taka-yamakoshi
commited on
Commit
•
50ce4f4
1
Parent(s):
8f32fbf
first pass
Browse files
app.py
CHANGED
@@ -58,6 +58,41 @@ def clear_data():
|
|
58 |
for key in st.session_state:
|
59 |
del st.session_state[key]
|
60 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
def show_annotated_sentence(sent,option_locs=[],mask_locs=[]):
|
62 |
disp_style = '"font-family:san serif; color:Black; font-size: 20px"'
|
63 |
prefix = f'<p style={disp_style}><span style="font-weight:bold">'
|
@@ -90,43 +125,45 @@ if __name__=='__main__':
|
|
90 |
sent_1 = st.text_input('Sentence 1',value='It is better to play a prank on Samuel than Craig because he gets angry less often.')
|
91 |
sent_2 = st.text_input('Sentence 2',value='It is better to play a prank on Samuel than Craig because he gets angry more often.')
|
92 |
if st.button('Tokenize'):
|
93 |
-
st.session_state['page_status'] = '
|
94 |
st.session_state['sent_1'] = sent_1
|
95 |
st.session_state['sent_2'] = sent_2
|
96 |
main_area.empty()
|
97 |
|
98 |
-
if st.session_state['page_status']=='
|
99 |
with main_area.container():
|
100 |
sent_1 = st.session_state['sent_1']
|
101 |
sent_2 = st.session_state['sent_2']
|
102 |
-
if 'masked_pos_1' not in st.session_state:
|
103 |
-
st.session_state['masked_pos_1'] = []
|
104 |
-
if 'masked_pos_2' not in st.session_state:
|
105 |
-
st.session_state['masked_pos_2'] = []
|
106 |
|
107 |
st.write('2. Select sites to mask out and click "Confirm"')
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
st.session_state['masked_pos_1'].remove(word_id)
|
119 |
-
st.markdown(show_annotated_sentence(decoded_sent,mask_locs=st.session_state['masked_pos_1']), unsafe_allow_html = True)
|
120 |
|
|
|
|
|
|
|
|
|
|
|
|
|
121 |
|
122 |
if st.session_state['page_status']=='analysis':
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
|
|
|
|
|
58 |
for key in st.session_state:
|
59 |
del st.session_state[key]
|
60 |
|
61 |
+
def annotate_mask(sent_id,sent):
|
62 |
+
st.write(f'Sentence {sent_id}')
|
63 |
+
input_sent = tokenizer(sent).input_ids
|
64 |
+
decoded_sent = [tokenizer.decode([token]) for token in input_sent[1:-1]]
|
65 |
+
char_nums = [len(word)+2 for word in decoded_sent]
|
66 |
+
cols = st.columns(char_nums)
|
67 |
+
mask_locs = []
|
68 |
+
for word_id,(col,word) in enumerate(zip(cols,decoded_sent)):
|
69 |
+
with col:
|
70 |
+
if st.button(word,key=f'word_{word_id}'):
|
71 |
+
if word_id not in mask_locs:
|
72 |
+
mask_locs.append(word_id)
|
73 |
+
else:
|
74 |
+
mask_locs.remove(word_id)
|
75 |
+
st.markdown(show_annotated_sentence(decoded_sent,mask_locs=mask_locs), unsafe_allow_html = True)
|
76 |
+
return mask_locs
|
77 |
+
|
78 |
+
def annotate_options(sent_id,sent):
|
79 |
+
st.write(f'Sentence {sent_id}')
|
80 |
+
input_sent = tokenizer(sent).input_ids
|
81 |
+
decoded_sent = [tokenizer.decode([token]) for token in input_sent[1:-1]]
|
82 |
+
char_nums = [len(word)+2 for word in decoded_sent]
|
83 |
+
cols = st.columns(char_nums)
|
84 |
+
option_locs = []
|
85 |
+
for word_id,(col,word) in enumerate(zip(cols,decoded_sent)):
|
86 |
+
with col:
|
87 |
+
if st.button(word,key=f'word_{word_id}'):
|
88 |
+
if word_id not in option_locs:
|
89 |
+
option_locs.append(word_id)
|
90 |
+
else:
|
91 |
+
option_locs.remove(word_id)
|
92 |
+
st.markdown(show_annotated_sentence(decoded_sent,option_locs=option_locs,
|
93 |
+
mask_locs=st.session_state[f'mask_locs_{sent_id}']), unsafe_allow_html = True)
|
94 |
+
return option_locs
|
95 |
+
|
96 |
def show_annotated_sentence(sent,option_locs=[],mask_locs=[]):
|
97 |
disp_style = '"font-family:san serif; color:Black; font-size: 20px"'
|
98 |
prefix = f'<p style={disp_style}><span style="font-weight:bold">'
|
|
|
125 |
sent_1 = st.text_input('Sentence 1',value='It is better to play a prank on Samuel than Craig because he gets angry less often.')
|
126 |
sent_2 = st.text_input('Sentence 2',value='It is better to play a prank on Samuel than Craig because he gets angry more often.')
|
127 |
if st.button('Tokenize'):
|
128 |
+
st.session_state['page_status'] = 'annotate_mask'
|
129 |
st.session_state['sent_1'] = sent_1
|
130 |
st.session_state['sent_2'] = sent_2
|
131 |
main_area.empty()
|
132 |
|
133 |
+
if st.session_state['page_status']=='annotate_mask':
|
134 |
with main_area.container():
|
135 |
sent_1 = st.session_state['sent_1']
|
136 |
sent_2 = st.session_state['sent_2']
|
|
|
|
|
|
|
|
|
137 |
|
138 |
st.write('2. Select sites to mask out and click "Confirm"')
|
139 |
+
st.session_state[f'mask_locs_1'] = annotate_mask(1,sent_1)
|
140 |
+
st.session_state[f'mask_locs_2'] = annotate_mask(2,sent_2)
|
141 |
+
if st.button('Confirm'):
|
142 |
+
st.session_state['page_status'] = 'annotate_options'
|
143 |
+
main_area.empty()
|
144 |
+
|
145 |
+
if st.session_state['page_status'] == 'annotate_options':
|
146 |
+
with main_area.container():
|
147 |
+
sent_1 = st.session_state['sent_1']
|
148 |
+
sent_2 = st.session_state['sent_2']
|
|
|
|
|
149 |
|
150 |
+
st.write('2. Select options click "Confirm"')
|
151 |
+
st.session_state[f'option_locs_1'] = annotate_options(1,sent_1)
|
152 |
+
st.session_state[f'option_locs_2'] = annotate_options(2,sent_2)
|
153 |
+
if st.button('Confirm'):
|
154 |
+
st.session_state['page_status'] = 'analysis'
|
155 |
+
main_area.empty()
|
156 |
|
157 |
if st.session_state['page_status']=='analysis':
|
158 |
+
with main_area.container():
|
159 |
+
sent_1 = st.session_state['sent_1']
|
160 |
+
sent_2 = st.session_state['sent_2']
|
161 |
+
|
162 |
+
input_ids_1 = tokenizer(sent_1).input_ids
|
163 |
+
input_ids_2 = tokenizer(sent_2).input_ids
|
164 |
+
input_ids = torch.tensor([input_ids_1,input_ids_2])
|
165 |
+
|
166 |
+
outputs = SkeletonAlbertForMaskedLM(model,input_ids,interventions = {0:{'lay':[(8,1,[0,1])]}})
|
167 |
+
logprobs = F.log_softmax(outputs['logits'], dim = -1)
|
168 |
+
preds = [torch.multinomial(torch.exp(probs), num_samples=1).squeeze(dim=-1) for probs in logprobs[0]]
|
169 |
+
st.write([tokenizer.decode([token]) for token in preds])
|