updated lb metrics
Browse files- leaderboard.py +47 -3
leaderboard.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
from glob import glob
|
2 |
-
from sklearn.metrics import accuracy_score, recall_score
|
3 |
import os
|
4 |
import pandas as pd
|
5 |
|
@@ -56,7 +56,7 @@ def get_duration_scores(df):
|
|
56 |
lb = pd.DataFrame({"Sample": columns, "Num Samples": samples_tested, "Accuracy": acc_scores})
|
57 |
return lb
|
58 |
|
59 |
-
def
|
60 |
|
61 |
columns = list(df[df.label != 'real'].algorithm.unique())
|
62 |
samples_tested = []
|
@@ -75,8 +75,52 @@ def get_algorithm_scores(df):
|
|
75 |
lb = pd.DataFrame({"Sample": columns, "Num Samples": samples_tested, "Recall": rec_scores})
|
76 |
return lb
|
77 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
def build_leaderboard(results_path = 'results'):
|
79 |
full_df = get_merged_df(results_path)
|
80 |
full_df_mapped = map_df(full_df)
|
81 |
-
leaderboard =
|
82 |
return leaderboard
|
|
|
1 |
from glob import glob
|
2 |
+
from sklearn.metrics import accuracy_score, recall_score, f1_score
|
3 |
import os
|
4 |
import pandas as pd
|
5 |
|
|
|
56 |
lb = pd.DataFrame({"Sample": columns, "Num Samples": samples_tested, "Accuracy": acc_scores})
|
57 |
return lb
|
58 |
|
59 |
+
def get_algorithm_scores_v1(df):
|
60 |
|
61 |
columns = list(df[df.label != 'real'].algorithm.unique())
|
62 |
samples_tested = []
|
|
|
75 |
lb = pd.DataFrame({"Sample": columns, "Num Samples": samples_tested, "Recall": rec_scores})
|
76 |
return lb
|
77 |
|
78 |
+
def get_algorithm_scores_v2(df):
|
79 |
+
|
80 |
+
columns = list(df[df.label != 'real'].algorithm.unique())
|
81 |
+
columns2 = list(df[df.label != 'real'].label.unique())
|
82 |
+
samples_tested = []
|
83 |
+
acc_scores = []
|
84 |
+
tpr_scores = []
|
85 |
+
tnr_scores = [float('nan')]*(len(columns) + len(columns2))
|
86 |
+
f1_scores = [float('nan')]*(len(columns) + len(columns2))
|
87 |
+
|
88 |
+
for c in columns:
|
89 |
+
mask = (df.algorithm == c)
|
90 |
+
sel_df = df[mask]
|
91 |
+
|
92 |
+
samples_tested.append(len(sel_df))
|
93 |
+
tpr_scores.append(round(recall_score(sel_df.gnd_truth.values, sel_df.pred.values, pos_label=1), 3))
|
94 |
+
|
95 |
+
|
96 |
+
for c in columns2:
|
97 |
+
mask = (df.label == c)
|
98 |
+
sel_df = df[mask]
|
99 |
+
|
100 |
+
samples_tested.append(len(sel_df))
|
101 |
+
tpr_scores.append(round(recall_score(sel_df.gnd_truth.values, sel_df.pred.values, pos_label=1), 3))
|
102 |
+
|
103 |
+
mask = (df.label != "real")
|
104 |
+
sel_df = df[mask]
|
105 |
+
|
106 |
+
tpr_scores.append(round(recall_score(sel_df.gnd_truth.values, sel_df.pred.values, pos_label=1), 3))
|
107 |
+
|
108 |
+
mask = (df.label == "real")
|
109 |
+
sel_df = df[mask]
|
110 |
+
|
111 |
+
tnr_scores.append(round(recall_score(sel_df.gnd_truth.values, sel_df.pred.values, pos_label=0), 3))
|
112 |
+
|
113 |
+
sel_df = df.copy()
|
114 |
+
samples_tested.append(len(sel_df))
|
115 |
+
f1_scores.append(round(f1_score(sel_df.gnd_truth.values, sel_df.pred.values, average="macro"), 3))
|
116 |
+
|
117 |
+
|
118 |
+
lb = pd.DataFrame({"Sample": columns + columns2 + ["overall (real + fake)"], "Num Samples": samples_tested,
|
119 |
+
"TPR": tpr_scores, "TNR": tnr_scores, "F1": f1_scores})
|
120 |
+
return lb
|
121 |
+
|
122 |
def build_leaderboard(results_path = 'results'):
|
123 |
full_df = get_merged_df(results_path)
|
124 |
full_df_mapped = map_df(full_df)
|
125 |
+
leaderboard = get_algorithm_scores_v2(full_df_mapped)
|
126 |
return leaderboard
|