def test_simple_model_initialisation_1():
"""Test how the table counts are initialised in a simple model.
A simple model that has emitted one of each base from the empty context
must have one table for each base in the root context and no other tables."""
model = cacto.cactomodelfromseqs(('A', 'C', 'G', 'T'))
t = model.t.copy()
t[model.prefixindex.topdown().value.id] -= 1
assert (0 == t).all()
def test_simple_model_initialisation_1():
"""Test how the table counts are initialised in a simple model.
A simple model that has emitted one of each base from the empty context
must have one table for each base in the root context and no other tables."""
model = cacto.cactomodelfromseqs(('A', 'C', 'G', 'T'))
t = model.t.copy()
t[model.prefixindex.topdown().value.id] -= 1
assert (0 == t).all()
def test_seqs_log_likelihood():
for trainingseqs in prefix_seq_sets:
model = cacto.cactomodelfromseqs(trainingseqs)
for predictionseqs in prefix_seq_sets:
logging.info(
'likelihood/base: %.3f',
math.exp(
model.seqsloglikelihood(predictionseqs) /
sum(map(len, predictionseqs))))
def test_model_initialisation_1():
"""Test how the table counts are initialised."""
model = cacto.cactomodelfromseqs(('CGAT',))
seqan.traverse.depthfirsttraversal(model.prefixindex, model.log_table_counts)
t = model.t.copy()
i_cga = model.prefixindex.topdown()
if not i_cga.goDown('CGA'[::-1]):
raise ValueError('Should have been able to find prefix "CGA"')
assert ([0,0,0,1] == t[i_cga.value.id]).all()
def test_model_initialisation_1():
"""Test how the table counts are initialised."""
model = cacto.cactomodelfromseqs(('CGAT', ))
seqan.traverse.depthfirsttraversal(model.prefixindex,
model.log_table_counts)
t = model.t.copy()
i_cga = model.prefixindex.topdown()
if not i_cga.goDown('CGA'[::-1]):
raise ValueError('Should have been able to find prefix "CGA"')
assert ([0, 0, 0, 1] == t[i_cga.value.id]).all()
def _test_empty_model_predictions():
"""Currently dumps core due to seqan bug."""
seqs = tuple('', )
model = cacto.cactomodelfromseqs(seqs)
#
# No matter what the context we should see p = 1/4
#
for u in (
'',
'A',
'GC',
):
x = cacto.Value('A')
logger.info('p(%s|%s) = %.3e', x, u, model.p_x_given_u(x, u))
assert abs(.25 - model.p(x, u)) < 1e-15
def _test_empty_model_predictions():
"""Currently dumps core due to seqan bug."""
seqs = tuple('',)
model = cacto.cactomodelfromseqs(seqs)
#
# No matter what the context we should see p = 1/4
#
for u in (
'',
'A',
'GC',
):
x = cacto.Value('A')
logger.info('p(%s|%s) = %.3e', x, u, model.p_x_given_u(x, u))
assert abs(.25 - model.p(x, u)) < 1e-15
def test_simple_model_predictions():
seqs = (
'A',
'C',
'G',
'T',
)
model = cacto.cactomodelfromseqs(seqs)
#
# No matter what the context we should see p(x|u) = 1/4
#
for u in (
'',
'A',
'GC',
):
x = cacto.Value('A')
logger.info('p(%s|%s) = %.3e', x, u, model.p_x_given_u(x, u))
p = model.p_x_given_u(x, u)
if abs(.25 - model.p_x_given_u(x, u)) >= 1e-15:
raise ValueError('p not close to 1/4')
def test_simple_model_predictions():
seqs = (
'A',
'C',
'G',
'T',
)
model = cacto.cactomodelfromseqs(seqs)
#
# No matter what the context we should see p(x|u) = 1/4
#
for u in (
'',
'A',
'GC',
):
x = cacto.Value('A')
logger.info('p(%s|%s) = %.3e', x, u, model.p_x_given_u(x, u))
p = model.p_x_given_u(x, u)
if abs(.25 - model.p_x_given_u(x, u)) >= 1e-15:
raise ValueError('p not close to 1/4')
def test_model_predictions():
import seqan
for seqs, test_xs_us in prediction_sets:
model = cacto.cactomodelfromseqs(seqs)
posterior = model.calculateposterior()
for x, u in test_xs_us:
logging.debug('%s|%s', x, u)
p = model.p_x_given_u(cacto.Value(x), u)
i = model._locate_context(u, topdownhistory=True)
post = posterior[i.value.id]
p2 = model.p_xord_given_ui(cacto.Value(x).ordValue, i)
# Check that the three different methods of calculating likelihoods
# give similar results
assertareclose(p, p2)
assertareclose(p, post[cacto.Value(x).ordValue])
assertareclose(1., post.sum()) # Check posterior adds to 1
if False: # Choose whether to build graph or not
import seqan.io.graphtool
builder = seqan.io.graphtool.Builder(model.prefixindex)
seqan.io.graphtool.GT.graph_draw(
builder.graph,
pos=seqan.io.graphtool.GT.sfdp_layout(builder.graph),
vertex_size=2,
vertex_fill_color="lightgrey",
vertex_font_size=8,
vertex_text=builder.map_vertices(
lambda it: '{0} {1} {2} {3}'.format(*map(
int, model._su(it)))),
vertex_pen_width=seqan.io.graphtool.root_vertex_property(
builder),
edge_text=seqan.io.graphtool.edge_labels_for_output(builder),
edge_color=seqan.io.graphtool.color_edges_by_first_symbol(
builder),
edge_end_marker="none",
edge_pen_width=2,
#edge_dash_style=seqan.io.graphtool.dash_non_suffix_edges(builder, suffix),
#edge_pen_width=builder.edge_lengths,
#output="graphtool.png"
)
def test_model_predictions():
import seqan
for seqs, test_xs_us in prediction_sets:
model = cacto.cactomodelfromseqs(seqs)
posterior = model.calculateposterior()
for x, u in test_xs_us:
logging.debug('%s|%s', x, u)
p = model.p_x_given_u(cacto.Value(x), u)
i = model._locate_context(u, topdownhistory=True)
post = posterior[i.value.id]
p2 = model.p_xord_given_ui(cacto.Value(x).ordValue, i)
# Check that the three different methods of calculating likelihoods
# give similar results
assertareclose(p, p2)
assertareclose(p, post[cacto.Value(x).ordValue])
assertareclose(1., post.sum()) # Check posterior adds to 1
if False: # Choose whether to build graph or not
import seqan.io.graphtool
builder = seqan.io.graphtool.Builder(model.prefixindex)
seqan.io.graphtool.GT.graph_draw(
builder.graph,
pos=seqan.io.graphtool.GT.sfdp_layout(builder.graph),
vertex_size=2,
vertex_fill_color="lightgrey",
vertex_font_size=8,
vertex_text=builder.map_vertices(
lambda it: '{0} {1} {2} {3}'.format(*map(int, model._su(it)))),
vertex_pen_width=seqan.io.graphtool.root_vertex_property(builder),
edge_text=seqan.io.graphtool.edge_labels_for_output(builder),
edge_color=seqan.io.graphtool.color_edges_by_first_symbol(builder),
edge_end_marker="none",
edge_pen_width=2,
#edge_dash_style=seqan.io.graphtool.dash_non_suffix_edges(builder, suffix),
#edge_pen_width=builder.edge_lengths,
#output="graphtool.png"
)
def test_seqs_log_likelihood():
for trainingseqs in prefix_seq_sets:
model = cacto.cactomodelfromseqs(trainingseqs)
for predictionseqs in prefix_seq_sets:
logging.info('likelihood/base: %.3f',
math.exp(model.seqsloglikelihood(predictionseqs)/sum(map(len, predictionseqs))))