def summarize(d):
print('newton delta:')
print(arbplf_newton_delta(json.dumps(d)))
print('deriv:')
print(arbplf_deriv(json.dumps(d)))
print('ll:')
print(arbplf_ll(json.dumps(d)))
def myderiv(d):
"""
Provides a dict -> pandas.DataFrame wrapper of the pure JSON arbplf_deriv.
"""
s = arbplf_deriv(json.dumps(d))
df = pd.read_json(StringIO(s), orient='split', precise_float=True)
return df
def summarize(d):
print('newton delta:')
print(arbplf_newton_delta(json.dumps(d)))
print('deriv:')
print(arbplf_deriv(json.dumps(d)))
print('ll:')
print(arbplf_ll(json.dumps(d)))
def test_truncated_deriv():
d = copy.deepcopy(D)
d['model_and_data']['probability_array'][0][-1] = [0, 1]
s = arbplf_deriv(json.dumps(d))
df = pd.read_json(StringIO(s), orient='split', precise_float=True)
actual = df.set_index('edge').value.values
# compute the desired closed form solution
T = sum(rates)
# derivative of log(1 - exp(-T))
desired = np.reciprocal(expm1(T))
# compare actual and desired result
assert_allclose(actual, desired)
def generic_objective(get_s, X):
s_in = get_s(X)
neg_ll = partial(generic_neg_ll, get_s)
# compute negative log likelihood
s_out = arbplf_ll(s_in)
df = pd.read_json(StringIO(s_out), orient='split', precise_float=True)
y = -df.value.values[0]
# explicitly compute derivatives with respect to edge rate coefficients
s_out = arbplf_deriv(s_in)
df = pd.read_json(StringIO(s_out), orient='split', precise_float=True)
d = -df.set_index('edge').value.values
edge_count = len(_edges)
D = np.concatenate((d, [0]*3))
request = np.array([0]*edge_count + [1]*3)
_finite_diffs(neg_ll, request, D, X, y)
print(y, D)
return y, D
def myderiv(d):
s = arbplf_deriv(json.dumps(d))
df = pd.read_json(StringIO(s), orient="split", precise_float=True)
return df
def myderiv(d):
s = arbplf_deriv(json.dumps(d))
df = pd.read_json(StringIO(s), orient='split', precise_float=True)
return df
