unitorch.cli.models
Classification
Tip
core/postprocess/classification/binary_score in configuration file to use the postprocess function.
Return positive-class score for binary classification.
Source code in src/unitorch/cli/models/classification_utils.py
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79 | @register_process("core/postprocess/classification/binary_score")
def _binary_score(
self,
outputs: ClassificationOutputs,
):
"""Return positive-class score for binary classification."""
assert outputs.outputs.dim() == 2
results = outputs.to_pandas()
assert results.shape[0] == 0 or results.shape[0] == outputs.outputs.shape[0]
outputs = outputs.outputs.numpy()
if self.act_fn is not None:
outputs = self.act_fn(outputs)
if outputs.ndim == 2:
pscore = outputs[:, 1] if outputs.shape[-1] > 1 else outputs[:, 0]
results["pscore"] = pscore.tolist()
else:
results["pscore"] = outputs.tolist()
return WriterOutputs(results)
|
Tip
core/postprocess/classification/score in configuration file to use the postprocess function.
Return argmax class and max score for multi-class classification.
Source code in src/unitorch/cli/models/classification_utils.py
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103 | @register_process("core/postprocess/classification/score")
def _classifier_score(
self,
outputs: ClassificationOutputs,
):
"""Return argmax class and max score for multi-class classification."""
assert outputs.outputs.dim() == 2
results = outputs.to_pandas()
assert results.shape[0] == 0 or results.shape[0] == outputs.outputs.shape[0]
outputs = outputs.outputs.numpy()
if self.act_fn is not None:
outputs = self.act_fn(outputs)
results["pscore"] = outputs.max(-1)
results["pclass"] = outputs.argmax(-1)
if self.id2label is not None:
results["pclass"] = results["pclass"].map(self.id2label)
if self.return_scores:
results["scores"] = outputs.tolist()
return WriterOutputs(results)
|
Tip
core/postprocess/classification/embedding in configuration file to use the postprocess function.
Postprocess embedding outputs, writing each embedding field to results.
Source code in src/unitorch/cli/models/classification_utils.py
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 | @register_process("core/postprocess/classification/embedding")
def _embedding(
self,
outputs: EmbeddingOutputs,
):
"""Postprocess embedding outputs, writing each embedding field to results."""
results = outputs.to_pandas()
assert results.shape[0] == 0 or results.shape[0] == outputs.embedding.shape[0]
embedding = outputs.embedding.numpy()
if embedding.ndim > 2:
embedding = embedding.reshape(embedding.size(0), -1)
results["embedding"] = embedding.tolist()
embedding1 = outputs.embedding1.numpy()
if embedding1.size > 0:
if embedding1.ndim > 2:
embedding1 = embedding1.reshape(embedding1.size(0), -1)
results["embedding1"] = embedding1.tolist()
embedding2 = outputs.embedding2.numpy()
if embedding2.size > 0:
if embedding2.ndim > 2:
embedding2 = embedding2.reshape(embedding2.size(0), -1)
results["embedding2"] = embedding2.tolist()
embedding3 = outputs.embedding3.numpy()
if embedding3.size > 0:
if embedding3.ndim > 2:
embedding3 = embedding3.reshape(embedding3.size(0), -1)
results["embedding3"] = embedding3.tolist()
embedding4 = outputs.embedding4.numpy()
if embedding4.size > 0:
if embedding4.ndim > 2:
embedding4 = embedding4.reshape(embedding4.size(0), -1)
results["embedding4"] = embedding4.tolist()
return WriterOutputs(results)
|
Tip
core/postprocess/classification/embedding/string in configuration file to use the postprocess function.
Postprocess embedding outputs as space-joined string representations.
Source code in src/unitorch/cli/models/classification_utils.py
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 | @register_process("core/postprocess/classification/embedding/string")
def _embedding_string(
self,
outputs: EmbeddingOutputs,
):
"""Postprocess embedding outputs as space-joined string representations."""
results = outputs.to_pandas()
assert results.shape[0] == 0 or results.shape[0] == outputs.embedding.shape[0]
embedding = outputs.embedding.numpy()
if embedding.ndim > 2:
embedding = embedding.reshape(embedding.size(0), -1)
results["embedding"] = embedding.tolist()
results["embedding"] = results["embedding"].map(
lambda x: " ".join([str(i) for i in x])
)
embedding1 = outputs.embedding1.numpy()
if embedding1.size > 0:
if embedding1.ndim > 2:
embedding1 = embedding1.reshape(embedding1.size(0), -1)
results["embedding1"] = embedding1.tolist()
results["embedding1"] = results["embedding1"].map(
lambda x: " ".join([str(i) for i in x])
)
embedding2 = outputs.embedding2.numpy()
if embedding2.size > 0:
if embedding2.ndim > 2:
embedding2 = embedding2.reshape(embedding2.size(0), -1)
results["embedding2"] = embedding2.tolist()
results["embedding2"] = results["embedding2"].map(
lambda x: " ".join([str(i) for i in x])
)
embedding3 = outputs.embedding3.numpy()
if embedding3.size > 0:
if embedding3.ndim > 2:
embedding3 = embedding3.reshape(embedding3.size(0), -1)
results["embedding3"] = embedding3.tolist()
results["embedding3"] = results["embedding3"].map(
lambda x: " ".join([str(i) for i in x])
)
embedding4 = outputs.embedding4.numpy()
if embedding4.size > 0:
if embedding4.ndim > 2:
embedding4 = embedding4.reshape(embedding4.size(0), -1)
results["embedding4"] = embedding4.tolist()
results["embedding4"] = results["embedding4"].map(
lambda x: " ".join([str(i) for i in x])
)
return WriterOutputs(results)
|