precompute lineq timeseries during boolean evaluation

stl/boolean_eval.py

@@ -8,23 +8,32 @@ import funcy as fn
 from lenses import lens
 
 import stl.ast
+import stl
 
 oo = float('inf')
 
+def pointwise_sat(phi):
+    ap_names = [z.id.name for z in stl.utils.AP_lens(phi).get_all()]
+    def _eval_stl(x, t):
+        evaluated = stl.utils.eval_lineqs(phi, x)
+        evaluated.update(fn.project(x, ap_names))
+        return eval_stl(phi)(evaluated, t)
+    return _eval_stl
+
 @singledispatch
-def pointwise_sat(stl):
+def eval_stl(stl):
     raise NotImplementedError
 
 
-@pointwise_sat.register(stl.Or)
+@eval_stl.register(stl.Or)
 def _(stl):
-    fs = [pointwise_sat(arg) for arg in stl.args]
+    fs = [eval_stl(arg) for arg in stl.args]
     return lambda x, t: any(f(x, t) for f in fs)
 
 
-@pointwise_sat.register(stl.And)
+@eval_stl.register(stl.And)
 def _(stl):
-    fs = [pointwise_sat(arg) for arg in stl.args]
+    fs = [eval_stl(arg) for arg in stl.args]
     return lambda x, t: all(f(x, t) for f in fs)
 
 
@@ -33,7 +42,10 @@ def get_times(x, tau, lo=None, hi=None):
         lo = min(v.first()[0] for v in x.values())
     if hi is None or hi is oo:
         hi = max(v.last()[0] for v in x.values())
+    try:
         end = min(v.domain.end() for v in x.values())
+    except:
+        import pdb; pdb.set_trace()
     hi = hi + tau if hi + tau <= end else end
     lo = lo + tau if lo + tau <= end else end
 
@@ -46,10 +58,10 @@ def get_times(x, tau, lo=None, hi=None):
     return sorted(set(fn.pluck(0, all_times)))
 
 
-@pointwise_sat.register(stl.Until)
+@eval_stl.register(stl.Until)
 def _(stl):
     def _until(x, t):
-        f1, f2 = pointwise_sat(stl.arg1), pointwise_sat(stl.arg2)
+        f1, f2 = eval_stl(stl.arg1), eval_stl(stl.arg2)
         for tau in get_times(x, t):
             if not f1(x, tau):
                 return f2(x, tau)
@@ -65,23 +77,23 @@ def eval_unary_temporal_op(phi, always=True):
         return lambda x, t: retval
     if hi == lo:
         return lambda x, t: f(x, t)
-    f = pointwise_sat(phi.arg) 
+    f = eval_stl(phi.arg) 
     return lambda x, t: fold(f(x, tau) for tau in get_times(x, t, lo, hi))
 
 
-@pointwise_sat.register(stl.F)
+@eval_stl.register(stl.F)
 def _(phi):
     return eval_unary_temporal_op(phi, always=False)
 
 
-@pointwise_sat.register(stl.G)
+@eval_stl.register(stl.G)
 def _(phi):
     return eval_unary_temporal_op(phi, always=True)
 
 
-@pointwise_sat.register(stl.Neg)
+@eval_stl.register(stl.Neg)
 def _(stl):
-    f = pointwise_sat(stl.arg)
+    f = eval_stl(stl.arg)
     return lambda x, t: not f(x, t)
 
 
@@ -94,22 +106,20 @@ op_lookup = {
 }
 
 
-@pointwise_sat.register(stl.AtomicPred)
+@eval_stl.register(stl.AtomicPred)
 def _(stl):
     return lambda x, t: x[str(stl.id)][t]
 
 
-@pointwise_sat.register(stl.LinEq)
-def _(stl):
-    op = op_lookup[stl.op]
-    return lambda x, t: op(eval_terms(stl, x, t), stl.const)
+@eval_stl.register(stl.LinEq)
+def _(lineq):
+    return lambda x, t: x[lineq][t]
 
 
 def eval_terms(lineq, x, t):
-    psi = lens(lineq).terms.each_().modify(eval_term(x, t))
-    return sum(psi.terms)
+    terms = lens(lineq).terms.each_().get_all()
+    return sum(eval_term(term, x, t) for term in terms)
 
 
-def eval_term(x, t):
-    # TODO(lift interpolation much higher)
-    return lambda term: term.coeff*x[term.id.name][t]
+def eval_term(term, x, t):
+    return float(term.coeff)*x[term.id.name][t]

stl/synth.py

@@ -1,7 +1,7 @@
 import operator as op
 
 from stl.utils import set_params, param_lens
-from stl import pointwise_sat
+from stl.boolean_eval import pointwise_sat
 
 from lenses import lens
 

stl/utils.py

@@ -6,6 +6,7 @@ from functools import reduce
 from lenses import lens, Lens
 import funcy as fn
 import sympy
+import traces
 
 import stl.ast
 from stl.ast import (LinEq, And, Or, NaryOpSTL, F, G, Interval, Neg,
@@ -138,6 +139,43 @@ def inline_context(phi, context):
     # TODO: this is hack to flatten the AST. Fix!
     return stl.parse(str(phi))
 
+op_lookup = {
+    ">": op.gt,
+    ">=": op.ge,
+    "<": op.lt,
+    "<=": op.le,
+    "=": op.eq,
+}
+
+def get_times(x):
+    times = set.union(*({t for t, _ in v.items()} for v in x.values()))
+    return sorted(times)
+
+
+def eval_lineq(lineq, x, times=None, compact=True):
+    if times is None:
+        times = get_times(x)
+
+    def eval_term(term, t):
+        return float(term.coeff)*x[term.id.name][t]
+
+    output = traces.TimeSeries(domain=traces.Domain(times[0], times[-1]))
+    terms = lens(lineq).terms.each_().get_all()
+    for t in times:
+        lhs = sum(eval_term(term, t) for term in terms)
+        output[t] = op_lookup[lineq.op](lhs, lineq.const)
+        
+    if compact:
+        output.compact()
+    return output
+
+def eval_lineqs(phi, x, times=None):
+    if times is None:
+        times = get_times(x)
+    lineqs = set(lineq_lens(phi).get_all())
+    return {lineq: eval_lineq(lineq, x, times=times) for lineq in lineqs}
+
+
 # EDSL
 
 def alw(phi, *, lo, hi):