def evaluate():
fw = open("../result/eval.results", "w")
flag = True
counter = 0
for df in data_files:
counter += 1
df = os.path.join(data_dir, df)
jstr = open(df).read()
res_dict = proxy.geolocate_cli(jstr, True)
err = res_dict["error"]
if err:
print err
continue
sname = res_dict["sname"].lower()
oconf = res_dict["oconf"]
# here to ensure 10 geotagged tweets, and 50% hurdle
if oconf != 2:
continue
pc = res_dict["pc"]
oc = res_dict["oc"]
plat, plon = lib_grid_search.lookup_coords(pc)
olat, olon = lib_grid_search.lookup_coords(oc)
error = gcd_dist.calc_dist_degree(plat, plon, olat, olon)
error_list.append(error)
if oc[-2:] == pc[-2:]:
cc_list.append(True)
text_pred = res_dict["text_pred"]
loc_pred = res_dict["loc_pred"]
tz_pred = res_dict["tz_pred"]
tweet_num = len(res_dict["tweets"])
rec = "{0}\t{1}\t{2}\t{3}\t{4}\t{5}\t{6}\t{7}\n".format(
sname, oc, pc, text_pred, loc_pred, tz_pred, error, tweet_num)
print counter, rec,
fw.write(rec)
divisor = float(len(error_list))
acc = len([error for error in error_list if error == 0.0]) / divisor
cacc = sum(1.0 for c in cc_list if c) / divisor
acc161 = len([error for error in error_list if error <= 161.0]) / divisor
mean = sum(error_list) / divisor
median = sorted(error_list)[int(divisor) / 2]
fw.write("Acc: {0}\n".format(acc))
print "Acc: {0}\n".format(acc),
fw.write("Acc161: {0}\n".format(acc161))
print "Acc161: {0}\n".format(acc161),
fw.write("AccCountry: {0}\n".format(cacc))
print "AccCountry: {0}\n".format(cacc),
fw.write("Median: {0}\n".format(median))
print "Median: {0}\n".format(median),
fw.write("Mean: {0}\n".format(mean))
print "Mean: {0}\n".format(mean),
fw.close()
print "All finished"
def evaluate():
fw = open("../result/eval.results", "w")
flag = True
counter = 0
for df in data_files:
counter += 1
df = os.path.join(data_dir, df)
jstr = open(df).read()
res_dict = proxy.geolocate_cli(jstr, True)
err = res_dict["error"]
if err:
print err
continue
sname = res_dict["sname"].lower()
oconf = res_dict["oconf"]
# here to ensure 10 geotagged tweets, and 50% hurdle
if oconf != 2:
continue
pc = res_dict["pc"]
oc = res_dict["oc"]
plat, plon = lib_grid_search.lookup_coords(pc)
olat, olon = lib_grid_search.lookup_coords(oc)
error = gcd_dist.calc_dist_degree(plat, plon, olat, olon)
error_list.append(error)
if oc[-2:] == pc[-2:]:
cc_list.append(True)
text_pred = res_dict["text_pred"]
loc_pred = res_dict["loc_pred"]
tz_pred = res_dict["tz_pred"]
tweet_num = len(res_dict["tweets"])
rec = "{0}\t{1}\t{2}\t{3}\t{4}\t{5}\t{6}\t{7}\n".format(sname, oc, pc, text_pred, loc_pred, tz_pred, error, tweet_num)
print counter, rec,
fw.write(rec)
divisor = float(len(error_list))
acc = len([error for error in error_list if error == 0.0]) / divisor
cacc = sum(1.0 for c in cc_list if c) / divisor
acc161 = len([error for error in error_list if error <= 161.0]) / divisor
mean = sum(error_list) / divisor
median = sorted(error_list)[int(divisor)/2]
fw.write("Acc: {0}\n".format(acc))
print "Acc: {0}\n".format(acc),
fw.write("Acc161: {0}\n".format(acc161))
print "Acc161: {0}\n".format(acc161),
fw.write("AccCountry: {0}\n".format(cacc))
print "AccCountry: {0}\n".format(cacc),
fw.write("Median: {0}\n".format(median))
print "Median: {0}\n".format(median),
fw.write("Mean: {0}\n".format(mean))
print "Mean: {0}\n".format(mean),
fw.close()
print "All finished"
def geolocate(data, enable_cache=True):
"""
Input:
1. user screen name
2. user timeline JSON data (assuming the same user name)
Ouptut:
GT-dict
Supported Options: caching;
"""
# identify input format
sname = None #user screen name
gt_dict = None #JSON format result
err_msg = None
parsed_data, err_msg = parse_input(data)
if isinstance(parsed_data, basestring):
sname = parsed_data
elif isinstance(parsed_data, dict):
gt_dict = parsed_data
sname = parsed_data["sname"]
else:
gt_dict = {"error": err_msg}
err_msg = "{0} {1}".format(err_msg, data)
lib_log.error(err_msg)
return gt_dict
sname = sname.lower()
print "Predict:", sname
# using cache?
if enable_cache:
cached = seek_cache(sname)
if cached:
return cached
# crawl data if input is user name
if not gt_dict:
gt_dict, err_msg = twitter_adapter.parse_user_timeline(sname)
if err_msg:
gt_dict = {"error": err_msg}
err_msg = "{0} {1}".format(err_msg, sname)
lib_log.error(err_msg)
return gt_dict
# sequential classifier
text_pred = text_decoder.predict(
feature_adapter.extract_text_features(gt_dict["tweets"]))
loc_pred = loc_decoder.predict(
feature_adapter.extract_ngram_features(gt_dict["loc"]))
tz_pred = tz_decoder.predict(
feature_adapter.extract_ngram_features(gt_dict["tz"]))
#desc_pred = desc_decoder.predict(feature_adapter.extract_ngram_features(gt_dict["desc"]))
#rname_pred = rname_decoder.predict(feature_adapter.extract_ngram_features(gt_dict["rname"]))
print "L0 predictions:", text_pred, loc_pred, tz_pred
#print text_pred, loc_pred, tz_pred, desc_pred, rname_pred
text_id = city_adapter.get_id_by_city(text_pred) + fea_num * 0
loc_id = city_adapter.get_id_by_city(loc_pred) + fea_num * 1
tz_id = city_adapter.get_id_by_city(tz_pred) + fea_num * 2
#desc_id = city_adapter.get_id_by_city(desc_pred) + fea_num * 3
#rname_id = city_adapter.get_id_by_city(rname_pred) + fea_num * 4
#libsvm_rec = "{0}:1 {1}:1 {2}:1\n".format(text_id, loc_id, tz_id)
#libsvm_rec = "{0}:1 {1}:1 {2}:1 {3}:1 {4}:1\n".format(text_id, loc_id, tz_id, desc_id, rname_id)
#print libsvm_rec
fea_vec_liblinear = [{text_id: 1, loc_id: 1, tz_id: 1}]
#fea_vec_liblinear = [{text_id:1, loc_id:1, tz_id:1, desc_id:1, rname_id:1}]
print "L1 data:", fea_vec_liblinear
p_label, p_acc, p_val = liblinearutil.predict([1], fea_vec_liblinear,
stacked_model)
slr_id = int(p_label[0]) # stacked logistic regression prediction label
slr_pred = city_adapter.get_city_by_id(slr_id)
print "L1 prediction:", slr_pred
gt_dict["pc"] = slr_pred
slr_lat, slr_lon = lib_grid_search.lookup_coords(slr_pred)
print "L1 prediction coordinates:", slr_lat, slr_lon
gt_dict["plat"] = slr_lat
gt_dict["plon"] = slr_lon
gt_dict["errdist"] = int(
gcd_dist.calc_dist_degree(slr_lat, slr_lon, gt_dict["olat"],
gt_dict["olon"])) if gt_dict["oc"] else None
assert (not err_msg)
gt_dict["error"] = err_msg
gt_dict["text_pred"] = text_pred
gt_dict["loc_pred"] = loc_pred
gt_dict["tz_pred"] = tz_pred
if enable_cache:
open("{0}/cache/{1}".format(pkg_path, sname),
"w").write("{0}\n".format(json.dumps(gt_dict)))
return gt_dict
def geolocate(data, enable_cache = True):
"""
Input:
1. user screen name
2. user timeline JSON data (assuming the same user name)
Ouptut:
GT-dict
"""
# identify input format
sname = None #user screen name
gt_dict = None #JSON format result
err_msg = None
parsed_data, err_msg = parse_input(data)
if isinstance(parsed_data, basestring):
sname = parsed_data
elif isinstance(parsed_data, dict):
gt_dict = parsed_data
sname = parsed_data["sname"]
else:
gt_dict = {"error":err_msg}
err_msg = "{0} {1}".format(err_msg, data)
lib_log.error(err_msg)
return gt_dict
sname = sname.lower()
print "Predict:", sname
# crawl data if input is user name
if not gt_dict:
gt_dict, err_msg = twitter_adapter.parse_user_timeline(sname)
if err_msg:
gt_dict = {"error":err_msg}
err_msg = "{0} {1}".format(err_msg, sname)
lib_log.error(err_msg)
return gt_dict
# sequential classifier
text_features = feature_adapter.extract_text_features(gt_dict["tweets"])
text_pred = text_decoder.predict(text_features)
loc_pred = loc_decoder.predict(feature_adapter.extract_ngram_features(gt_dict["loc"]))
tz_pred = tz_decoder.predict(feature_adapter.extract_ngram_features(gt_dict["tz"]))
print "L0 predictions:", text_pred, loc_pred, tz_pred
#print text_pred, loc_pred, tz_pred, desc_pred, rname_pred
text_id = city_adapter.get_id_by_city(text_pred) + fea_num * 0
loc_id = city_adapter.get_id_by_city(loc_pred) + fea_num * 1
tz_id = city_adapter.get_id_by_city(tz_pred) + fea_num * 2
fea_vec_liblinear = [{text_id:1, loc_id:1, tz_id:1}]
print "L1 data:", fea_vec_liblinear
p_label, p_acc, p_val = liblinearutil.predict([1], fea_vec_liblinear, stacked_model)
slr_id = int(p_label[0]) # stacked logistic regression prediction label
slr_pred = city_adapter.get_city_by_id(slr_id)
print "L1 prediction:", slr_pred
gt_dict["pc"] = slr_pred
slr_lat, slr_lon = lib_grid_search.lookup_coords(slr_pred)
print "L1 prediction coordinates:", slr_lat, slr_lon
gt_dict["liw"] = text_features
gt_dict["plat"] = slr_lat
gt_dict["plon"] = slr_lon
gt_dict["errdist"] = int(gcd_dist.calc_dist_degree(slr_lat, slr_lon, gt_dict["olat"], gt_dict["olon"])) if gt_dict["oc"] else None
assert(not err_msg)
gt_dict["error"] = err_msg
gt_dict["text_pred"] = text_pred
gt_dict["loc_pred"] = loc_pred
gt_dict["tz_pred"] = tz_pred
if enable_cache:
open("{0}/cache/{1}".format(pkg_path, sname), "w").write("{0}\n".format(json.dumps(gt_dict)))
return gt_dict
