diff --git a/evaluate_baseline.py b/evaluate_baseline.py
index 2058636fabf79c4f71c59ef26f7eeb3131413e2b..ed07554f29b68a7d2d5f21546c28acf5bff49117 100644
--- a/evaluate_baseline.py
+++ b/evaluate_baseline.py
@@ -1,6 +1,6 @@
from joblib import load
import pandas as pd
-from lib.utils_geo import haversine_pd,latlon2healpix
+from lib.utils_geo import latlon2healpix
import argparse
import os
diff --git a/evaluate_geocoder.py b/evaluate_geocoder.py
index e57c2b3013c58e5049f47256fb0cd4fcc9331238..a47a68a746ed9bdeeec0b666ca36ea0f7990f81e 100644
--- a/evaluate_geocoder.py
+++ b/evaluate_geocoder.py
@@ -45,13 +45,20 @@ def accuracy_at_k(geocoding_df,geocoder,k=100):
geocoding_df["distanceKM"] = haversine_pd(geocoding_df.longitude,geocoding_df.latitude,geocoding_df.pred_longitude,geocoding_df.pred_latitude)
return (geocoding_df.distanceKM <k).sum()/len(geocoding_df)
+def median_distance_error(geocoding_df,geocoder):
+ lons,lats = g.get_coords(geocoding_df.toponym.values,geocoding_df.toponym_context.values)
+ geocoding_df["pred_latitude"] = lats
+ geocoding_df["pred_longitude"] = lons
+ geocoding_df["distanceKM"] = haversine_pd(geocoding_df.longitude,geocoding_df.latitude,geocoding_df.pred_longitude,geocoding_df.pred_latitude)
+ return geocoding_df.distanceKM.median()
+
res_ = []
for mod in tqdm(model_available):
index_fn = MODELS_DIR + mod +"_index"
model_fn = MODELS_DIR + mod +".h5"
g = Geocoder(model_fn, index_fn)
- res_.append([mod,accuracy_at_k(geocoding_df,g,100),accuracy_at_k(geocoding_df,g,50),accuracy_at_k(geocoding_df,g,20)])
+ res_.append([mod,accuracy_at_k(geocoding_df,g,161),accuracy_at_k(geocoding_df,g,100),accuracy_at_k(geocoding_df,g,50),accuracy_at_k(geocoding_df,g,20),median_distance_error(geocoding_df,g)])
-pd.DataFrame(res_,columns="dataset accuracy@100km accuracy@50km accuracy@20km".split()).to_csv(OUTPUT_FN,sep="\t",index=None)
+pd.DataFrame(res_,columns="dataset accuracy@161km accuracy@100km accuracy@50km accuracy@20km MDE".split()).to_csv(OUTPUT_FN,sep="\t",index=None)
diff --git a/generate_dataset.py b/generate_dataset.py
index 90224c8997d3b3ab370ff29e5e5d8ee05cd96be1..39e688cb186ffb1c0fc027ccf42e0d1376b4015f 100644
--- a/generate_dataset.py
+++ b/generate_dataset.py
@@ -3,7 +3,7 @@ import argparse
import pandas as pd
import numpy as np
-from helpers import read_geonames
+from lib.helpers import read_geonames
from lib.utils_geo import latlon2healpix
from tqdm import tqdm
@@ -166,6 +166,10 @@ print('# cooc_pairs: ', len(cooc_pairs))
print('# adjacent_pairs: ', len(adjacent_pairs))
print('# inclusion_pairs: ', len(inclusion_pairs))
+#ADD metadata
+cooc_pairs["sampling"] = args.cooc_sampling
+adjacent_pairs["sampling"] = args.adj_sampling
+
# SAVE DATA
inclusion_pairs.to_csv("{0}_inclusion.csv".format(PREFIX), sep="\t")
adjacent_pairs.to_csv("{0}_adjacent.csv".format(PREFIX), sep="\t")
diff --git a/helpers.py b/lib/helpers.py
similarity index 90%
rename from helpers.py
rename to lib/helpers.py
index b6cf164a1d711d7968010f4d126b3ea1088d17de..ae19955f450a2bed75a3e8bcd2e56f8560dacfe4 100644
--- a/helpers.py
+++ b/lib/helpers.py
@@ -177,7 +177,7 @@ import time
import os
class EpochTimer(Callback):
- def __init__(self,log_filename):
+ def __init__(self,log_filename,is_p,is_c,is_i,sampling_p,sampling_c):
self.epoch = 0
self.timer = time.time()
# if os.path.exists(log_filename):
@@ -185,7 +185,9 @@ class EpochTimer(Callback):
# self.epoch = pd.read_csv(log_filename).Epoch.max()
# else:
self.output = open(log_filename,'a')
- self.output.write("{0},{1},{2},{3},{4},{5}\n".format("Epoch","Execution Time","Loss","Val_Loss","Accuracy","Accuracy_val"))
+ header_config = "is_P,is_C,is_I"
+ self.config_model = ",".join(list(map(str,[is_p,is_c,is_i,sampling_p,sampling_c])))
+ self.output.write("{0},{1},{2},{3},{4},{5},{6}\n".format("Epoch","Execution Time","Loss","Val_Loss","Accuracy","Accuracy_val",header_config))
self.output.flush()
def on_epoch_begin(self,epoch, logs={}):
@@ -193,7 +195,7 @@ class EpochTimer(Callback):
def on_epoch_end(self, epoch, logs={}):
end_time = time.time() - self.timer
- self.output.write("{0},{1},{2},{3},{4},{5}\n".format(self.epoch,end_time,logs["loss"],logs["val_loss"],logs["compute_metric"],logs["val_compute_metric"]))
+ self.output.write("{0},{1},{2},{3},{4},{5},{6}\n".format(self.epoch,end_time,logs["loss"],logs["val_loss"],logs["compute_metric"],logs["val_compute_metric"],self.config_model))
self.output.flush()
self.epoch += 1
diff --git a/lib/utils_geo.py b/lib/utils_geo.py
index 00b8a1a2fba238cd5b01ad3710a0dd8b4f2aec8d..3ee3833b441f3939b9b7b47d75f8a54fbdb9b404 100644
--- a/lib/utils_geo.py
+++ b/lib/utils_geo.py
@@ -11,7 +11,7 @@ from tqdm import tqdm
import pandas as pd, numpy as np
from numba import njit
-from helpers import read_geonames
+from .helpers import read_geonames
from tqdm import tqdm
from joblib import Parallel,delayed
diff --git a/train_baseline.py b/train_baseline.py
index 653c326902d4ea92f457fb8ec74f825cc4b6b6b3..f3609727c303f50ba359601a830f9c87f69bb962 100644
--- a/train_baseline.py
+++ b/train_baseline.py
@@ -69,7 +69,7 @@ classifier_dict = {
parameters = {
"naive-bayes":[{"alpha":[0,1]}],
- "svm":[{"kernel":["rbf","poly"], 'gamma': [1e-1,1e-2,1e-3, 1,10,100]}],
+ "svm":[{"kernel":["linear","rbf","poly"], 'gamma': [1e-1,1e-2,1e-3, 1,10,100]}],
"sgd":[{"penalty":["l1","l2"],"loss":["hinge","modified_huber","log"]}],
"knn":[{"n_neighbors":list(range(4,8)),"p":[1,2]}],
"decision-tree": [{"criterion":["gini","entropy"]}],
diff --git a/train_geocoder.py b/train_geocoder.py
index eaa5193b252bfaf2c3e5b598e24d9ab66378ea72..a89d2988e17ba37312786009008b4e5194ad3830 100644
--- a/train_geocoder.py
+++ b/train_geocoder.py
@@ -17,7 +17,7 @@ from lib.ngram_index import NgramIndex
from lib.word_index import WordIndex
from lib.utils import ConfigurationReader
from lib.utils_geo import accuracy_k,haversine_tf_1circle
-from helpers import EpochTimer
+from lib.helpers import EpochTimer
from lib.datageneratorv4 import DataGenerator
# Logging
@@ -66,11 +66,11 @@ EMBEDDING_FN = "outputs/{0}_embedding.npy".format(PREFIX_OUTPUT_FN)
PREFIX_OUTPUT_FN+="_{0}".format(EPOCHS)
if args.adjacency:
- PREFIX_OUTPUT_FN += "_A"
+ PREFIX_OUTPUT_FN += "_P"
if args.inclusion:
PREFIX_OUTPUT_FN += "_I"
if args.wikipedia:
- PREFIX_OUTPUT_FN += "_P"
+ PREFIX_OUTPUT_FN += "_C"
MODEL_OUTPUT_FN = "outputs/{0}.h5".format(PREFIX_OUTPUT_FN)
INDEX_FN = "outputs/{0}_index".format(PREFIX_OUTPUT_FN)
@@ -82,15 +82,20 @@ HISTORY_FN = "outputs/{0}.csv".format(PREFIX_OUTPUT_FN)
#############################################################################################
data_used = []
-
+sampling_adj = None
+sampling_cooc = None
if args.wikipedia:
data_used.append(pd.read_csv(COOC_FN,sep="\t"))
+ if "sampling" in data_used[-1].columns:
+ sampling_cooc = data_used[-1]["sampling"].unique()[0]
if args.inclusion:
data_used.append(pd.read_csv(INCLUSION_FN,sep="\t"))
if args.adjacency:
data_used.append(pd.read_csv(ADJACENT_FN, sep="\t"))
+ if "sampling" in data_used[-1].columns:
+ sampling_adj = data_used[-1]["sampling"].unique()[0]
if len(data_used) <1:
print("No Type of toponyms indicated. Stopping the program...")
@@ -205,8 +210,8 @@ print(model.summary())
checkpoint = ModelCheckpoint(MODEL_OUTPUT_FN , monitor='loss', verbose=1,
save_best_only=save_best_only, mode='auto', period=1)
-epoch_timer = EpochTimer(HISTORY_FN)
+epoch_timer = EpochTimer(HISTORY_FN,is_p=args.adjacency,is_c=args.wikipedia,is_i=args.inclusion,sampling_c=sampling_cooc,sampling_p=sampling_adj)
history = model.fit(training_generator,verbose=True,
diff --git a/wikipediageocoding.ipynb b/wikipediageocoding.ipynb
index 91dc00b7003a06f915635ab958208601c9022a9c..4b1febdf3d94f8b76be5a8c2338dfb51c0ecdeaa 100644
--- a/wikipediageocoding.ipynb
+++ b/wikipediageocoding.ipynb
@@ -37,6 +37,13 @@
"model_ngram_index_path = \"data/FR20_cooc_4_30_P_index.txt\""
]
},
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {},
+ "outputs": [],
+ "source": []
+ },
{
"cell_type": "code",
"execution_count": 3,