-
Notifications
You must be signed in to change notification settings - Fork 252
/
test_models_temp.py
215 lines (191 loc) · 8.59 KB
/
test_models_temp.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
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
import pytest
# TODO: TEC-1819: Enable coexisting eager and graph tf tests:
# see https://stackoverflow.com/questions/50143896/both-eager-and-graph-execution-in-tensorflow-tests
pytestmark = pytest.mark.skip("Skipping to avoid failures in latent feature model tests caused by tf.eager."
"Will use Autograph feature in KnowEvolve code in the future to get rid of eager.")
import numpy as np
from ampligraph.evaluation.metrics import mr_score
import tensorflow as tf
X_bis = np.array([[2.0, 18.0, 0.0, 0.0],
[153.0, 5.0, 9.0, 0.0],
[103.0, 0.0, 82.0, 0.0], # seen in the train
[153.0, 5.0, 56.0, 0.0],
[5.0, 17.0, 60.0, 0.0],
[9.0, 6.0, 153.0, 24.0],
[153.0, 5.0, 3.0, 0.0],
[44.0, 10.0, 1.0, 0.0], # with 0 seen
[3.0, 6.0, 153.0, 0.0],
[242.0, 0.0, 8.0, 24.0],
[7.0, 1.0, 170.0, 24.0],
[56.0, 6.0, 153.0, 0.0],
[211.0, 6.0, 40.0, 24.0],
[60.0, 17.0, 5.0, 0.0],
[170.0, 1.0, 7.0, 0.0], # seen
[55.0, 118.0, 43.0, 0.0],
[36.0, 6.0, 153.0, 0.0],
[40.0, 5.0, 211.0, 24.0],
[257.0, 0.0, 66.0, 0.0],
[9.0, 8.0, 26.0, 0.0] # also seen
])
X_mask = np.array([-1, -1, -1, -1, 1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1])
def test_load_ICEWS_r():
datas=load_ICEWS_reduced()
assert datas['train'][1,2]==54
def test_KnowEvolve_running():
from ampligraph.temporal import KnowEvolve
X = np.array([
[0, 4, 7, 5.257244305824649],
[4, 0, 1, 5.424818694879746],
[9, 0, 6, 5.479774175414377],
[8, 0, 6, 5.528975403788779],
[6, 1, 9, 5.731247947359487],
[3, 3, 7, 5.741413035549908],
[6, 4, 0, 5.749306503402599],
[6, 2, 1, 5.907582200331214],
[8, 3, 2, 6.2633789148495005],
[7, 1, 9, 6.3488699640368225],
[7, 1, 5, 6.3869577883721975],
[5, 2, 2, 6.468355387856009],
[5, 4, 7, 7.074951676048747],
[4, 3, 3, 7.114139126265265], ])
KnE = KnowEvolve(max_iter=2, batch_size=3, lr=0.1, nb_nodes=10, nb_rel=5, dim_embed=6, dim_rel=3)
KnE.fit(X)
assert KnE.is_fitted
def test_load_and_fit_ICEWS_reduced():
from ampligraph.temporal import KnowEvolve
datas = load_ICEWS_reduced()
X=datas['train']
assert not np.any(np.isnan(X))
KnE = KnowEvolve(max_iter=1, batch_size=20, lr=0.02, nb_nodes=500, nb_rel=260, dim_embed=10, dim_rel=5)
KnE.fit(X[:1000])
assert KnE.is_fitted
def test_predict():
from ampligraph.temporal import KnowEvolve
datas = load_ICEWS_reduced()
X = datas['train']
KnE = KnowEvolve(max_iter=1, batch_size=20, lr=0.02, nb_nodes=500, nb_rel=260, dim_embed=10, dim_rel=5)
KnE.fit(X[:500])
scores = KnE.predict(X_bis, X_mask, update_embeddings=False)
assert (tf.reduce_all(scores>0))
scores = KnE.predict(X_bis, X_mask, update_embeddings=True)
assert (tf.reduce_all(scores > 0))
def test_predict_time():
from ampligraph.temporal import KnowEvolve
datas = load_ICEWS_reduced()
X = datas['train']
KnE = KnowEvolve(max_iter=1, batch_size=20, lr=0.02, nb_nodes=500, nb_rel=260, dim_embed=10, dim_rel=5)
KnE.fit(X[:500])
expctd_time = KnE.predict_time(X_bis)
assert (tf.reduce_all(expctd_time > 0))
def test_feed_running():
from ampligraph.temporal import KnowEvolve
datas = load_ICEWS_reduced()
X = datas['train']
KnE = KnowEvolve(max_iter=1, batch_size=20, lr=0.02, nb_nodes=500, nb_rel=260, dim_embed=10, dim_rel=5)
KnE.fit(X[:500])
KnE.feed(X_bis)
def test_save_and_restore():
from ampligraph.temporal import KnowEvolve
path = '/media/sf_Shared_with_VM/savingTest/'
name = 'model1'
datas = load_ICEWS_reduced()
X = datas['train']
KnE = KnowEvolve(max_iter=1, batch_size=20, lr=0.02, nb_nodes=500, nb_rel=260, dim_embed=10, dim_rel=5)
KnE.fit(X[:100])
KnE.save(path, name)
# need to try with differents parameters, like batch_size=25
kne_bis = KnowEvolve(max_iter=2, batch_size=20, lr=0.02, nb_nodes=500, nb_rel=260, dim_embed=10, dim_rel=5)
kne_bis.restore(path, name)
kne_bis_variables = [kne_bis.model.rEmb, kne_bis.model.RREmb, kne_bis.model.V, kne_bis.model.Whh, kne_bis.model.Wh,
kne_bis.model.Wts, kne_bis.model.Wto]
KnE_variables = [KnE.model.rEmb, KnE.model.RREmb, KnE.model.V, KnE.model.Whh, KnE.model.Wh, KnE.model.Wts,
KnE.model.Wto]
for t1, t2 in list(zip(kne_bis_variables, KnE_variables)):
print('testing')
assert (tf.squeeze(tf.reduce_all(tf.equal(t1, t2))))
assert (tf.reduce_all(tf.equal(kne_bis.predict(X[100:120], X_mask), KnE.predict(X[100:120], X_mask))))
def test_later_training():
from ampligraph.temporal import KnowEvolve
#work without the tensorflow seed, means the optimizer is well stored
path = '/media/sf_Shared_with_VM/savingTest/'
name = 'model1'
datas = load_ICEWS_reduced()
X = datas['train']
KnE = KnowEvolve(max_iter=1, batch_size=20, lr=0.02, nb_nodes=500, nb_rel=260, dim_embed=10, dim_rel=5)
KnE.fit(X[:100])
KnE.save(path, name)
kne_bis = KnowEvolve(max_iter=2, batch_size=20, lr=0.02, nb_nodes=500, nb_rel=260, dim_embed=10, dim_rel=5)
kne_bis.restore(path, name)
#second training, indepedent
KnE.fit(X[:100])
kne_bis.fit(X[:100])
#test
kne_bis_variables = [kne_bis.model.rEmb, kne_bis.model.RREmb, kne_bis.model.V, kne_bis.model.Whh, kne_bis.model.Wh,
kne_bis.model.Wts, kne_bis.model.Wto]
KnE_variables = [KnE.model.rEmb, KnE.model.RREmb, KnE.model.V, KnE.model.Whh, KnE.model.Wh, KnE.model.Wts,
KnE.model.Wto]
for t1, t2 in list(zip(kne_bis_variables, KnE_variables)):
print('testing')
assert (tf.squeeze(tf.reduce_all(tf.equal(t1, t2))))
assert (tf.reduce_all(tf.equal(kne_bis.predict(X[100:120], X_mask), KnE.predict(X[100:120], X_mask))))
def test_preprocess_predict():
from ampligraph.temporal import KnowEvolve
X_bis = np.array([[2.0, 18.0, 0.0, 0.0], #-1
[153.0, 5.0, 9.0, 0.0], #-1
[103.0, 0.0, 82.0, 0.0], #-1 # seen in the train
[153.0, 5.0, 56.0, 0.0], #-1
[5.0, 17.0, 60.0, 0.0], #1
[9.0, 6.0, 153.0, 24.0], #-1
[153.0, 5.0, 3.0, 0.0], #-1
[44.0, 10.0, 1.0, 0.0], #-1 # with 0 seen
[3.0, 6.0, 153.0, 0.0], #1
[242.0, 0.0, 8.0, 24.0], #1
[7.0, 1.0, 170.0, 24.0], #-1
[56.0, 6.0, 153.0, 0.0], #1
[211.0, 6.0, 40.0, 24.0], #-1
[60.0, 17.0, 5.0, 0.0], #-1
[170.0, 1.0, 7.0, 0.0], #-1 # seen
[55.0, 118.0, 43.0, 0.0], #-1
[36.0, 6.0, 153.0, 0.0], #-1
[40.0, 5.0, 211.0, 24.0], #-1
[257.0, 0.0, 66.0, 0.0], #-1
[9.0, 8.0, 26.0, 0.0] #-1 # also seen
])
X_mask = np.array([-1, -1, -1, -1, 1, -1, -1, -1, 1, 1, -1, 1, -1, -1, -1, -1, -1, -1, -1, -1])
KnE = KnowEvolve(max_iter=1, batch_size=20, lr=0.02, nb_nodes=500, nb_rel=260, dim_embed=10, dim_rel=5)
batches, cutting_in_batches, X_idx=KnE.preprocess_predict(X_bis, X_mask)
assert(len(batches)==2)
assert(len(batches[0])==15)
assert(len(batches[1])==5)
assert(cutting_in_batches[0]==12)
assert(cutting_in_batches[1] == 4)
assert(X_idx[19]==9)
def test_previous_table():
from ampligraph.temporal import KnowEvolve
X = np.array([
[0, 4, 7, 5.],
[4, 0, 1, 5.],
[9, 0, 6, 5.],
[8, 0, 13, 5.2],
[6, 1, 9, 5.7],
[3, 3, 7, 5.8],
[6, 4, 0, 5.9],
[6, 2, 1, 6.0],
[11, 3, 2, 6.26],
[7, 1, 9, 6.29],
[7, 1, 5, 6.38],
[5, 2, 2, 6.4],
[5, 4, 7, 7.0],
[4, 3, 3, 7.0], ])
KnE = KnowEvolve(max_iter=2, batch_size=3, lr=0.1, nb_nodes=20, nb_rel=5, dim_embed=6, dim_rel=3)
KnE.fit(X)
tp=KnE.tp
rp=KnE.rp
assert(tp[11,:].tolist()==[-24.0, 6.26])
assert(tp[2,:].tolist()==[6.26, 6.4])
assert(tp[15, :].tolist()==[-24.0, -24.0])
assert(rp[5]==4.0)
assert(rp[15]==-1.0)
def test_mar_ranks():
ranks=[8,9,10,3,1,6]
assert(mr_score(ranks) == 6 + 1 / 6)