def test_kmeans(self):
def test_impl(numCenter, numIter, N, D):
A = np.ones((N, D))
centroids = np.zeros((numCenter, D))
for l in range(numIter):
dist = np.array([[
sqrt(np.sum((A[i, :] - centroids[j, :])**2))
for j in range(numCenter)
] for i in range(N)])
labels = np.array([dist[i, :].argmin() for i in range(N)])
centroids = np.array([[
np.sum(A[labels == i, j]) / np.sum(labels == i)
for j in range(D)
] for i in range(numCenter)])
return centroids
hpat_func = hpat.jit(test_impl)
n = 11
np.testing.assert_allclose(hpat_func(1, 1, n, 2),
test_impl(1, 1, n, 2))
self.assertEqual(count_array_OneDs(), 4)
self.assertEqual(count_array_OneD_Vars(), 1)
self.assertEqual(count_parfor_OneDs(), 5)
self.assertEqual(count_parfor_OneD_Vars(), 1)
def test_intraday(self):
def test_impl(nsyms):
max_num_days = 100
all_res = 0.0
for i in hpat.prange(nsyms):
s_open = 20 * np.ones(max_num_days)
s_low = 28 * np.ones(max_num_days)
s_close = 19 * np.ones(max_num_days)
df = pd.DataFrame({
'Open': s_open,
'Low': s_low,
'Close': s_close
})
df['Stdev'] = df['Close'].rolling(window=90).std()
df['Moving Average'] = df['Close'].rolling(window=20).mean()
df['Criteria1'] = (df['Open'] -
df['Low'].shift(1)) < -df['Stdev']
df['Criteria2'] = df['Open'] > df['Moving Average']
df['BUY'] = df['Criteria1'] & df['Criteria2']
df['Pct Change'] = (df['Close'] - df['Open']) / df['Open']
df['Rets'] = df['Pct Change'][df['BUY'] == True]
all_res += df['Rets'].mean()
return all_res
hpat_func = hpat.jit(test_impl)
n = 11
self.assertEqual(hpat_func(n), test_impl(n))
self.assertEqual(count_array_OneDs(), 0)
self.assertEqual(count_parfor_OneDs(), 1)
def test_dist_input(self):
def test_impl(A):
return len(A)
hpat_func = hpat.jit(distributed=['A'])(test_impl)
n = 128
arr = np.ones(n)
np.testing.assert_allclose(hpat_func(arr) / self.num_ranks, test_impl(arr))
self.assertEqual(count_array_OneDs(), 1)
def test_series_head_index_parallel1(self):
def test_impl(S):
return S.head(3)
S = pd.Series([6,9,2,3,6,4,5], ['a','ab','abc','c','f','hh',''])
hpat_func = hpat.jit(distributed={'S'})(test_impl)
start, end = get_start_end(len(S))
pd.testing.assert_series_equal(hpat_func(S[start:end]), test_impl(S))
self.assertTrue(count_array_OneDs()>0)
def test_series_head_index_parallel1(self):
'''Verifies head method for distributed Series with integer index'''
def test_impl(S):
return S.head(3)
hpat_func = hpat.jit(distributed={'S'})(test_impl)
S = pd.Series([6, 9, 2, 3, 6, 4, 5], [8, 1, 6, 0, 9, 1, 3])
start, end = get_start_end(len(S))
pd.testing.assert_series_equal(hpat_func(S[start:end]), test_impl(S))
self.assertTrue(count_array_OneDs() > 0)
def test_rebalance(self):
def test_impl(N):
A = np.arange(n)
B = A[A > 10]
C = hpat.distributed_api.rebalance_array(B)
return C.sum()
hpat_func = hpat.jit(test_impl)
n = 128
np.testing.assert_allclose(hpat_func(n), test_impl(n))
self.assertEqual(count_array_OneDs(), 3)
self.assertEqual(count_parfor_OneDs(), 2)
def test_array_reduce(self):
def test_impl(N):
A = np.ones(3)
B = np.ones(3)
for i in numba.prange(N):
A += B
return A
hpat_func = hpat.jit(test_impl)
n = 128
np.testing.assert_allclose(hpat_func(n), test_impl(n))
self.assertEqual(count_array_OneDs(), 0)
self.assertEqual(count_parfor_OneDs(), 1)
def test_join1_seq(self):
def test_impl(n):
df1 = pd.DataFrame({'key1': np.arange(n)+3, 'A': np.arange(n)+1.0})
df2 = pd.DataFrame({'key2': 2*np.arange(n)+1, 'B': n+np.arange(n)+1.0})
df3 = pd.merge(df1, df2, left_on='key1', right_on='key2')
return df3.B
hpat_func = hpat.jit(test_impl)
n = 11
self.assertEqual(hpat_func(n).sum(), test_impl(n).sum())
self.assertEqual(count_array_OneDs(), 0)
self.assertEqual(count_parfor_OneDs(), 0)
n = 11111
self.assertEqual(hpat_func(n).sum(), test_impl(n).sum())
def test_series_head_parallel1(self):
'''Verifies head method for distributed Series with string data and no index'''
def test_impl(S):
return S.head(7)
hpat_func = hpat.jit(distributed={'S'})(test_impl)
# need to test different lenghts, as head's size is fixed and implementation
# depends on relation of size of the data per processor to output data size
for n in range(1, 5):
S = pd.Series(['a', 'ab', 'abc', 'c', 'f', 'hh', ''] * n)
start, end = get_start_end(len(S))
pd.testing.assert_series_equal(hpat_func(S[start:end]), test_impl(S))
self.assertTrue(count_array_OneDs() > 0)
def test_cumsum(self):
def test_impl(n):
df = pd.DataFrame({'A': np.ones(n), 'B': np.random.ranf(n)})
Ac = df.A.cumsum()
return Ac.sum()
hpat_func = hpat.jit(test_impl)
n = 11
self.assertEqual(hpat_func(n), test_impl(n))
self.assertEqual(count_array_REPs(), 0)
self.assertEqual(count_array_OneDs(), 2)
self.assertEqual(count_parfor_REPs(), 0)
self.assertEqual(count_parfor_OneDs(), 2)
self.assertTrue(dist_IR_contains('dist_cumsum'))
def test_dist_return(self):
def test_impl(N):
A = np.arange(N)
return A
hpat_func = hpat.jit(locals={'A:return': 'distributed'})(test_impl)
n = 128
dist_sum = hpat.jit(lambda a: hpat.distributed_api.dist_reduce(
a, np.int32(hpat.distributed_api.Reduce_Type.Sum.value)))
dist_sum(1) # run to compile
np.testing.assert_allclose(dist_sum(hpat_func(n).sum()),
test_impl(n).sum())
self.assertEqual(count_array_OneDs(), 1)
self.assertEqual(count_parfor_OneDs(), 1)
def test_box_dist_return(self):
def test_impl(n):
df = pd.DataFrame({'A': np.ones(n), 'B': np.arange(n)})
return df
hpat_func = hpat.jit(distributed={'df'})(test_impl)
n = 11
hres, res = hpat_func(n), test_impl(n)
self.assertEqual(count_array_OneDs(), 3)
self.assertEqual(count_parfor_OneDs(), 2)
dist_sum = hpat.jit(lambda a: hpat.distributed_api.dist_reduce(
a, np.int32(hpat.distributed_api.Reduce_Type.Sum.value)))
dist_sum(1) # run to compile
np.testing.assert_allclose(dist_sum(hres.A.sum()), res.A.sum())
np.testing.assert_allclose(dist_sum(hres.B.sum()), res.B.sum())
def test_rebalance_loop(self):
def test_impl(N):
A = np.arange(n)
B = A[A > 10]
s = 0
for i in range(3):
s += B.sum()
return s
hpat_func = hpat.jit(test_impl)
n = 128
np.testing.assert_allclose(hpat_func(n), test_impl(n))
self.assertEqual(count_array_OneDs(), 4)
self.assertEqual(count_parfor_OneDs(), 2)
self.assertIn('allgather', list(hpat_func.inspect_llvm().values())[0])
def test_logistic_regression(self):
def test_impl(n, d):
iterations = 3
X = np.ones((n, d)) + .5
Y = np.ones(n)
D = X.shape[1]
w = np.ones(D) - 0.5
for i in range(iterations):
w -= np.dot(((1.0 / (1.0 + np.exp(-Y * np.dot(X, w))) - 1.0) * Y), X)
return w
hpat_func = hpat.jit(test_impl)
n = 11
d = 4
np.testing.assert_allclose(hpat_func(n, d), test_impl(n, d))
self.assertEqual(count_array_OneDs(), 3)
self.assertEqual(count_parfor_OneDs(), 3)
def test_dist_return_tuple(self):
def test_impl(N):
A = np.arange(N)
B = np.arange(N) + 1.5
return A, B
hpat_func = hpat.jit(locals={'A:return': 'distributed',
'B:return': 'distributed'})(test_impl)
n = 128
dist_sum = hpat.jit(
lambda a: hpat.distributed_api.dist_reduce(
a, np.int32(hpat.distributed_api.Reduce_Type.Sum.value)))
dist_sum(1.0) # run to compile
np.testing.assert_allclose(
dist_sum((hpat_func(n)[0] + hpat_func(n)[1]).sum()), (test_impl(n)[0] + test_impl(n)[1]).sum())
self.assertEqual(count_array_OneDs(), 2)
self.assertEqual(count_parfor_OneDs(), 2)
def test_linear_regression(self):
def test_impl(N, D):
p = 2
iterations = 3
X = np.ones((N, D)) + .5
Y = np.ones((N, p))
alphaN = 0.01 / N
w = np.zeros((D, p))
for i in range(iterations):
w -= alphaN * np.dot(X.T, np.dot(X, w) - Y)
return w
hpat_func = hpat.jit(test_impl)
n = 11
d = 4
np.testing.assert_allclose(hpat_func(n, d), test_impl(n, d))
self.assertEqual(count_array_OneDs(), 5)
self.assertEqual(count_parfor_OneDs(), 3)
def test_join_left_parallel1(self):
"""
"""
def test_impl(A1, B1, C1, A2, B2, D2):
df1 = pd.DataFrame({'A': A1, 'B': B1, 'C': C1})
df2 = pd.DataFrame({'A': A2, 'B': B2, 'D': D2})
df3 = df1.merge(df2, on=('A', 'B'))
return df3.C.sum() + df3.D.sum()
hpat_func = hpat.jit(
locals={
'A1:input': 'distributed',
'B1:input': 'distributed',
'C1:input': 'distributed',
})(test_impl)
df1 = pd.DataFrame({
'A': [3, 1, 1, 3, 4],
'B': [1, 2, 3, 2, 3],
'C': [7, 8, 9, 4, 5]
})
df2 = pd.DataFrame({
'A': [2, 1, 4, 4, 3],
'B': [1, 3, 2, 3, 2],
'D': [1, 2, 3, 4, 8]
})
start, end = get_start_end(len(df1))
h_A1 = df1.A.values[start:end]
h_B1 = df1.B.values[start:end]
h_C1 = df1.C.values[start:end]
h_A2 = df2.A.values
h_B2 = df2.B.values
h_D2 = df2.D.values
p_A1 = df1.A.values
p_B1 = df1.B.values
p_C1 = df1.C.values
p_A2 = df2.A.values
p_B2 = df2.B.values
p_D2 = df2.D.values
h_res = hpat_func(h_A1, h_B1, h_C1, h_A2, h_B2, h_D2)
p_res = test_impl(p_A1, p_B1, p_C1, p_A2, p_B2, p_D2)
self.assertEqual(h_res, p_res)
self.assertEqual(count_array_OneDs(), 3)
def test_kde(self):
def test_impl(n):
X = np.ones(n)
b = 0.5
points = np.array([-1.0, 2.0, 5.0])
N = points.shape[0]
exps = 0
for i in hpat.prange(n):
p = X[i]
d = (-(p - points)**2) / (2 * b**2)
m = np.min(d)
exps += m - np.log(b * N) + np.log(np.sum(np.exp(d - m)))
return exps
hpat_func = hpat.jit(test_impl)
n = 11
np.testing.assert_approx_equal(hpat_func(n), test_impl(n))
self.assertEqual(count_array_OneDs(), 1)
self.assertEqual(count_parfor_OneDs(), 2)
def test_logistic_regression_acc(self):
def test_impl(N, D):
iterations = 3
g = 2 * np.ones(D) - 1
X = 2 * np.ones((N, D)) - 1
Y = ((np.dot(X, g) > 0.0) == (np.ones(N) > .90)) + .0
w = 2 * np.ones(D) - 1
for i in range(iterations):
w -= np.dot(((1.0 / (1.0 + np.exp(-Y * np.dot(X, w))) - 1.0) * Y), X)
R = np.dot(X, w) > 0.0
accuracy = np.sum(R == Y) / N
return accuracy
hpat_func = hpat.jit(test_impl)
n = 11
d = 4
np.testing.assert_approx_equal(hpat_func(n, d), test_impl(n, d))
self.assertEqual(count_array_OneDs(), 3)
self.assertEqual(count_parfor_OneDs(), 4)
def test_array_reduce(self):
binops = ['+=', '*=', '+=', '*=', '|=', '|=']
dtypes = ['np.float32', 'np.float32', 'np.float64', 'np.float64', 'np.int32', 'np.int64']
for (op, typ) in zip(binops, dtypes):
func_text = """def f(n):
A = np.arange(0, 10, 1, {})
B = np.arange(0 + 3, 10 + 3, 1, {})
for i in numba.prange(n):
A {} B
return A
""".format(typ, typ, op)
loc_vars = {}
exec(func_text, {'np': np, 'numba': numba}, loc_vars)
test_impl = loc_vars['f']
hpat_func = hpat.jit(test_impl)
n = 128
np.testing.assert_allclose(hpat_func(n), test_impl(n))
self.assertEqual(count_array_OneDs(), 0)
self.assertEqual(count_parfor_OneDs(), 1)