Update app.py
Browse files
app.py
CHANGED
@@ -15,6 +15,7 @@
|
|
15 |
__author__ = 'Dmitry Ustalov'
|
16 |
__license__ = 'Apache 2.0'
|
17 |
|
|
|
18 |
from typing import IO, Tuple, List, cast, Dict, Set, Callable
|
19 |
|
20 |
import gradio as gr
|
@@ -68,28 +69,38 @@ def bradley_terry(wins: npt.NDArray[np.int64], ties: npt.NDArray[np.int64],
|
|
68 |
return p
|
69 |
|
70 |
|
71 |
-
def centrality(algorithm: Callable[
|
72 |
-
wins: npt.NDArray[np.int64], ties: npt.NDArray[np.int64]
|
73 |
-
tolerance: float = 10e-6, limit: int = 100) -> npt.NDArray[np.float64]:
|
74 |
A = wins + .5 * ties
|
75 |
|
76 |
G = nx.from_numpy_array(A, create_using=nx.DiGraph)
|
77 |
|
78 |
-
scores: Dict[int, float] = algorithm(G
|
79 |
|
80 |
p = np.array([scores[i] for i in range(len(G))])
|
81 |
|
82 |
return p
|
83 |
|
84 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
def eigen(wins: npt.NDArray[np.int64], ties: npt.NDArray[np.int64],
|
86 |
seed: int = 0, tolerance: float = 10e-6, limit: int = 100) -> npt.NDArray[np.float64]:
|
87 |
-
|
|
|
|
|
88 |
|
89 |
|
90 |
def pagerank(wins: npt.NDArray[np.int64], ties: npt.NDArray[np.int64],
|
91 |
seed: int = 0, tolerance: float = 10e-6, limit: int = 100) -> npt.NDArray[np.float64]:
|
92 |
-
|
|
|
|
|
93 |
|
94 |
|
95 |
# https://gist.github.com/dustalov/41678b70c40ba5a55430fa5e77b121d9#file-newman-py
|
@@ -141,6 +152,7 @@ def newman(wins: npt.NDArray[np.int64], ties: npt.NDArray[np.int64],
|
|
141 |
|
142 |
|
143 |
ALGORITHMS = {
|
|
|
144 |
'Bradley-Terry (1952)': bradley_terry,
|
145 |
'Eigenvector (1986)': eigen,
|
146 |
'PageRank (1998)': pagerank,
|
@@ -156,7 +168,7 @@ def largest_strongly_connected_component(df: pd.DataFrame) -> Set[str]:
|
|
156 |
return cast(Set[str], largest)
|
157 |
|
158 |
|
159 |
-
def handler(file: IO[bytes], algorithm: str, filtered: bool, seed: int) -> Tuple[pd.DataFrame, Figure]:
|
160 |
if file is None:
|
161 |
raise gr.Error('File must be uploaded')
|
162 |
|
@@ -219,6 +231,10 @@ def handler(file: IO[bytes], algorithm: str, filtered: bool, seed: int) -> Tuple
|
|
219 |
df_result.sort_values(by=['rank', 'score'], ascending=[True, False], inplace=True)
|
220 |
df_result.reset_index(inplace=True)
|
221 |
|
|
|
|
|
|
|
|
|
222 |
df_pairwise = pd.DataFrame(data=scores[:, np.newaxis] / (scores + scores[:, np.newaxis]),
|
223 |
index=index, columns=index)
|
224 |
df_pairwise = df_pairwise.reindex(labels=df_result['item'], columns=df_result['item'], copy=False)
|
@@ -249,6 +265,12 @@ def main() -> None:
|
|
249 |
'This option keeps only the largest strongly-connected component (SCC) of the input graph. '
|
250 |
'Some items might be missing as a result of this filtering.'
|
251 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
252 |
gr.Number(
|
253 |
label='Seed',
|
254 |
precision=0
|
@@ -264,6 +286,7 @@ def main() -> None:
|
|
264 |
)
|
265 |
],
|
266 |
examples=[
|
|
|
267 |
['food.csv', 'Bradley-Terry (1952)', False],
|
268 |
['food.csv', 'Eigenvector (1986)', False],
|
269 |
['food.csv', 'PageRank (1998)', False],
|
|
|
15 |
__author__ = 'Dmitry Ustalov'
|
16 |
__license__ = 'Apache 2.0'
|
17 |
|
18 |
+
from functools import partial
|
19 |
from typing import IO, Tuple, List, cast, Dict, Set, Callable
|
20 |
|
21 |
import gradio as gr
|
|
|
69 |
return p
|
70 |
|
71 |
|
72 |
+
def centrality(algorithm: Callable[[nx.DiGraph], Dict[int, float]],
|
73 |
+
wins: npt.NDArray[np.int64], ties: npt.NDArray[np.int64]) -> npt.NDArray[np.float64]:
|
|
|
74 |
A = wins + .5 * ties
|
75 |
|
76 |
G = nx.from_numpy_array(A, create_using=nx.DiGraph)
|
77 |
|
78 |
+
scores: Dict[int, float] = algorithm(G)
|
79 |
|
80 |
p = np.array([scores[i] for i in range(len(G))])
|
81 |
|
82 |
return p
|
83 |
|
84 |
|
85 |
+
def counting(wins: npt.NDArray[np.int64], ties: npt.NDArray[np.int64],
|
86 |
+
seed: int = 0, tolerance: float = 10e-6, limit: int = 100) -> npt.NDArray[np.float64]:
|
87 |
+
M = wins + .5 * ties
|
88 |
+
|
89 |
+
return M.sum(axis=0)
|
90 |
+
|
91 |
+
|
92 |
def eigen(wins: npt.NDArray[np.int64], ties: npt.NDArray[np.int64],
|
93 |
seed: int = 0, tolerance: float = 10e-6, limit: int = 100) -> npt.NDArray[np.float64]:
|
94 |
+
algorithm = partial(nx.algorithms.eigenvector_centrality_numpy, max_iter=limit, tol=tolerance)
|
95 |
+
|
96 |
+
return centrality(algorithm, wins, ties)
|
97 |
|
98 |
|
99 |
def pagerank(wins: npt.NDArray[np.int64], ties: npt.NDArray[np.int64],
|
100 |
seed: int = 0, tolerance: float = 10e-6, limit: int = 100) -> npt.NDArray[np.float64]:
|
101 |
+
algorithm = partial(nx.algorithms.pagerank, max_iter=limit, tol=tolerance)
|
102 |
+
|
103 |
+
return centrality(algorithm, wins, ties)
|
104 |
|
105 |
|
106 |
# https://gist.github.com/dustalov/41678b70c40ba5a55430fa5e77b121d9#file-newman-py
|
|
|
152 |
|
153 |
|
154 |
ALGORITHMS = {
|
155 |
+
'Counting': counting,
|
156 |
'Bradley-Terry (1952)': bradley_terry,
|
157 |
'Eigenvector (1986)': eigen,
|
158 |
'PageRank (1998)': pagerank,
|
|
|
168 |
return cast(Set[str], largest)
|
169 |
|
170 |
|
171 |
+
def handler(file: IO[bytes], algorithm: str, filtered: bool, truncated: bool, seed: int) -> Tuple[pd.DataFrame, Figure]:
|
172 |
if file is None:
|
173 |
raise gr.Error('File must be uploaded')
|
174 |
|
|
|
231 |
df_result.sort_values(by=['rank', 'score'], ascending=[True, False], inplace=True)
|
232 |
df_result.reset_index(inplace=True)
|
233 |
|
234 |
+
if truncated:
|
235 |
+
df_result = pd.concat((df_result.head(5), df_result.tail(5)), copy=False)
|
236 |
+
df_result = df_result[~df_result.index.duplicated(keep='last')]
|
237 |
+
|
238 |
df_pairwise = pd.DataFrame(data=scores[:, np.newaxis] / (scores + scores[:, np.newaxis]),
|
239 |
index=index, columns=index)
|
240 |
df_pairwise = df_pairwise.reindex(labels=df_result['item'], columns=df_result['item'], copy=False)
|
|
|
265 |
'This option keeps only the largest strongly-connected component (SCC) of the input graph. '
|
266 |
'Some items might be missing as a result of this filtering.'
|
267 |
),
|
268 |
+
gr.Checkbox(
|
269 |
+
value=False,
|
270 |
+
label='Truncate Output',
|
271 |
+
info='Perform the entire computation but output only five head and five tail items, '
|
272 |
+
'avoiding overlap.'
|
273 |
+
),
|
274 |
gr.Number(
|
275 |
label='Seed',
|
276 |
precision=0
|
|
|
286 |
)
|
287 |
],
|
288 |
examples=[
|
289 |
+
['food.csv', 'Counting', False],
|
290 |
['food.csv', 'Bradley-Terry (1952)', False],
|
291 |
['food.csv', 'Eigenvector (1986)', False],
|
292 |
['food.csv', 'PageRank (1998)', False],
|