def build_extrinsic_tables(target, source, env):
files = source[0].read()
rows = []
for (language, pack), setups in files.iteritems():
for setup, (asr_fname, kws_fname) in setups.iteritems():
with meta_open(asr_fname) as asr_fd, meta_open(kws_fname) as kws_fd:
asr = ASRResults(asr_fd)
kws = KWSResults(kws_fd)
rows.append([language, setup] + [asr.get(x) for x in ["error", "substitutions", "deletions", "insertions"]] + [kws.get(x) for x in ["pmiss", "mtwv"]])
with meta_open(target[0].rstr(), "w") as ofd:
body = "\n".join([r" & ".join([str(x) for x in row]) + r" \\" for row in rows])
ofd.write(r"""
\begin{tabular}{|*{2}{l|}*{6}{r|}}
\hline
Language & Augmentation & \multicolumn{4}{|c|}{ASR} & \multicolumn{2}{|c|}{KWS} \\
& & Errors & Subs & Dels & Ins & PMiss & MTWV \\
\hline
%s
\hline
\end{tabular}
""" % (body))
return None
def build_site(target, source, env):
properties, figures, results = [x.read() for x in source[0:3]]
languages = set([x[0] for x in figures.keys()])
lookup = {"PRE" : "Prefixes",
"STM" : "Stems",
"SUF" : "Suffixes",
}
packs = ["Limited"]
base_path = os.path.dirname(target[0].rstr())
try:
os.makedirs(pjoin(base_path, "images"))
except:
pass
with meta_open(pjoin(base_path, "theme.css"), "w") as ofd:
ofd.write("body {text-align : center; vertical-align : top;}\n")
ofd.write("table {text-align : center; vertical-align : top;}\n")
ofd.write("tr {text-align : center; vertical-align : top;}\n")
ofd.write("td {text-align : center; vertical-align : top;}\n")
with meta_open(target[0].rstr(), "w") as ofd:
xml = et.TreeBuilder()
xml.start("html", {})
xml.start("head", {}), xml.start("link", {"rel" : "stylesheet", "type" : "text/css", "href" : "theme.css"}), xml.end("link"), xml.end("head")
xml.start("body", {}), xml.start("table", {})
for language in languages:
language_properties = properties[(language, "Limited")]
with meta_open(language_properties["prefixes"]) as prefix_fd, meta_open(language_properties["stems"]) as stem_fd, meta_open(language_properties["suffixes"]) as suffix_fd:
babel_output = {"PRE" : [l.strip().split()[0] for l in prefix_fd if "<epsilon>" not in l],
"STM" : [l.strip().split()[0] for l in stem_fd if "<epsilon>" not in l],
"SUF" : [l.strip().split()[0] for l in suffix_fd if "<epsilon>" not in l],
}
with meta_open(language_properties["limited_vocabulary"]) as lim_fd, meta_open(language_properties["dev_vocabulary"]) as dev_fd:
lim_vocab = set(FrequencyList(lim_fd).make_conservative().keys())
dev_vocab = set(FrequencyList(dev_fd).make_conservative().keys())
lim_vocab_size = len(lim_vocab)
dev_vocab_size = len(dev_vocab)
both_vocabs = len(lim_vocab.union(dev_vocab))
avg_len_lim_vocab = sum(map(len, lim_vocab)) / float(len(lim_vocab))
avg_len_dev_vocab = sum(map(len, dev_vocab)) / float(len(dev_vocab))
with meta_open(language_properties["morfessor_input"]) as input_fd, meta_open(language_properties["morfessor_output"]) as output_fd:
input_vocab = FrequencyList({w : int(c) for c, w in [x.strip().split() for x in input_fd]})
morf_output = MorfessorOutput(output_fd)
morf_analysis_counts = len(morf_output)
morf_morph_counts = {k : len(v) for k, v in morf_output.morphs.iteritems()}
morf_morph_lengths = {k : sum(map(len, v)) / float(len(v)) for k, v in morf_output.morphs.iteritems() if len(v) > 0}
xml.start("tr", {}), xml.start("td", {}), xml.start("h1", {}), xml.data(language.title()), xml.end("h1"), xml.end("td"), xml.end("tr")
xml.start("table", {})
xml.start("tr", {}), [(xml.start("td", {}), xml.start("h3", {}), xml.data("%s information" % (x)), xml.end("h3"), xml.end("td")) for x in ["Language", "Morfessor", "BabelGUM"]], xml.end("tr")
xml.start("tr", {})
# Language information
xml.start("td", {})
xml.start("table", {"border" : "1"})
#xml.start("tr", {}), [(xml.start("td", {}), xml.data(x), xml.end("td")) for x in ["Pack", "Vocabulary size"]], xml.end("tr")
xml.start("tr", {}), xml.start("td", {}), xml.end("td"), xml.start("td", {}), xml.data("Count"), xml.end("td"), xml.start("td", {}), xml.data("Average length"), xml.end("td"), xml.end("tr")
xml.start("tr", {}), xml.start("td", {}), xml.data("Limited vocab"), xml.end("td"), xml.start("td", {}), xml.data("%d" % (lim_vocab_size)), xml.end("td"), xml.start("td", {}), xml.data("%.2f" % (avg_len_lim_vocab)), xml.end("td"), xml.end("tr")
xml.start("tr", {}), xml.start("td", {}), xml.data("Dev vocab"), xml.end("td"), xml.start("td", {}), xml.data("%d" % (dev_vocab_size)), xml.end("td"), xml.start("td", {}), xml.data("%.2f" % (avg_len_dev_vocab)), xml.end("td"), xml.end("tr")
#for name, values in [(lookup[x[0]], x[1]) for x in sorted(morf_output.morphs.iteritems())]:
# xml.start("tr", {})
#xml.start("td", {}), xml.data(name), xml.end("td")
#xml.start("td", {}), xml.data(str(len(values))), xml.end("td")
#xml.start("td", {}), xml.data(avg_len), xml.end("td")
# xml.end("tr")
xml.end("table")
xml.end("td")
# Morfessor information
xml.start("td", {})
xml.start("table", {"border" : "1"})
xml.start("tr", {}), [(xml.start("td", {}), xml.data(x), xml.end("td")) for x in ["Type", "Count", "Average length"]], xml.end("tr")
for name, values in [(lookup[x[0]], x[1]) for x in sorted(morf_output.morphs.iteritems())]:
if len(values) > 0:
avg_len = "%.2f" % (sum(map(len, values)) / float(len(values)))
else:
avg_len = ""
xml.start("tr", {})
xml.start("td", {}), xml.data(name), xml.end("td")
xml.start("td", {}), xml.data(str(len(values))), xml.end("td")
xml.start("td", {}), xml.data(avg_len), xml.end("td")
xml.end("tr")
xml.end("table")
xml.end("td")
# BabelGUM information
xml.start("td", {})
xml.start("table", {"border" : "1"})
xml.start("tr", {}), [(xml.start("td", {}), xml.data(x), xml.end("td")) for x in ["Type", "Count", "Average length"]], xml.end("tr")
for name, values in [(lookup[x[0]], x[1]) for x in sorted(babel_output.iteritems())]:
if len(values) > 0:
avg_len = "%.2f" % (sum(map(len, values)) / float(len(values)))
else:
avg_len = ""
xml.start("tr", {})
xml.start("td", {}), xml.data(name), xml.end("td")
xml.start("td", {}), xml.data(str(len(values))), xml.end("td")
xml.start("td", {}), xml.data(avg_len), xml.end("td")
xml.end("tr")
xml.end("table")
xml.end("td")
xml.end("tr")
xml.end("table")
# graphs of IV increase and OOV reduction, type-based and token-based
xml.start("tr", {}), xml.start("td", {}), xml.start("h3", {}), xml.data("Intrinsic performance evaluation"), xml.end("h3"), xml.end("td"), xml.end("tr")
xml.start("tr", {}), xml.start("td", {})
xml.start("table", {})
for pack in packs:
image_file = "%s_%s.png" % (language, pack)
shutil.copy(figures[(language, pack)], pjoin(base_path, "images", image_file))
xml.start("tr", {}), xml.start("td", {}), xml.start("img", {"src" : pjoin("images", image_file)}), xml.end("img"), xml.end("td"), xml.end("tr")
xml.end("table")
xml.end("td"), xml.end("tr")
# word error rate for ASR and maximum term-weighted value for KWS
xml.start("tr", {}), xml.start("td", {}), xml.start("h3", {}), xml.data("Extrinsic performance evaluation"), xml.end("h3"), xml.end("td"), xml.end("tr")
xml.start("tr", {}), xml.start("td", {}), xml.start("table", {"border" : "1"})
xml.start("tr", {}), [(xml.start("td", {}), xml.data(x), xml.end("td")) for x in ["Augmentation", "Error", "Substitutions", "Deletions", "Insertions", "PMiss", "MTWV"]], xml.end("tr")
for name, values in sorted(results[(language, "Limited")].iteritems()):
with meta_open(values["ASR"]) as asr_fd, meta_open(values["KWS"]) as kws_fd:
asr = ASRResults(asr_fd)
kws = KWSResults(kws_fd)
xml.start("tr", {})
xml.start("td", {}), xml.data(name), xml.end("td")
[(xml.start("td", {}), xml.data("%.3f" % (asr.get(x))), xml.end("td")) for x in ["error", "substitutions", "deletions", "insertions"]]
[(xml.start("td", {}), xml.data("%.3f" % (kws.get(x))), xml.end("td")) for x in ["pmiss", "mtwv"]]
xml.end("tr")
xml.end("table")
xml.end("td"), xml.end("tr")
xml.end("table"), xml.end("body")
xml.end("html")
ofd.write(et.tostring(xml.close()))
return None
