Floating point error if optimized code used #689

fcr · 2018-07-29T09:50:06Z

Run GCAL_Tutorial.ipynb after building the optimized code.
During the stage of "Rebuilding snapshot 'lissom_oo_or_10000.typ' " generally get:

FloatingPointError Traceback (most recent call last)
in ()
9 model_dir = param.resolve_path('examples' if model=='GCAL' else 'models', path_to_file=False)
10 runscript(os.path.join(model_dir, basename + ".ty")) # Load the model
---> 11 topo.sim.views = c(times=times)
12 save_snapshot(snapshot_name)
13

/Users/fred/Development/topographica/external/featuremapper/featuremapper/collector.pyc in call(self, attrtree, times, strict)
541 try:
542 for i, t in enumerate(np.diff(times)):
--> 543 interval_hook(float(t))
544
545 # An empty attrtree buffer stops analysis repeatedly

/Users/fred/Development/topographica/external/holoviews/holoviews/ipython/widgets.pyc in call(self, value)
203 remaining = value - completed
204 if remaining != 0:
--> 205 self.run_hook(remaining)
206 super(RunProgress, self).call(100)
207

/Users/fred/Development/topographica/topo/base/simulation.pyc in run(self, duration, until)
1345 event = self.events.pop(0)
1346 self.debug("Delivering %s",event)
-> 1347 event(self)
1348 did_event=True
1349

/Users/fred/Development/topographica/topo/base/simulation.pyc in call(self, sim)
446
447 def call(self,sim):
--> 448 self.conn.dest.input_event(self.conn,self.data)
449
450 def repr(self):

/Users/fred/Development/topographica/topo/sheet/init.pyc in input_event(self, conn, data)
314 proj.activity *= 0.0
315 self.mask.reset()
--> 316 super(SettlingCFSheet,self).input_event(conn,data)
317
318

/Users/fred/Development/topographica/topo/base/projection.pyc in input_event(self, conn, data)
467 """
468 self.verbose("Received input from %s on dest_port %s via connection %s.", conn.src.name,conn.dest_port,conn.name)
--> 469 self.present_input(data,conn)
470 self.new_input = True
471

/Users/fred/Development/topographica/topo/base/projection.pyc in present_input(self, input_activity, conn)
596 is called.
597 """
--> 598 conn.activate(input_activity)
599
600

/Users/fred/Development/topographica/topo/base/cf.pyc in activate(self, input_activity)
811 iaf(input_activity)
812 self.input_buffer = input_activity
--> 813 self.activity *=0.0
814 self.response_fn(CFIter(self), input_activity, self.activity, self.strength)
815 for of in self.output_fns:

FloatingPointError: invalid value encountered in multiply

Note that this does not occur if the non optimized code is used (but the process takes forever).

philippjfr · 2018-08-04T23:09:46Z

Presumably the optimized code is the weave version of the code? Maybe you could try to switch to the cythonized version in that case. You can see those being added here:

topographica/topo/submodel/scal.py

Line 217 in f945a8b

projection.CFProjection.response_fn=optimized.CFPRF_DotProduct_cython()

fcr · 2018-08-05T06:10:14Z

AFAIK I am using the Cython code:

python setup.py build_ext
In file included from /Users/fred/anaconda/envs/python2/lib/python2.7/site-packages/numpy/core/include/numpy/ndarraytypes.h:1818:0,
from /Users/fred/anaconda/envs/python2/lib/python2.7/site-packages/numpy/core/include/numpy/ndarrayobject.h:18,
from /Users/fred/anaconda/envs/python2/lib/python2.7/site-packages/numpy/core/include/numpy/arrayobject.h:4,
from topo/optimized/optimized.c:567:
/Users/fred/anaconda/envs/python2/lib/python2.7/site-packages/numpy/core/include/numpy/npy_1_7_deprecated_api.h:15:2: warning: #warning "Using deprecated NumPy API, disable it by " "#defining NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION" [-Wcpp]
#warning "Using deprecated NumPy API, disable it by "
^~~~~~~
cc1plus: warning: command line option '-Wstrict-prototypes' is valid for C/ObjC but not for C++

fcr · 2018-08-05T07:58:34Z

Correction.
Okay, seems that weave was active.
I just remembered that I installed weave when I was first trying to run topographica a few weeks ago, after I ran the LISSOM tutorial and I got this message:
"Caution: Unable to import Weave. Will use non-optimized versions of most components.
Note: Inline-optimized components are currently disabled; see topo.misc.inlinec".

This was in spite of building with Cython as you saw in my previous message. This is my first time encountering weave.

When I uninstall weave I get back the "Unable to import Weave" message again.

Now that you showed how to activate the cython code in the model I'll give it a try.

fcr · 2018-08-05T08:29:41Z

Ran with the Cython code. Got the same error:
`FloatingPointError Traceback (most recent call last)
in ()
11 print(model_path)
12 runscript(model_path) # Load the model
---> 13 topo.sim.views = c(times=times)
14 save_snapshot(snapshot_name)
15