-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluate_odp.py
182 lines (147 loc) · 5.5 KB
/
evaluate_odp.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
import json
from datetime import datetime
from typing import Dict, List
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from sklearn.linear_model import Lasso
from sklearn.metrics import adjusted_rand_score
from tqdm import tqdm
from search_clustering.pipeline import KNNPipeline
from search_clustering.presets import get_odp_params, make_pipelines
from search_clustering.utils.odp_239 import (
DEFAULT_EMBEDDINGS,
align_clusters_by_label,
create_odp239_splits,
embed_odp239_labels_in_splits,
embed_target_name,
read_odp239_to_df,
subtopic_recall,
)
def evaluate(data: dict, params: dict):
pipelines = make_pipelines(params)
results: Dict[str, List[str]] = {
"id": [],
"silhouette": [],
"ari": [],
"recall": [],
"outliers": [],
"time": [],
}
for identifier, pipeline in tqdm(pipelines.items(), desc="Pipelines"):
silhouette_pipe = []
ari_pipe = []
recall_pipe = []
outliers_pipe = []
time_pipe = []
for cat in data.keys():
start_time = datetime.now()
_, clusters, labels, silhouette = pipeline.fit_transform(
data[cat]["data"], verbose=False, visualize=False
)
delta = datetime.now() - start_time
ari = adjusted_rand_score(data[cat]["target"], clusters)
label_embeddings = np.vstack(
[embed_target_name(label, DEFAULT_EMBEDDINGS).cpu() for label in labels]
)
aligned_clusters = align_clusters_by_label(
data[cat]["target_embeddings"],
label_embeddings,
clusters,
n_neighbors=1,
)
recall = subtopic_recall(data[cat]["target"], aligned_clusters)
outliers = len(clusters[clusters == -1]) / len(clusters)
silhouette_pipe.append(silhouette)
ari_pipe.append(ari)
recall_pipe.append(recall)
outliers_pipe.append(outliers)
time_pipe.append(delta.total_seconds())
def make_str(values: List[float]) -> str:
mean = round(float(np.mean(values)), 2)
std = round(float(np.std(values)), 2)
return f"{mean} ± {std}"
results["id"].append(identifier)
results["silhouette"].append(make_str(silhouette_pipe))
results["ari"].append(make_str(ari_pipe))
results["recall"].append(make_str(recall_pipe))
results["outliers"].append(make_str(outliers_pipe))
results["time"].append(make_str(time_pipe))
return results
def evaluate_detailed(data: dict, pipeline: KNNPipeline) -> pd.DataFrame:
category = []
support = []
n_clusters = []
ari = []
recall = []
for cat in data.keys():
_, clusters, labels, _ = pipeline.fit_transform(
data[cat]["data"], verbose=False, visualize=False
)
label_embeddings = np.vstack(
[embed_target_name(label, DEFAULT_EMBEDDINGS).cpu() for label in labels]
)
aligned_clusters = align_clusters_by_label(
data[cat]["target_embeddings"],
label_embeddings,
clusters,
n_neighbors=1,
)
category.append(cat)
support.append(len(data[cat]["data"]))
n_clusters.append(max(clusters))
ari.append(adjusted_rand_score(data[cat]["target"], clusters))
recall.append(subtopic_recall(data[cat]["target"], aligned_clusters))
return pd.DataFrame(
{
"category": category,
"support": support,
"n_clusters": n_clusters,
"ari": ari,
"recall": recall,
}
)
def plot_correlations(df: pd.DataFrame, title: str = ""):
df["n_clusters"] /= df["n_clusters"].max()
corr = df.corr()["support"]
x = np.linspace(df["support"].min(), df["support"].max(), num=2).reshape(-1, 1)
lr = Lasso()
fig, ax = plt.subplots(figsize=(5, 5))
labels = {"n_clusters": "#Clusters", "ari": "ARI", "recall": "Recall"}
for col in corr.index[1:]:
lr.fit(df["support"].to_numpy().reshape(-1, 1), df[col])
ax.scatter(df["support"], df[col], alpha=0.75)
ax.plot(
x,
lr.predict(x),
linestyle="dashed",
label=rf"{labels[col]}, $\rho = {corr[col]:.4f}$",
)
ax.set_xlabel("Number of Documents")
ax.set_ylim(0.0, 1.0)
ax.legend()
ax.set_title(title)
fig.show()
def read_results(filename: str = f"evaluation_odp.json") -> pd.DataFrame:
with open(f"results/{filename}", "r") as f:
results = json.loads(f.read())
# preprocessing and labeling equal for all pipelines, remove
for i in range(len(results["id"])):
id_i = results["id"][i]
results["id"][i] = id_i[id_i.find("_") + 1 : id_i.rfind("_")]
df = pd.DataFrame(results)
# use pipeline parameters as hierarchical multindex
split = df["id"].str.split("_")
split = pd.DataFrame(
split.to_list(), columns=["embedding", "reduction", "clustering"]
)
df.index = pd.MultiIndex.from_frame(split[["embedding", "clustering", "reduction"]])
df = df.sort_index(level=["embedding"])
return df.drop(columns=["id"])
if __name__ == "__main__":
df = read_odp239_to_df()
data = create_odp239_splits(df)
data = embed_odp239_labels_in_splits(data)
results = evaluate(data, get_odp_params("all"))
with open("results/evaluation_odp_gpu.json", "w") as f:
f.write(json.dumps(results, indent=2))