def test_incorporate_sampleid_existing():
rng = np.random.RandomState()
trcrpm = TRCRP_Mixture(chains=1, lag=2, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
new_frame = pd.DataFrame(
[
[nan, nan, 99.],
[44., nan, nan],
[21., nan, 88.],
],
index=[1, 3, 9],
columns=['a', 'b', 'c'],
)
trcrpm.incorporate(new_frame)
tabulated_data = trcrpm.engine.states[0].data_array()
expected_data = [
[nan, nan, 86., nan, nan, 57., nan, nan, 65.],
[nan, 86., 19., nan, 57., 62., nan, 65., 99.],
[86., 19., 17., 57., 62., 41., 65., 99., 17.],
[19., 17., 44., 62., 41., 7., 99., 17., 30.],
[17., 44., 75., 41., 7., 1., 17., 30., 12.],
[44., 75., 45., 7., 1., nan, 30., 12., 72.],
[75., 45., 48., 1., nan, 77., 12., 72., 8.],
[45., 48., 29., nan, 77., nan, 72., 8., 86.],
[48., 29., 83., 77., nan, 46., 8., 86., 38.],
[29., 83., 21., nan, 46., 54., 86., 38., 88.],
]
assert np.allclose(tabulated_data, expected_data, equal_nan=True)
def test_timepoint_to_rowid():
rng = np.random.RandomState(2)
trcrpm = TRCRP_Mixture(chains=1, lag=0, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
for i in xrange(len(FRAME)):
assert trcrpm._timepoint_to_rowid(i) == i
assert trcrpm._timepoint_to_rowid(i) == i
def test_incorporate_sampleid_non_contiguous():
rng = np.random.RandomState(1)
trcrpm = TRCRP_Mixture(chains=1, lag=2, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
# Now incorporate an non-contiguous row skipping last_sampled_id+1.
last_sampled_id = max(FRAME.index)
frame_new = pd.DataFrame([[11, 12, float('nan')]],
columns=['a', 'b', 'c'],
index=[last_sampled_id + 2])
trcrpm.incorporate(frame_new)
tabulated_data = trcrpm.engine.states[0].data_array()
expected_data = [
[nan, nan, 86., nan, nan, 57., nan, nan, 65.],
[nan, 86., 19., nan, 57., 62., nan, 65., nan],
[86., 19., 17., 57., 62., 41., 65., nan, 17.],
[19., 17., nan, 62., 41., 7., nan, 17., 30.],
[17., nan, 75., 41., 7., 1., 17., 30., 12.],
[nan, 75., 45., 7., 1., nan, 30., 12., 72.],
[75., 45., 48., 1., nan, 77., 12., 72., 8.],
[45., 48., 29., nan, 77., nan, 72., 8., 86.],
[48., 29., 83., 77., nan, 46., 8., 86., 38.],
[29., 83., nan, nan, 46., 54., 86., 38., nan],
[nan, nan, 11, 54., nan, 12, nan, nan, nan],
]
assert np.allclose(tabulated_data, expected_data, equal_nan=True)
def test_tabulate_lagged_data():
rng = np.random.RandomState(2)
trcrpm = TRCRP_Mixture(chains=1, lag=0, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
tabulated_data = trcrpm.engine.states[0].data_array()
expected_data = DATA_RAW
assert np.allclose(tabulated_data, DATA_RAW, equal_nan=True)
trcrpm = TRCRP_Mixture(chains=1, lag=2, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
tabulated_data = trcrpm.engine.states[0].data_array()
expected_data = [
[nan, nan, 86., nan, nan, 57., nan, nan, 65.],
[nan, 86., 19., nan, 57., 62., nan, 65., nan],
[86., 19., 17., 57., 62., 41., 65., nan, 17.],
[19., 17., nan, 62., 41., 7., nan, 17., 30.],
[17., nan, 75., 41., 7., 1., 17., 30., 12.],
[nan, 75., 45., 7., 1., nan, 30., 12., 72.],
[75., 45., 48., 1., nan, 77., 12., 72., 8.],
[45., 48., 29., nan, 77., nan, 72., 8., 86.],
[48., 29., 83., 77., nan, 46., 8., 86., 38.],
[29., 83., nan, nan, 46., 54., 86., 38., nan],
]
assert np.allclose(tabulated_data, expected_data, equal_nan=True)
def test_get_dependence_probabilities_crash():
rng = np.random.RandomState(1)
trcrpm = TRCRP_Mixture(chains=4, lag=3, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
nvars = len(trcrpm.variables)
dependencies = trcrpm.dependence_probability_pairwise()
assert np.allclose(dependencies, np.ones((nvars, nvars)))
def test_get_temporal_regimes_crash():
rng = np.random.RandomState(1)
trcrpm = TRCRP_Mixture(chains=4, lag=3, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
for variable in trcrpm.variables:
regimes_all = trcrpm.get_temporal_regimes(variable)
assert np.shape(regimes_all) == (trcrpm.chains, len(trcrpm.dataset))
regimes_some = trcrpm.get_temporal_regimes(variable,
timepoints=[0, 1, 2])
assert np.shape(regimes_some) == (trcrpm.chains, 3)
def test_serialize_trcrp_mixture():
rng = np.random.RandomState(1)
trcrpm = TRCRP_Mixture(chains=4, lag=3, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
metadata = trcrpm.to_metadata()
modulename, attributename = metadata['factory']
module = importlib.import_module(modulename)
builder = getattr(module, attributename)
trcrpm2 = builder.from_metadata(metadata, seed=1)
assert isinstance(trcrpm2, TRCRP_Mixture)
def test_trcrp_mixture_all_dependent():
rng = np.random.RandomState(2)
trcrpm = TRCRP_Mixture(chains=3, lag=3, variables=FRAME.columns, rng=rng)
with pytest.raises(ValueError):
# No data incorporated yet.
trcrpm.resample_all(steps=10)
trcrpm.incorporate(FRAME)
trcrpm.resample_all(steps=10)
for state in trcrpm.engine.states:
assert len(state.views) == 1
trcrpm.resample_hyperparameters(steps=5)
regimes_a = trcrpm.get_temporal_regimes('a')
regimes_b = trcrpm.get_temporal_regimes('b')
regimes_c = trcrpm.get_temporal_regimes('c')
assert np.all(regimes_a == regimes_b)
assert np.all(regimes_a == regimes_c)
def test_simulate_dimensions():
rng = np.random.RandomState(1)
trcrpm = TRCRP_Mixture(chains=4, lag=3, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
def check_dims_correct(timepoints, variables, nsamples, simulator):
samples = simulator(timepoints, variables, nsamples, multiprocess=0)
# Number of chains is 4.
assert len(samples) == nsamples * 4
for sample in samples:
assert len(sample) == len(timepoints)
for subsample in sample:
assert len(subsample) == len(variables)
for simulator in [trcrpm.simulate, trcrpm.simulate_ancestral]:
check_dims_correct([1], ['a'], 4, simulator)
check_dims_correct([1], ['a', 'b'], 7, simulator)
check_dims_correct([1, 2], ['a'], 1, simulator)
check_dims_correct([1, 2, 3], ['c', 'b'], 10, simulator)
check_dims_correct([9, 10, 11, 12], ['c', 'b'], 10, simulator)
def test_incorporate_sampleid_wedged():
rng = np.random.RandomState(1)
trcrpm = TRCRP_Mixture(chains=1, lag=2, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
# Now incorporate an non-contiguous row skipping last_sampled_id+1.
last_sampled_id = max(FRAME.index)
frame_new = pd.DataFrame([[11, 12, float('nan')]],
columns=['a', 'b', 'c'],
index=[last_sampled_id + 2])
trcrpm.incorporate(frame_new)
# Now bring back the missing observation at last_sampled_id+1.
# XXX The xfailure happens in this step.
frame_new = pd.DataFrame([[3, 3, 3]],
columns=['a', 'b', 'c'],
index=[last_sampled_id + 1])
trcrpm.incorporate(frame_new)
tabulated_data = trcrpm.engine.states[0].data_array()
expected_data = [
[nan, nan, 86., nan, nan, 57., nan, nan, 65.],
[nan, 86., 19., nan, 57., 62., nan, 65., nan],
[86., 19., 17., 57., 62., 41., 65., nan, 17.],
[19., 17., nan, 62., 41., 7., nan, 17., 30.],
[17., nan, 75., 41., 7., 1., 17., 30., 12.],
[nan, 75., 45., 7., 1., nan, 30., 12., 72.],
[75., 45., 48., 1., nan, 77., 12., 72., 8.],
[45., 48., 29., nan, 77., nan, 72., 8., 86.],
[48., 29., 83., 77., nan, 46., 8., 86., 38.],
[29., 83., nan, nan, 46., 54., 86., 38., nan],
[nan, 3., 11., 54., 3., 12., nan, 3., nan], # row updated.
[83., nan, 3., 46., 54., 3., 38., nan, 3.], # new row.
]
assert np.allclose(tabulated_data, expected_data, equal_nan=True)
def test_dependence_constraints():
rng = np.random.RandomState(2)
# All variables in TRCRP_Mixture are dependent.
trcrpm = TRCRP_Mixture(chains=1, lag=0, variables=FRAME.columns, rng=rng)
assert trcrpm._get_variable_dependence_constraints() == [[0, 1, 2]]
# All variables in TRCRP_Mixture are dependent.
trcrpm = TRCRP_Mixture(chains=1, lag=2, variables=FRAME.columns, rng=rng)
assert trcrpm._get_variable_dependence_constraints() == \
[[2,1,0, 5,4,3, 8,7,6]]
# All variables in TRCRP_Mixture are dependent.
trcrpm = TRCRP_Mixture(chains=1, lag=5, variables=FRAME.columns, rng=rng)
assert trcrpm._get_variable_dependence_constraints() == \
[[5,4,3,2,1,0, 11,10,9,8,7,6, 17,16,15,14,13,12]]
# Lag 0 dependencies should skip singleton 'c'.
trcrpm = Hierarchical_TRCRP_Mixture(chains=1,
lag=0,
variables=FRAME.columns,
rng=rng,
dependencies=[['a', 'b'], ['c']])
assert trcrpm._get_variable_dependence_constraints() == [[0, 1]]
# Lag 0 dependencies with all constrained.
trcrpm = Hierarchical_TRCRP_Mixture(chains=1,
lag=0,
variables=FRAME.columns,
rng=rng,
dependencies=[['a', 'b', 'c']])
assert trcrpm._get_variable_dependence_constraints() == [[0, 1, 2]]
# Lag 1 dependencies with two constraints.
trcrpm = Hierarchical_TRCRP_Mixture(chains=1,
lag=1,
variables=FRAME.columns,
rng=rng,
dependencies=[['a', 'b']])
assert trcrpm._get_variable_dependence_constraints() == [[5, 4],
[1, 0, 3, 2]]
# Lag 1 dependency with single constraint.
trcrpm = Hierarchical_TRCRP_Mixture(chains=1,
lag=1,
variables=FRAME.columns,
rng=rng,
dependencies=[['a']])
assert trcrpm._get_variable_dependence_constraints() == [[3, 2], [5, 4],
[1, 0]]
# Multiple customer dependencies. Capturing a runtime error since
# incorporate is going to use multiprocess so parallel_map captures the
# ValueError and throws a RuntimeError.
with pytest.raises(RuntimeError):
trcrpm = Hierarchical_TRCRP_Mixture(chains=1,
lag=0,
variables=FRAME.columns,
rng=rng,
dependencies=[['a', 'c'],
['a', 'b']])
trcrpm.incorporate(FRAME)
def test_get_cgpm_constraints():
rng = np.random.RandomState(2)
trcrpm = TRCRP_Mixture(chains=1, lag=3, variables=FRAME.columns, rng=rng)
trcrpm.incorporate(FRAME)
# For reference, lag=3 implies the following tabulation
# column a column b column c
# 0 1 2 3 4 5 6 7 8 9 10 11
# Incorporated timepoints should have no constraints.
for timepoint in trcrpm.dataset.index:
assert trcrpm._get_cgpm_constraints(timepoint) is None
# Fresh timepoint 10 should have appropriate constraints.
constraints_10 = trcrpm._get_cgpm_constraints(10)
assert constraints_10 == {
# column a column b column c
0: 29,
1: 83,
5: 46,
6: 54,
8: 86,
9: 38
}
# Incorporating data into timepoint 9 should update constraints for 10.
frame_new = pd.DataFrame([[-12, nan, -12]],
columns=['a', 'b', 'c'],
index=[9])
trcrpm.incorporate(frame_new)
constraints_10 = trcrpm._get_cgpm_constraints(10)
assert constraints_10 == {
# column a column b column c
0: 29,
1: 83,
2: -12,
5: 46,
6: 54,
8: 86,
9: 38,
10: -12
}
# Fresh timepoint 12 should have appropriate constraints.
constraints_12 = trcrpm._get_cgpm_constraints(12)
assert constraints_12 == {
# column a column b column c
0: -12,
4: 54,
8: -12
}
# Incorporating data into timepoint 11 should update constraints for 12.
frame_new = pd.DataFrame([[-44, -48, -47]],
columns=['a', 'b', 'c'],
index=[11])
trcrpm.incorporate(frame_new)
constraints_12 = trcrpm._get_cgpm_constraints(12)
assert constraints_12 == {
# column a column b column c
0: -12,
2: -44,
4: 54,
6: -48,
8: -12,
10: -47
}
from trcrpm import TRCRP_Mixture
from collections import Counter
print "f**k yeah"
data = pd.read_csv("./data/anomaly0245.csv", index_col=0)
data = data.iloc[156600:240000].reset_index(drop=True)
# Setup the placekeeping and initilizing variables
chain = 0
x, eng_val, states, num_states = [], [], [], []
i = 0
step = 30
print(i)
rng = np.random.RandomState(1)
model = TRCRP_Mixture(chains=1, lag=10, variables=data.columns, rng=rng)
model.incorporate(data[i:i + step])
model.resample_all(seconds=10)
model.resample_hyperparameters(seconds=10)
s = model.get_temporal_regimes('anomaly')[chain]
num_states = step * [len(sorted(set(s)))]
states = list(s[i:i + step])
eng_val = data.iloc[i:i + step, 0].tolist()
x = list(range(i, i + step))
for i in range(step, len(data) - step, step):
model.incorporate(data[i:i + step])
model.resample_all(seconds=10)
model.resample_hyperparameters(seconds=10)
s = model.get_temporal_regimes("anomaly")[chain]
num_states = step * [len(sorted(set(s)))]
states = list(s[i:i + step])
