model of DCN pyramidal neuron
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 

205 lines
7.5 KiB

from __future__ import print_function
import os, sys, pickle, pprint
import numpy as np
import pyqtgraph as pg
from .. import AUDIT_TESTS
class UserTester(object):
"""
Base class for testing when a human is required to verify the results.
When a test is passed by the user, its output is saved and used as a basis
for future tests. If future test results do not match the stored results,
then the user is asked to decide whether to fail the test, or pass the
test and store new results.
Subclasses must reimplement run_test() to return a dictionary of results
to store. Optionally, compare_results and audit_result may also be
reimplemented to customize the testing behavior.
By default, test results are stored in a 'test_data' directory relative
to the file that defines the UserTester subclass in use.
"""
data_dir = "test_data"
def __init__(self, key, *args, **kwds):
"""Initialize with a string *key* that provides a short, unique
description of this test. All other arguments are passed to run_test().
*key* is used to determine the file name for storing test results.
"""
self.audit = AUDIT_TESTS
self.key = key
self.rtol = 1e-3
self.args = args
self.assert_test_info(*args, **kwds)
def run_test(self, *args, **kwds):
"""
Exceute the test. All arguments are taken from __init__.
Return a picklable dictionary of test results.
"""
raise NotImplementedError()
def compare_results(self, key, info, expect):
"""
Compare *result* of the current test against the previously stored
result *expect*. If *expect* is None, then no previous result was
stored.
If *result* and *expect* do not match, then raise an exception.
"""
# Check test structures are the same
assert type(info) is type(expect)
if hasattr(info, "__len__"):
assert len(info) == len(expect)
if isinstance(info, dict):
for k in info:
assert k in expect
for k in expect:
assert k in info
self.compare_results(k, info[k], expect[k])
elif isinstance(info, list):
for i in range(len(info)):
self.compare_results(key, info[i], expect[i])
elif isinstance(info, np.ndarray):
assert info.shape == expect.shape
if len(info) == 0:
return
# assert info.dtype == expect.dtype
if info.dtype.fields is None:
intnan = -9223372036854775808 # happens when np.nan is cast to int
inans = np.isnan(info) | (info == intnan)
enans = np.isnan(expect) | (expect == intnan)
assert np.all(inans == enans)
mask = ~inans
if not np.allclose(info[mask], expect[mask], rtol=self.rtol):
print(
"\nComparing data array, shapes match: ",
info.shape == expect.shape,
)
print("Model tested: %s, measure: %s" % (self.key, key))
# print( 'args: ', dir(self.args[0]))
print("Array expected: ", expect[mask])
print("Array received: ", info[mask])
try:
self.args[0].print_all_mechs()
except:
print("args[0] is string: ", self.args[0])
assert np.allclose(info[mask], expect[mask], rtol=self.rtol)
else:
for k in info.dtype.fields.keys():
self.compare_results(k, info[k], expect[k])
elif np.isscalar(info):
if not np.allclose(info, expect, rtol=self.rtol):
print("Comparing Scalar data, model: %s, measure: %s" % (self.key, key))
# print 'args: ', dir(self.args[0])
print(
"Expected: ",
expect,
", received: ",
info,
" relative tolerance: ",
self.rtol,
)
if isinstance(self.args[0], str):
pass
# print ': ', str
else:
self.args[0].print_all_mechs()
assert np.allclose(info, expect, rtol=self.rtol)
else:
try:
assert info == expect
except AssertionError:
raise
except Exception:
raise NotImplementedError(
"Cannot compare objects of type %s" % type(info)
)
def audit_result(self, info, expect):
""" Display results and ask the user to decide whether the test passed.
Return True for pass, False for fail.
If *expect* is None, then no previous test results were stored.
"""
app = pg.mkQApp()
print("\n=== New test results for %s: ===\n" % self.key)
pprint.pprint(info)
# we use DiffTreeWidget to display differences between large data structures, but
# this is not present in mainline pyqtgraph yet
if hasattr(pg, "DiffTreeWidget"):
win = pg.DiffTreeWidget()
else:
from cnmodel.util.difftreewidget import DiffTreeWidget
win = DiffTreeWidget()
win.resize(800, 800)
win.setData(expect, info)
win.show()
print("Store new test results? [y/n]")
yn = raw_input()
win.hide()
return yn.lower().startswith("y")
def assert_test_info(self, *args, **kwds):
"""
Test *cell* and raise exception if the results do not match prior
data.
"""
result = self.run_test(*args, **kwds)
expect = self.load_test_result()
try:
assert expect is not None
self.compare_results(None, result, expect)
except:
if not self.audit:
if expect is None:
raise Exception(
"No prior test results for test '%s'. "
"Run test.py --audit store new test data." % self.key
)
else:
raise
store = self.audit_result(result, expect)
if store:
self.save_test_result(result)
else:
raise Exception("Rejected test results for '%s'" % self.key)
def result_file(self):
"""
Return a file name to be used for storing / retrieving test results
given *self.key*.
"""
modfile = sys.modules[self.__class__.__module__].__file__
path = os.path.dirname(modfile)
return os.path.join(path, self.data_dir, self.key + ".pk")
def load_test_result(self):
"""
Load prior test results for *self.key*.
If there are no prior results, return None.
"""
fn = self.result_file()
if os.path.isfile(fn):
return pickle.load(open(fn, "rb"), encoding="latin1")
return None
def save_test_result(self, result):
"""
Store test results for *self.key*.
Th e*result* argument must be picklable.
"""
fn = self.result_file()
dirname = os.path.dirname(fn)
if not os.path.isdir(dirname):
os.mkdir(dirname)
pickle.dump(result, open(fn, "wb"))