def run_baseline(treebank_name, outdir=None, trainfile=None, testfile=None):
if not outdir: outdir= config.exp + treebank_name
malteval = Malteval()
TM = TreebankTransformer(treebank_name=treebank_name)
"""Train and parse"""
train_gold = TM.trainfile
test_gold = TM.testfile
TM._parser.train(train_gold)
parsed_baseline = outdir + '/dev_parsed_baseline.conll'
TM._parser.parse(test_gold, parsed_baseline)
"""RESULTS"""
uas, las= malteval.accuracy(test_gold,parsed_baseline)
output = "%s;%s;%s\n"%(treebank_name,las,uas)
return output
def plot_dep_scores(indir, treebank_number, dep, fig=None, outfile="Figures/figure.png"):
gold = indir + "test_gold.conll"
baseline = indir + "dev_parsed_baseline.conll"
transf = indir + "dev_parsed.ud.conll"
malteval = Malteval()
mb = malteval.deprel_matrix(gold, baseline)
mt = malteval.deprel_matrix(gold, transf)
"""The matplotlib thing"""
x = np.array([malt[6].strip("\n") for malt in mb])
p1 = np.array([float(malt[0]) if malt[0] is not "-" else 0.0 for malt in mb]) # sorry
p2 = np.array([float(malt[0]) if malt[0] is not "-" else 0.0 for malt in mt]) # not sorry
r1 = np.array([float(malt[1]) if malt[1] is not "-" else 0.0 for malt in mb])
r2 = np.array([float(malt[1]) if malt[1] is not "-" else 0.0 for malt in mt])
f1 = np.array([(2 * p * r / (p + r)) if 0.0 not in (p, r) else 0.0 for p, r in zip(p1, r1)])
f2 = np.array([(2 * p * r / (p + r)) if 0.0 not in (p, r) else 0.0 for p, r in zip(p2, r2)])
n1 = np.array([int(malt[2]) if malt[2] is not "-" else 0 for malt in mb])
n2 = np.array([int(malt[2]) if malt[2] is not "-" else 0 for malt in mt])
me1 = np.array([0.98 / sqrt(n) if n != 0 else 0.0 for n in n1])
me2 = np.array([0.98 / sqrt(n) if n != 0 else 0.0 for n in n2])
allinfo = [(xi, fi1, fi2, mei1, mei2) for (xi, fi1, fi2, mei1, mei2) in zip(x, f1, f2, me1, me2) if xi == dep]
x = [tup[0] for tup in allinfo]
f1 = [tup[1] for tup in allinfo]
f2 = [tup[2] for tup in allinfo]
me1 = [tup[3] for tup in allinfo]
me2 = [tup[4] for tup in allinfo]
if not fig:
fig, ax = plt.subplots()
else:
ax = fig.add_subplot(1, 1, 1)
if len(x) == 0:
return fig, ax
bar_width = 0.45
last_index = treebank_number * 3 * bar_width
index = last_index + np.arange(len(x))
error_config = {"ecolor": "0.3"}
ax.bar(
index,
f1,
bar_width,
color="darkblue",
error_kw=error_config,
yerr=me1,
label="baseline" if treebank_number == 0 else "",
)
ax.bar(
index + bar_width,
f2,
bar_width,
color="skyblue",
error_kw=error_config,
yerr=me2,
label="transformed" if treebank_number == 0 else "",
)
if treebank_number == 0:
plt.xlabel("%s deprel" % dep)
plt.ylabel("Attachment score")
plt.legend(bbox_to_anchor=(0.95, 0.92), bbox_transform=plt.gcf().transFigure)
return (fig, ax)