0.0
for sv in negClass:
supVec = sv["supVec"]
if len(supVec) != len(datum):
raise PFARuntimeException(
"support vectors must have same length as datum",
self.errcodeBase + 1, self.name, pos)
coeff = sv["coeff"]
negClassScore += callfcn(state, scope, kernel,
[supVec, datum]) * coeff
posClassScore = 0.0
for sv in posClass:
supVec = sv["supVec"]
if len(supVec) != len(datum):
raise PFARuntimeException(
"support vectors must have same length as datum",
self.errcodeBase + 1, self.name, pos)
coeff = sv["coeff"]
posClassScore += callfcn(state, scope, kernel,
[supVec, datum]) * coeff
return negClassScore + posClassScore + const
class Sigmoid(LibFcn):
name = prefix + "sigmoid"
sig = Sig([{"x": P.Array(P.Double())},
{"y": P.Array(P.Double())},
{"gamma": P.Double()},
{"intercept": P.Double()}], P.Double())
errcodeBase = 23030
def __call__(self, state, scope, pos, paramTypes, x, y, gamma, intercept):
return math.tanh(gamma * dot(x, y, self.errcodeBase + 0, self.name, pos) + intercept)
class TanH(LibFcn):
name = prefix + "tanh"
sig = Sig([{"x": P.Double()}], P.Double())
errcodeBase = 27260
def genpy(self, paramTypes, args, pos):
return "math.tanh({0})".format(*args)
def __call__(self, state, scope, pos, paramTypes, x):
return math.tanh(x)
class ACos(LibFcn):
name = prefix + "acos"
sig = Sig([{"x": P.Double()}], P.Double())
errcodeBase = 27030
def __call__(self, state, scope, pos, paramTypes, x):
try:
return math.acos(x)
except ValueError:
return float("nan")
class RandomDouble(LibFcn):
name = prefix + "double"
sig = Sig([{"low": P.Double()}, {"high": P.Double()}], P.Double())
errcodeBase = 34030
def __call__(self, state, scope, pos, paramTypes, low, high):
if high <= low:
raise PFARuntimeException("high must be greater than low",
self.errcodeBase + 0, self.name, pos)
return state.rand.uniform(low, high)
class Tan(LibFcn):
name = prefix + "tan"
sig = Sig([{"x": P.Double()}], P.Double())
errcodeBase = 27250
def __call__(self, state, scope, pos, paramTypes, x):
if math.isinf(x) or math.isnan(x):
return float("nan")
else:
return math.tan(x)