Martijn van Beers
commited on
Commit
·
4b1b415
1
Parent(s):
9e7d7f8
Try gradio explanations
Browse files
app.py
CHANGED
@@ -156,7 +156,7 @@ def show_explanation(model, input_ids, attention_mask, index=None, start_layer=0
|
|
156 |
output, expl = generate_relevance(
|
157 |
model, input_ids, attention_mask, index=index, start_layer=start_layer
|
158 |
)
|
159 |
-
print(output.shape, expl.shape)
|
160 |
# normalize scores
|
161 |
scaler = PyTMinMaxScalerVectorized()
|
162 |
|
@@ -177,23 +177,34 @@ def show_explanation(model, input_ids, attention_mask, index=None, start_layer=0
|
|
177 |
tokens = tokenizer.convert_ids_to_tokens(input_ids[record].flatten())[
|
178 |
1 : 0 - ((attention_mask[record] == 0).sum().item() + 1)
|
179 |
]
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
classification,
|
186 |
-
classification,
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
195 |
|
196 |
-
def
|
197 |
text_batch = [input_text]
|
198 |
encoding = tokenizer(text_batch, return_tensors="pt")
|
199 |
input_ids = encoding["input_ids"].to(device)
|
@@ -202,14 +213,12 @@ def run(input_text):
|
|
202 |
# true class is positive - 1
|
203 |
true_class = 1
|
204 |
|
205 |
-
|
206 |
-
return html
|
207 |
-
|
208 |
|
209 |
-
|
210 |
-
fn=
|
211 |
inputs="text",
|
212 |
-
outputs="
|
213 |
title="RoBERTa Explanability",
|
214 |
description="Quick demo of a version of [Hila Chefer's](https://github.com/hila-chefer) [Transformer-Explanability](https://github.com/hila-chefer/Transformer-Explainability/) but without the layerwise relevance propagation (as in [Transformer-MM_explainability](https://github.com/hila-chefer/Transformer-MM-Explainability/)) for a RoBERTa model.",
|
215 |
examples=[
|
@@ -220,5 +229,6 @@ iface = gradio.Interface(
|
|
220 |
"I really didn't like this movie. Some of the actors were good, but overall the movie was boring"
|
221 |
],
|
222 |
],
|
|
|
223 |
)
|
224 |
-
|
|
|
156 |
output, expl = generate_relevance(
|
157 |
model, input_ids, attention_mask, index=index, start_layer=start_layer
|
158 |
)
|
159 |
+
#print(output.shape, expl.shape)
|
160 |
# normalize scores
|
161 |
scaler = PyTMinMaxScalerVectorized()
|
162 |
|
|
|
177 |
tokens = tokenizer.convert_ids_to_tokens(input_ids[record].flatten())[
|
178 |
1 : 0 - ((attention_mask[record] == 0).sum().item() + 1)
|
179 |
]
|
180 |
+
vis_data_records.append(list(zip(tokens, nrm.tolist())))
|
181 |
+
#print([(tokens[i], nrm[i].item()) for i in range(len(tokens))])
|
182 |
+
# vis_data_records.append(
|
183 |
+
# visualization.VisualizationDataRecord(
|
184 |
+
# nrm,
|
185 |
+
# output[record][classification],
|
186 |
+
# classification,
|
187 |
+
# classification,
|
188 |
+
# index,
|
189 |
+
# 1,
|
190 |
+
# tokens,
|
191 |
+
# 1,
|
192 |
+
# )
|
193 |
+
# )
|
194 |
+
# return visualize_text(vis_data_records)
|
195 |
+
return vis_data_records
|
196 |
+
|
197 |
+
|
198 |
+
def sentence_sentiment(input_text):
|
199 |
+
text_batch = [input_text]
|
200 |
+
encoding = tokenizer(text_batch, return_tensors="pt")
|
201 |
+
input_ids = encoding["input_ids"].to(device)
|
202 |
+
attention_mask = encoding["attention_mask"].to(device)
|
203 |
+
output = model(input_ids=input_ids, attention_mask=attention_mask)[0]
|
204 |
+
index = output.argmax(axis=-1).item()
|
205 |
+
return classification[index]
|
206 |
|
207 |
+
def sentiment_explanation_hila(input_text):
|
208 |
text_batch = [input_text]
|
209 |
encoding = tokenizer(text_batch, return_tensors="pt")
|
210 |
input_ids = encoding["input_ids"].to(device)
|
|
|
213 |
# true class is positive - 1
|
214 |
true_class = 1
|
215 |
|
216 |
+
return show_explanation(model, input_ids, attention_mask)
|
|
|
|
|
217 |
|
218 |
+
hila = gradio.Interface(
|
219 |
+
fn=sentence_sentiment,
|
220 |
inputs="text",
|
221 |
+
outputs="label",
|
222 |
title="RoBERTa Explanability",
|
223 |
description="Quick demo of a version of [Hila Chefer's](https://github.com/hila-chefer) [Transformer-Explanability](https://github.com/hila-chefer/Transformer-Explainability/) but without the layerwise relevance propagation (as in [Transformer-MM_explainability](https://github.com/hila-chefer/Transformer-MM-Explainability/)) for a RoBERTa model.",
|
224 |
examples=[
|
|
|
229 |
"I really didn't like this movie. Some of the actors were good, but overall the movie was boring"
|
230 |
],
|
231 |
],
|
232 |
+
interpretation=sentiment_explanation_hila
|
233 |
)
|
234 |
+
hila.launch()
|