def compare_strategies(length, ncat, op, dtype):
keys = ak.randint(0, ncat, length)
if dtype == 'int64':
vals = ak.randint(0, length // ncat, length)
elif dtype == 'bool':
vals = ak.zeros(length, dtype='bool')
for i in np.random.randint(0, length, ncat // 2):
vals[i] = True
else:
vals = ak.linspace(-1, 1, length)
print("Global groupby", end=' ')
start = time()
gg = ak.GroupBy(keys, False)
ggtime = time() - start
print(ggtime)
print("Global reduce", end=' ')
start = time()
gk, gv = gg.aggregate(vals, op)
grtime = time() - start
print(grtime)
print("Local groupby", end=' ')
start = time()
lg = ak.GroupBy(keys, True)
lgtime = time() - start
print(lgtime)
print("Local reduce", end=' ')
start = time()
lk, lv = lg.aggregate(vals, op)
lrtime = time() - start
print(lrtime)
print(f"Keys match? {(gk == lk).all()}")
print(f"Absolute diff of vals = {ak.abs(gv - lv).sum()}")
return ggtime, grtime, lgtime, lrtime
def check_linspace(N):
# create np version
a = np.linspace(10, 20, N)
# create ak version
b = ak.linspace(10, 20, N)
# print(a,b)
f = np.allclose(a, b.to_ndarray())
return pass_fail(f)
def check_linspace(N):
# create np version
a = ak.array(np.linspace(10, 20, N))
# create ak version
b = ak.linspace(10, 20, N)
# print(a,b)
c = a == b
# print(type(c),c)
return pass_fail(c.all())
def generate_arrays(length, nkeys, nvals, dtype='int64'):
keys = ak.randint(0, nkeys, length)
if dtype == 'int64':
vals = ak.randint(0, nvals, length)
elif dtype == 'bool':
vals = ak.zeros(length, dtype='bool')
for i in np.random.randint(0, length, nkeys // 2):
vals[i] = True
else:
vals = ak.linspace(-1, 1, length)
return keys, vals
def run_tests():
global pdarrays
pdarrays = {
'int64': ak.arange(1, SIZE + 1, 1),
'float64': ak.linspace(0, 1, SIZE),
'bool': (ak.arange(0, SIZE, 1) % 2) == 0
}
global ndarrays
ndarrays = {
'int64': np.arange(1, SIZE + 1, 1),
'float64': np.linspace(0, 1, SIZE),
'bool': (np.arange(0, SIZE, 1) % 2) == 0
}
global scalars
#scalars = {k: v[SIZE//2] for k, v in ndarrays.items()}
scalars = {'int64': 5, 'float64': 3.14159, 'bool': True}
dtypes = pdarrays.keys()
print("Dtypes: ", dtypes)
print("pdarrays: ")
for k, v in pdarrays.items():
print(k, ": ", v)
print("ndarrays: ")
for k, v in ndarrays.items():
print(k, ": ", v)
print("scalars: ")
for k, v in scalars.items():
print(k, ": ", v)
def do_op(lt, rt, ls, rs, isarkouda, oper):
evalstr = ''
if ls:
evalstr += 'scalars["{}"]'.format(lt)
else:
evalstr += '{}["{}"]'.format(('ndarrays', 'pdarrays')[isarkouda],
lt)
evalstr += ' {} '.format(oper)
if rs:
evalstr += 'scalars["{}"]'.format(rt)
else:
evalstr += '{}["{}"]'.format(('ndarrays', 'pdarrays')[isarkouda],
rt)
#print(evalstr)
res = eval(evalstr)
return res
results = {
'not_implemented': [],
'caught': [],
'wrong_dtype': [],
'wrong_value': [],
'failure': []
}
tests = 0
for ltype, rtype, op in product(dtypes, dtypes, ak.pdarray.BinOps):
for lscalar, rscalar in ((False, False), (False, True), (True, False)):
tests += 1
expression = "{}({}) {} {}({})".format(ltype, ('array',
'scalar')[lscalar],
op, rtype,
('array',
'scalar')[rscalar])
try:
npres = do_op(ltype, rtype, lscalar, rscalar, False, op)
except TypeError: # numpy doesn't implement operation
try:
akres = do_op(ltype, rtype, lscalar, rscalar, True, op)
except RuntimeError as e:
results['not_implemented'].append((expression, e))
continue
try:
akres = do_op(ltype, rtype, lscalar, rscalar, True, op)
except RuntimeError as e:
warnings.warn("Error computing {}\n{}".format(
expression, str(e)))
results['caught'].append((expression, e))
continue
try:
akrestype = akres.dtype
except Exception as e:
warnings.warn(
"Cannot detect return dtype of ak result: {} (np result: {})"
.format(akres, npres))
results['failure'].append((expression, e))
continue
if akrestype != npres.dtype:
restypes = "{}(np) vs. {}(ak)".format(npres.dtype, akrestype)
warnings.warn("dtype mismatch: {} = {}".format(
expression, restypes))
results['wrong_dtype'].append((expression, restypes))
continue
try:
akasnp = akres.to_ndarray()
except Exception as e:
warnings.warn("Could not convert to ndarray: {}".format(akres))
results['failure'].append((expression, e))
continue
if not np.allclose(akasnp, npres, equal_nan=True):
res = "np: {}\nak: {}".format(npres, akasnp)
warnings.warn("result mismatch: {} =\n{}".format(
expression, res))
results['wrong_value'].append((expression, res))
for errtype, errs in results.items():
print("{} {}".format(len(errs), errtype))
for expr, msg in errs:
print(expr)
print(msg)
print("{} differences from numpy in {} tests".format(
sum(len(errs)
for errs in results.values()) - len(results['not_implemented']),
tests))
print(iv) b = ak.zeros(iv.size,dtype=ak.int64) a[iv] = b print(a) ak.v = False a = ak.randint(10,30,40) vc = ak.value_counts(a) print(vc[0].size,vc[0]) print(vc[1].size,vc[1]) ak.v = False a = ak.arange(0,10,1) b = a[a<5] a = ak.linspace(0,9,10) b = a[a<5] print(b) ak.v = True ak.pdarrayIterThresh = 1000 a = ak.arange(0,10,1) print(list(a)) ak.v = False a = ak.randint(10,30,40) u = ak.unique(a) h = ak.histogram(a,bins=20) print(a) print(h.size,h) print(u.size,u)
def run_tests(verbose):
global pdarrays
pdarrays = {
'int64': ak.arange(0, SIZE, 1),
'float64': ak.linspace(0, 2, SIZE),
'bool': (ak.arange(0, SIZE, 1) % 2) == 0
}
global ndarrays
ndarrays = {
'int64': np.arange(0, SIZE, 1),
'float64': np.linspace(0, 2, SIZE),
'bool': (np.arange(0, SIZE, 1) % 2) == 0
}
global scalars
#scalars = {k: v[SIZE//2] for k, v in ndarrays.items()}
scalars = {'int64': 5, 'float64': 3.14159, 'bool': True}
dtypes = pdarrays.keys()
if verbose:
print("Operators: ", ak.pdarray.BinOps)
print("Dtypes: ", dtypes)
print("pdarrays: ")
for k, v in pdarrays.items():
print(k, ": ", v)
print("ndarrays: ")
for k, v in ndarrays.items():
print(k, ": ", v)
print("scalars: ")
for k, v in scalars.items():
print(k, ": ", v)
def do_op(lt, rt, ls, rs, isarkouda, oper):
evalstr = ''
if ls:
evalstr += 'scalars["{}"]'.format(lt)
else:
evalstr += '{}["{}"]'.format(('ndarrays', 'pdarrays')[isarkouda],
lt)
evalstr += ' {} '.format(oper)
if rs:
evalstr += 'scalars["{}"]'.format(rt)
else:
evalstr += '{}["{}"]'.format(('ndarrays', 'pdarrays')[isarkouda],
rt)
#print(evalstr)
res = eval(evalstr)
return res
results = {
'neither_implement': [], # (expression, ak_error)
'arkouda_minus_numpy': [], # (expression, ak_result, error_on_exec?)
'numpy_minus_arkouda': [], # (expression, ak_result, error_on_exec?)
'both_implement': []
} # (expression, ak_result, error_on_exec?, dtype_mismatch?, value_mismatch?)
tests = 0
for ltype, rtype, op in product(dtypes, dtypes, ak.pdarray.BinOps):
for lscalar, rscalar in ((False, False), (False, True), (True, False)):
tests += 1
expression = "{}({}) {} {}({})".format(ltype, ('array',
'scalar')[lscalar],
op, rtype,
('array',
'scalar')[rscalar])
try:
npres = do_op(ltype, rtype, lscalar, rscalar, False, op)
except TypeError: # numpy doesn't implement operation
try:
akres = do_op(ltype, rtype, lscalar, rscalar, True, op)
except RuntimeError as e:
if 'not implemented' in str(
e): # neither numpy nor arkouda implement
results['neither_implement'].append(
(expression, str(e)))
else: # arkouda implements with error, np does not implement
results['arkouda_minus_numpy'].append(
(expression, str(e), True))
continue
# arkouda implements but not numpy
results['arkouda_minus_numpy'].append(
(expression, str(akres), False))
continue
try:
akres = do_op(ltype, rtype, lscalar, rscalar, True, op)
except RuntimeError as e:
if 'not implemented' in str(
e): # numpy implements but not arkouda
results['numpy_minus_arkouda'].append(
(expression, str(e), True))
else: # both implement, but arkouda errors
results['both_implement'].append(
(expression, str(e), True, False, False))
continue
# both numpy and arkouda execute without error
try:
akrestype = akres.dtype
except Exception as e:
warnings.warn(
"Cannot detect return dtype of ak result: {} (np result: {})"
.format(akres, npres))
results['both_implement'].append(
(expression, str(akres), False, True, False))
continue
if akrestype != npres.dtype:
restypes = "{}(np) vs. {}(ak)".format(npres.dtype, akrestype)
#warnings.warn("dtype mismatch: {} = {}".format(expression, restypes))
results['both_implement'].append(
(expression, restypes, False, True, False))
continue
try:
akasnp = akres.to_ndarray()
except Exception as e:
warnings.warn("Could not convert to ndarray: {}".format(akres))
results['both_implement'].append(
(expression, str(akres), True, False, False))
continue
if not np.allclose(akasnp, npres, equal_nan=True):
res = "np: {}\nak: {}".format(npres, akasnp)
# warnings.warn("result mismatch: {} =\n{}".format(expression, res))
results['both_implement'].append(
(expression, res, False, False, True))
# Finally, both numpy and arkouda agree on result
results['both_implement'].append(
(expression, "", False, False, False))
print("# ops not implemented by numpy or arkouda: {}".format(
len(results['neither_implement'])))
if verbose:
for expression, err in results['neither_implement']:
print(expression)
print("# ops implemented by numpy but not arkouda: {}".format(
len(results['numpy_minus_arkouda'])))
if verbose:
for expression, err, flag in results['numpy_minus_arkouda']:
print(expression)
print("# ops implemented by arkouda but not numpy: {}".format(
len(results['arkouda_minus_numpy'])))
if verbose:
for expression, res, flag in results['arkouda_minus_numpy']:
print(expression, " -> ", res)
nboth = len(results['both_implement'])
print("# ops implemented by both: {}".format(nboth))
matches = 0
execerrors = []
dtypeerrors = []
valueerrors = []
for (expression, res, ex, dt, val) in results['both_implement']:
matches += not any((ex, dt, val))
if ex: execerrors.append((expression, res))
if dt: dtypeerrors.append((expression, res))
if val: valueerrors.append((expression, res))
print(" Matching results: {} / {}".format(matches, nboth))
print(" Arkouda execution errors: {} / {}".format(len(execerrors), nboth))
if verbose: print('\n'.join(map(': '.join, execerrors)))
print(" Dtype mismatches: {} / {}".format(len(dtypeerrors),
nboth))
if verbose: print('\n'.join(map(': '.join, dtypeerrors)))
print(" Value mismatches: {} / {}".format(len(valueerrors),
nboth))
if verbose: print('\n'.join(map(': '.join, valueerrors)))
return matches == nboth
