Spaces:
Runtime error
Runtime error
adds probability_emote page
Browse files- app.py +36 -28
- src/__init__.py +0 -0
- src/probability_emote.py +184 -0
- src/story_gen.py +221 -0
- src/story_gen_test.py +34 -0
app.py
CHANGED
@@ -1,5 +1,6 @@
|
|
1 |
import streamlit as st
|
2 |
-
from
|
|
|
3 |
import plotly.express as px
|
4 |
import random
|
5 |
import numpy as np
|
@@ -11,36 +12,41 @@ container_mode = st.sidebar.container()
|
|
11 |
container_guide = st.container()
|
12 |
container_param = st.sidebar.container()
|
13 |
container_button = st.sidebar.container()
|
14 |
-
|
15 |
mode = container_mode.radio(
|
16 |
"Select a mode",
|
17 |
('Probability Emote', 'Create Statistics', 'Play Storytelling'), index=0)
|
18 |
|
19 |
-
choices_first_sentence = [
|
20 |
-
'Custom',
|
21 |
-
'Hello, I\'m a language model,',
|
22 |
-
'So I suppose you want to ask me how I did it.',
|
23 |
-
'I always wanted to be a giraffe - until that night.',
|
24 |
-
'My first tutor was a dragon with a terrible sense of humor.',
|
25 |
-
'Doctors told her she could never diet again.',
|
26 |
-
'Memory is all around us, as well as within.',
|
27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
|
29 |
-
]
|
30 |
-
cfs = st.selectbox('Choose First Sentence', choices_first_sentence)
|
31 |
-
if cfs == 'Custom':
|
32 |
-
story_till_now = st.text_input(
|
33 |
-
label='First Sentence', key='first_sentence')
|
34 |
-
else:
|
35 |
-
st.session_state.first_sentence = cfs
|
36 |
-
story_till_now = cfs
|
37 |
-
first_sentence = story_till_now
|
38 |
-
first_emotion = gen.get_emotion(first_sentence)
|
39 |
|
40 |
-
|
41 |
-
|
42 |
-
if
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
|
|
|
|
|
|
|
|
|
44 |
num_generation = container_param.slider(
|
45 |
label='Number of generation', min_value=1, max_value=100, value=5, step=1)
|
46 |
num_tests = container_param.slider(
|
@@ -53,7 +59,7 @@ if mode == 'Create Statistics':
|
|
53 |
elif reaction_weight_mode == "Random":
|
54 |
reaction_weight = -1
|
55 |
if container_button.button('Analyse'):
|
56 |
-
gen.get_stats(story_till_now=
|
57 |
num_generation=num_generation, length=length, reaction_weight=reaction_weight, num_tests=num_tests)
|
58 |
# if len(gen.stories) > 0:
|
59 |
# for si, story in enumerate(gen.stories):
|
@@ -87,13 +93,14 @@ if mode == 'Create Statistics':
|
|
87 |
container_guide.markdown(
|
88 |
'### You selected statistics. Now set your parameters and click the `Analyse` button.')
|
89 |
elif mode == 'Play Storytelling':
|
90 |
-
|
|
|
91 |
if 'sentence_list' not in st.session_state:
|
92 |
st.session_state.sentence_list = [{'sentence': first_sentence,
|
93 |
'emotion': first_emotion['label'],
|
94 |
'score': first_emotion['score']}]
|
95 |
if 'full_story' not in st.session_state:
|
96 |
-
st.session_state.full_story =
|
97 |
container_button = container_button.columns([1, 1, 1])
|
98 |
heading_container = st.container()
|
99 |
col_turn, col_sentence, col_emo = st.columns([1, 8, 2])
|
@@ -129,5 +136,6 @@ elif mode == 'Play Storytelling':
|
|
129 |
st.session_state.sentence_list = [{'sentence': first_sentence,
|
130 |
'emotion': first_emotion['label'],
|
131 |
'score': first_emotion['score']}]
|
132 |
-
|
133 |
-
#
|
|
|
|
1 |
import streamlit as st
|
2 |
+
from src.story_gen import StoryGenerator
|
3 |
+
from src.probability_emote import run_pe
|
4 |
import plotly.express as px
|
5 |
import random
|
6 |
import numpy as np
|
|
|
12 |
container_guide = st.container()
|
13 |
container_param = st.sidebar.container()
|
14 |
container_button = st.sidebar.container()
|
|
|
15 |
mode = container_mode.radio(
|
16 |
"Select a mode",
|
17 |
('Probability Emote', 'Create Statistics', 'Play Storytelling'), index=0)
|
18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
+
def initialise_storytelling():
|
21 |
+
choices_first_sentence = [
|
22 |
+
'Custom',
|
23 |
+
'Hello, I\'m a language model,',
|
24 |
+
'So I suppose you want to ask me how I did it.',
|
25 |
+
'I always wanted to be a giraffe - until that night.',
|
26 |
+
'My first tutor was a dragon with a terrible sense of humor.',
|
27 |
+
'Doctors told her she could never diet again.',
|
28 |
+
'Memory is all around us, as well as within.',
|
29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
|
31 |
+
]
|
32 |
+
cfs = st.selectbox('Choose First Sentence', choices_first_sentence)
|
33 |
+
if cfs == 'Custom':
|
34 |
+
story_till_now = st.text_input(
|
35 |
+
label='First Sentence', key='first_sentence')
|
36 |
+
else:
|
37 |
+
st.session_state.first_sentence = cfs
|
38 |
+
story_till_now = cfs
|
39 |
+
first_sentence = story_till_now
|
40 |
+
first_emotion = gen.get_emotion(first_sentence)
|
41 |
+
|
42 |
+
length = container_param.slider(label='Length of the generated sentence',
|
43 |
+
min_value=1, max_value=100, value=10, step=1)
|
44 |
+
return first_sentence, first_emotion, length
|
45 |
|
46 |
+
|
47 |
+
if mode == 'Create Statistics':
|
48 |
+
first_sentence, first_emotion, length = initialise_storytelling()
|
49 |
+
# story_till_now = first_sentence
|
50 |
num_generation = container_param.slider(
|
51 |
label='Number of generation', min_value=1, max_value=100, value=5, step=1)
|
52 |
num_tests = container_param.slider(
|
|
|
59 |
elif reaction_weight_mode == "Random":
|
60 |
reaction_weight = -1
|
61 |
if container_button.button('Analyse'):
|
62 |
+
gen.get_stats(story_till_now=first_sentence,
|
63 |
num_generation=num_generation, length=length, reaction_weight=reaction_weight, num_tests=num_tests)
|
64 |
# if len(gen.stories) > 0:
|
65 |
# for si, story in enumerate(gen.stories):
|
|
|
93 |
container_guide.markdown(
|
94 |
'### You selected statistics. Now set your parameters and click the `Analyse` button.')
|
95 |
elif mode == 'Play Storytelling':
|
96 |
+
first_sentence, first_emotion, length = initialise_storytelling()
|
97 |
+
# story_till_now = first_sentence
|
98 |
if 'sentence_list' not in st.session_state:
|
99 |
st.session_state.sentence_list = [{'sentence': first_sentence,
|
100 |
'emotion': first_emotion['label'],
|
101 |
'score': first_emotion['score']}]
|
102 |
if 'full_story' not in st.session_state:
|
103 |
+
st.session_state.full_story = first_sentence
|
104 |
container_button = container_button.columns([1, 1, 1])
|
105 |
heading_container = st.container()
|
106 |
col_turn, col_sentence, col_emo = st.columns([1, 8, 2])
|
|
|
136 |
st.session_state.sentence_list = [{'sentence': first_sentence,
|
137 |
'emotion': first_emotion['label'],
|
138 |
'score': first_emotion['score']}]
|
139 |
+
elif mode == 'Probability Emote':
|
140 |
+
# container_mode.write('Let\'s play storytelling.')
|
141 |
+
run_pe(container_param)
|
src/__init__.py
ADDED
File without changes
|
src/probability_emote.py
ADDED
@@ -0,0 +1,184 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
|
3 |
+
import plotly.express as px
|
4 |
+
import plotly.graph_objects as go
|
5 |
+
|
6 |
+
import numpy as np
|
7 |
+
import numpy as np
|
8 |
+
|
9 |
+
@st.cache
|
10 |
+
def get_w(f, ec=0.86, rv=0.50):
|
11 |
+
result = (rv + f-1)/(ec + f-1)
|
12 |
+
result = np.clip(result, 0, 1)
|
13 |
+
print(f'w = {result}')
|
14 |
+
return result
|
15 |
+
|
16 |
+
@st.cache
|
17 |
+
def get_f(w, ec=0.86, rv=0.50):
|
18 |
+
result = 1+(ec*w-rv)/(1-w)
|
19 |
+
result = np.clip(result, 0, 1)
|
20 |
+
print(f'f = {result}')
|
21 |
+
return result
|
22 |
+
|
23 |
+
@st.cache
|
24 |
+
def get_pe(w, ec=0.86, f=0.50):
|
25 |
+
result = ec*w+(1-w)*(1-f)
|
26 |
+
result = np.clip(result, 0, 1)
|
27 |
+
print(f'f = {result}')
|
28 |
+
return result
|
29 |
+
|
30 |
+
xdata1 = np.arange(0, 1.1, step=0.01)
|
31 |
+
# rand = np.random.random_sample()
|
32 |
+
|
33 |
+
@st.cache
|
34 |
+
def proper_float(i):
|
35 |
+
return np.round(i, 2)
|
36 |
+
|
37 |
+
@st.cache
|
38 |
+
def get_text():
|
39 |
+
return '''
|
40 |
+
|
41 |
+
## Description
|
42 |
+
### Eric's proposal
|
43 |
+
> I would propose a scoring metric something like this:
|
44 |
+
> * `probability_emote = w * emotion_confidence + (1 - w) * frequency_penalty`
|
45 |
+
>
|
46 |
+
> Where:
|
47 |
+
> * emotion_confidence: a score from 0.0 to 1.0 representing the emotion model’s confidence in it’s classification results
|
48 |
+
> * frequency_penalty: a score from 0.0 to 1.0 where a high score penalizes frequent emotes
|
49 |
+
> * `frequency_penalty = 1 - emotion_frequency`
|
50 |
+
> * w: a weight from 0.0 to 1.0 that controls the balance between emotion_confidence and frequency_penalty
|
51 |
+
> * Then you generate a random number between 0.0 and 1.0 and emote if it is greater than probability_emote
|
52 |
+
> * You will have to set frequency_penalty and w through trial and error, but you can start with setting w=0.5 and giving the emotion classifier and frequency penalty equal weight.
|
53 |
+
> * Setting w=1.0 would disable the frequency penalty altogether
|
54 |
+
'''
|
55 |
+
|
56 |
+
@st.cache
|
57 |
+
def get_equation_text(w=0.5, ec=0.7, rand=None, emotion_frequency=None):
|
58 |
+
text = f'''
|
59 |
+
#### Equation
|
60 |
+
* frequency_penalty = 1 - emotion_frequency
|
61 |
+
* probability_emote = w * emotion_confidence + (1 - w) * frequency_penalty
|
62 |
+
* **probability_emote** = {proper_float(w)} * {proper_float(ec)} + {proper_float(1-w)} * frequency_penalty
|
63 |
+
|
64 |
+
'''
|
65 |
+
if rand is not None:
|
66 |
+
frequency_penalty=proper_float(1-emotion_frequency)
|
67 |
+
probability_emote=proper_float((w)*(ec)+(1-w)*frequency_penalty)
|
68 |
+
text = f'''
|
69 |
+
#### Equation
|
70 |
+
* frequency_penalty = 1 - emotion_frequency = 1 - {emotion_frequency} = {frequency_penalty}
|
71 |
+
* probability_emote = w * emotion_confidence + (1 - w) * frequency_penalty
|
72 |
+
* probability_emote = {proper_float(w)} * {proper_float(ec)} + {proper_float(1-w)} * {frequency_penalty}
|
73 |
+
* **probability_emote** = {probability_emote}
|
74 |
+
* Show_Emotion
|
75 |
+
= probability_emote > (Random value between 0 and 1)
|
76 |
+
* Random value = {rand}
|
77 |
+
* Show_Emotion = {probability_emote} > {rand}
|
78 |
+
* **Show_Emotion** = {probability_emote > rand}
|
79 |
+
'''
|
80 |
+
return text
|
81 |
+
|
82 |
+
def set_input(container_param,
|
83 |
+
label, key_slider, key_input,
|
84 |
+
min_value=0.,
|
85 |
+
max_value=1.,
|
86 |
+
value=.5,
|
87 |
+
step=.01,):
|
88 |
+
def slider2input():
|
89 |
+
st.session_state[key_input] = st.session_state[key_slider]
|
90 |
+
def input2slider():
|
91 |
+
st.session_state[key_slider] = st.session_state[key_input]
|
92 |
+
container_param=container_param.columns([1.1, 1])
|
93 |
+
slider_input = container_param[1].slider(
|
94 |
+
label=label,
|
95 |
+
min_value=min_value,
|
96 |
+
max_value=max_value,
|
97 |
+
value=value,
|
98 |
+
step=step,
|
99 |
+
key = key_slider,
|
100 |
+
on_change=slider2input)
|
101 |
+
number_input= container_param[0].number_input(
|
102 |
+
label='',
|
103 |
+
min_value=min_value,
|
104 |
+
max_value=max_value,
|
105 |
+
value=value,
|
106 |
+
step=step,
|
107 |
+
key = key_input,
|
108 |
+
on_change=input2slider)
|
109 |
+
return slider_input, number_input
|
110 |
+
|
111 |
+
def run_pe(container_param):
|
112 |
+
w_slider, w=set_input(container_param,
|
113 |
+
label='Weight w', key_slider='w_slider', key_input='w_input',
|
114 |
+
min_value=0.,
|
115 |
+
max_value=1.,
|
116 |
+
value=.5,
|
117 |
+
step=.01,)
|
118 |
+
score_slider, score=set_input(container_param,
|
119 |
+
label='Confidence Score', key_slider='score_slider', key_input='score_input',
|
120 |
+
min_value=0.,
|
121 |
+
max_value=1.,
|
122 |
+
value=.5,
|
123 |
+
step=.01,)
|
124 |
+
# score = container_param.slider(
|
125 |
+
# label='Confidence Score',
|
126 |
+
# min_value=0.,
|
127 |
+
# max_value=1.,
|
128 |
+
# value=.5,
|
129 |
+
# step=.01)
|
130 |
+
calculate_check = container_param.checkbox(label='Calculate', value=False)
|
131 |
+
if calculate_check:
|
132 |
+
emotion_frequency_slider, emotion_frequency=set_input(container_param,
|
133 |
+
label='Emotion Frequency', key_slider='emotion_frequency_slider_slider', key_input='emotion_frequency_slider_input',
|
134 |
+
min_value=0.,
|
135 |
+
max_value=1.,
|
136 |
+
value=.5,
|
137 |
+
step=.01,)
|
138 |
+
rand_slider, rand=set_input(container_param,
|
139 |
+
label='Weight w', key_slider='rand_slider', key_input='rand_input',
|
140 |
+
min_value=0.,
|
141 |
+
max_value=1.,
|
142 |
+
value=.5,
|
143 |
+
step=.01,)
|
144 |
+
else:
|
145 |
+
emotion_frequency='emotion_frequency'
|
146 |
+
rand=None
|
147 |
+
st.markdown(get_equation_text(w=w, ec=score, rand=rand, emotion_frequency=emotion_frequency))
|
148 |
+
fig = go.Figure()
|
149 |
+
# fig.add_trace(go.Scatter(x=xdata1, y=np.ones_like(xdata1)*rand,
|
150 |
+
# mode='markers', name='Random',
|
151 |
+
# line=dict(color='#ff8300', width=2)
|
152 |
+
# ))
|
153 |
+
if calculate_check:
|
154 |
+
dd = 0.01
|
155 |
+
fig.add_hline(y=rand, line_width=3, line_dash="dash", line_color="#ff8300")
|
156 |
+
fig.add_vline(x=emotion_frequency, line_width=3, line_dash="dash", line_color="green")
|
157 |
+
fig.add_trace(go.Scatter(
|
158 |
+
x=[emotion_frequency-dd, emotion_frequency+dd], y=[rand-dd, rand+dd],
|
159 |
+
mode='lines',
|
160 |
+
line=dict(color='#ee00ee', width=8)
|
161 |
+
),)
|
162 |
+
fig.add_trace(go.Scatter(
|
163 |
+
x=[emotion_frequency+dd, emotion_frequency-dd], y=[rand-dd, rand+dd],
|
164 |
+
mode='lines',
|
165 |
+
line=dict(color='#ee00ee', width=8)
|
166 |
+
),)
|
167 |
+
fig.add_trace(go.Scatter(
|
168 |
+
x=xdata1, y=get_pe(w=w, f=xdata1, ec=score),
|
169 |
+
mode='lines',
|
170 |
+
name='Probability-Emote',
|
171 |
+
line=dict(color='#00eeee', width=4)
|
172 |
+
),
|
173 |
+
)
|
174 |
+
|
175 |
+
fig.update_layout(
|
176 |
+
template='plotly_dark',
|
177 |
+
xaxis_range=[0., 1.],
|
178 |
+
yaxis_range=[0., 1.],
|
179 |
+
xaxis_title="Emotion Frequency",
|
180 |
+
yaxis_title="Probability Emote",
|
181 |
+
showlegend=False
|
182 |
+
)
|
183 |
+
st.plotly_chart(fig, use_container_width=True)
|
184 |
+
st.markdown(get_text())
|
src/story_gen.py
ADDED
@@ -0,0 +1,221 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
import sys
|
3 |
+
import time
|
4 |
+
|
5 |
+
import printj
|
6 |
+
from transformers import pipeline # , set_seed
|
7 |
+
import numpy as np
|
8 |
+
import pandas as pd
|
9 |
+
# import nltk
|
10 |
+
import re
|
11 |
+
import streamlit as st
|
12 |
+
|
13 |
+
|
14 |
+
class StoryGenerator:
|
15 |
+
def __init__(self):
|
16 |
+
self.initialise_models()
|
17 |
+
self.stats_df = pd.DataFrame(data=[], columns=[])
|
18 |
+
self.stories = []
|
19 |
+
self.data = []
|
20 |
+
|
21 |
+
@staticmethod
|
22 |
+
@st.cache(allow_output_mutation=True)
|
23 |
+
def get_generator():
|
24 |
+
return pipeline('text-generation', model='gpt2')
|
25 |
+
|
26 |
+
@staticmethod
|
27 |
+
@st.cache(allow_output_mutation=True)
|
28 |
+
def get_classifier():
|
29 |
+
return pipeline("text-classification",
|
30 |
+
model="j-hartmann/emotion-english-distilroberta-base", return_all_scores=True)
|
31 |
+
|
32 |
+
def initialise_models(self):
|
33 |
+
# start = time.time()
|
34 |
+
self.generator = self.get_generator()
|
35 |
+
self.classifier = self.get_classifier()
|
36 |
+
# initialising_time = time.time()-start
|
37 |
+
# print(f'Initialising Time: {initialising_time}')
|
38 |
+
# set_seed(42)
|
39 |
+
# sys.exit()
|
40 |
+
|
41 |
+
def reset():
|
42 |
+
self.clear_stories()
|
43 |
+
self.clear_stats()
|
44 |
+
|
45 |
+
def clear_stories(self):
|
46 |
+
self.data = []
|
47 |
+
self.stories = []
|
48 |
+
|
49 |
+
def clear_stats(self):
|
50 |
+
self.stats_df = pd.DataFrame(data=[], columns=[])
|
51 |
+
|
52 |
+
def get_emotion(self, text):
|
53 |
+
emotions = self.classifier(text)
|
54 |
+
emotion = max(emotions[0], key=lambda x: x['score'])
|
55 |
+
return emotion
|
56 |
+
|
57 |
+
@staticmethod
|
58 |
+
def get_num_token(text):
|
59 |
+
# return len(nltk.word_tokenize(text))
|
60 |
+
return len(re.findall(r'\w+', text))
|
61 |
+
|
62 |
+
@staticmethod
|
63 |
+
def check_show_emotion(confidence_score, frequency, w):
|
64 |
+
frequency_penalty = 1 - frequency
|
65 |
+
probability_emote = w * confidence_score + (1-w) * frequency_penalty
|
66 |
+
return probability_emote > np.random.random_sample()
|
67 |
+
|
68 |
+
def story(self,
|
69 |
+
story_till_now="Hello, I'm a language model,",
|
70 |
+
num_generation=4,
|
71 |
+
length=10):
|
72 |
+
# last_length = 0
|
73 |
+
|
74 |
+
for i in range(num_generation):
|
75 |
+
last_length = len(story_till_now)
|
76 |
+
genreate_robot_sentence = self.generator(story_till_now, max_length=self.get_num_token(story_till_now) +
|
77 |
+
length, num_return_sequences=1)
|
78 |
+
story_till_now = genreate_robot_sentence[0]['generated_text']
|
79 |
+
new_sentence = story_till_now[last_length:]
|
80 |
+
emotion = self.get_emotion(new_sentence)
|
81 |
+
# printj.yellow(f'Sentence {i}:')
|
82 |
+
# story_to_print = f'{printj.ColorText.cyan(story_till_now[:last_length])}{printj.ColorText.green(story_till_now[last_length:])}\n'
|
83 |
+
# print(story_to_print)
|
84 |
+
# printj.purple(f'Emotion: {emotion}')
|
85 |
+
return story_till_now, emotion
|
86 |
+
|
87 |
+
def next_sentence(self,
|
88 |
+
story_till_now="Hello, I'm a language model,",
|
89 |
+
length=10):
|
90 |
+
last_length = len(story_till_now)
|
91 |
+
genreate_robot_sentence = self.generator(story_till_now, max_length=self.get_num_token(story_till_now) +
|
92 |
+
length, num_return_sequences=1)
|
93 |
+
story_till_now = genreate_robot_sentence[0]['generated_text']
|
94 |
+
new_sentence = story_till_now[last_length:]
|
95 |
+
emotion = self.get_emotion(new_sentence)
|
96 |
+
return story_till_now, emotion, new_sentence
|
97 |
+
|
98 |
+
|
99 |
+
def auto_ist(self,
|
100 |
+
story_till_now="Hello, I'm a language model,",
|
101 |
+
num_generation=4,
|
102 |
+
length=20, reaction_weight=0.5):
|
103 |
+
stats_df = pd.DataFrame(data=[], columns=[])
|
104 |
+
stats_dict = dict()
|
105 |
+
num_reactions = 0
|
106 |
+
reaction_frequency = 0
|
107 |
+
emotion = self.get_emotion(story_till_now) # first line emotion
|
108 |
+
story_data = [{
|
109 |
+
'sentence': story_till_now,
|
110 |
+
'turn': 'first',
|
111 |
+
'emotion': emotion['label'],
|
112 |
+
'confidence_score': emotion['score'],
|
113 |
+
}]
|
114 |
+
for i in range(num_generation):
|
115 |
+
# Text generation for User
|
116 |
+
last_length = len(story_till_now)
|
117 |
+
printj.cyan(story_till_now)
|
118 |
+
printj.red.bold_on_white(
|
119 |
+
f'loop: {i}; generate user text; length: {last_length}')
|
120 |
+
genreate_user_sentence = self.generator(story_till_now, max_length=self.get_num_token(
|
121 |
+
story_till_now)+length, num_return_sequences=1)
|
122 |
+
story_till_now = genreate_user_sentence[0]['generated_text']
|
123 |
+
new_sentence_user = story_till_now[last_length:]
|
124 |
+
|
125 |
+
printj.red.bold_on_white(f'loop: {i}; check emotion')
|
126 |
+
# Emotion self.classifier for User
|
127 |
+
emotion_user = self.get_emotion(new_sentence_user)
|
128 |
+
if emotion_user['label'] == 'neutral':
|
129 |
+
show_emotion_user = False
|
130 |
+
else:
|
131 |
+
reaction_frequency = num_reactions/(i+1)
|
132 |
+
show_emotion_user = self.check_show_emotion(
|
133 |
+
confidence_score=emotion_user['score'], frequency=reaction_frequency, w=reaction_weight)
|
134 |
+
if show_emotion_user:
|
135 |
+
num_reactions += 1
|
136 |
+
|
137 |
+
story_data.append({
|
138 |
+
'sentence': new_sentence_user,
|
139 |
+
'turn': 'user',
|
140 |
+
'emotion': emotion_user['label'],
|
141 |
+
'confidence_score': emotion_user['score'],
|
142 |
+
})
|
143 |
+
stats_dict['sentence_no'] = i
|
144 |
+
stats_dict['turn'] = 'user'
|
145 |
+
stats_dict['sentence'] = new_sentence_user
|
146 |
+
stats_dict['show_emotion'] = show_emotion_user
|
147 |
+
stats_dict['emotion_label'] = emotion_user['label']
|
148 |
+
stats_dict['emotion_score'] = emotion_user['score']
|
149 |
+
stats_dict['num_reactions'] = num_reactions
|
150 |
+
stats_dict['reaction_frequency'] = reaction_frequency
|
151 |
+
stats_dict['reaction_weight'] = reaction_weight
|
152 |
+
stats_df = pd.concat(
|
153 |
+
[stats_df, pd.DataFrame(stats_dict, index=[f'idx_{i}'])])
|
154 |
+
# Text generation for Robot
|
155 |
+
last_length = len(story_till_now)
|
156 |
+
printj.cyan(story_till_now)
|
157 |
+
printj.red.bold_on_white(
|
158 |
+
f'loop: {i}; generate robot text; length: {last_length}')
|
159 |
+
genreate_robot_sentence = self.generator(story_till_now, max_length=self.get_num_token(
|
160 |
+
story_till_now)+length, num_return_sequences=1)
|
161 |
+
story_till_now = genreate_robot_sentence[0]['generated_text']
|
162 |
+
new_sentence_robot = story_till_now[last_length:]
|
163 |
+
emotion_robot = self.get_emotion(new_sentence_robot)
|
164 |
+
|
165 |
+
story_data.append({
|
166 |
+
'sentence': new_sentence_robot,
|
167 |
+
'turn': 'robot',
|
168 |
+
'emotion': emotion_robot['label'],
|
169 |
+
'confidence_score': emotion_robot['score'],
|
170 |
+
})
|
171 |
+
stats_dict['sentence_no'] = i
|
172 |
+
stats_dict['turn'] = 'robot'
|
173 |
+
stats_dict['sentence'] = new_sentence_robot
|
174 |
+
stats_dict['show_emotion'] = None
|
175 |
+
stats_dict['emotion_label'] = emotion_robot['label']
|
176 |
+
stats_dict['emotion_score'] = emotion_robot['score']
|
177 |
+
stats_dict['num_reactions'] = None
|
178 |
+
stats_dict['reaction_frequency'] = None
|
179 |
+
stats_dict['reaction_weight'] = None
|
180 |
+
stats_df = pd.concat(
|
181 |
+
[stats_df, pd.DataFrame(stats_dict, index=[f'idx_{i}'])])
|
182 |
+
|
183 |
+
return stats_df, story_till_now, story_data
|
184 |
+
|
185 |
+
def get_stats(self,
|
186 |
+
story_till_now="Hello, I'm a language model,",
|
187 |
+
num_generation=4,
|
188 |
+
length=20, reaction_weight=-1, num_tests=2):
|
189 |
+
use_random_w = reaction_weight == -1
|
190 |
+
# self.stories = []
|
191 |
+
try:
|
192 |
+
num_rows = max(self.stats_df.story_id)+1
|
193 |
+
except Exception:
|
194 |
+
num_rows = 0
|
195 |
+
for story_id in range(num_tests):
|
196 |
+
if use_random_w:
|
197 |
+
# reaction_weight = np.random.random_sample()
|
198 |
+
reaction_weight = np.round(np.random.random_sample(), 1)
|
199 |
+
stats_df0, _story_till_now, story_data = self.auto_ist(
|
200 |
+
story_till_now=story_till_now,
|
201 |
+
num_generation=num_generation,
|
202 |
+
length=length, reaction_weight=reaction_weight)
|
203 |
+
stats_df0.insert(loc=0, column='story_id', value=story_id+num_rows)
|
204 |
+
|
205 |
+
# stats_df0['story_id'] = story_id
|
206 |
+
self.stats_df = pd.concat([self.stats_df, stats_df0])
|
207 |
+
printj.yellow(f'story_id: {story_id}')
|
208 |
+
printj.green(stats_df0)
|
209 |
+
self.stories.append(_story_till_now)
|
210 |
+
self.data.append(story_data)
|
211 |
+
self.stats_df = self.stats_df.reset_index(drop=True)
|
212 |
+
print(self.stats_df)
|
213 |
+
|
214 |
+
def save_stats(self, path='pandas_simple.xlsx'):
|
215 |
+
writer = pd.ExcelWriter(path, engine='xlsxwriter')
|
216 |
+
|
217 |
+
# Convert the dataframe to an XlsxWriter Excel object.
|
218 |
+
self.stats_df.to_excel(writer, sheet_name='IST')
|
219 |
+
|
220 |
+
# Close the Pandas Excel writer and output the Excel file.
|
221 |
+
writer.save()
|
src/story_gen_test.py
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# %%
|
2 |
+
import printj
|
3 |
+
from story_gen import StoryGenerator
|
4 |
+
|
5 |
+
gen = StoryGenerator()
|
6 |
+
# # %%
|
7 |
+
# story_till_now, emotion = gen.story(story_till_now='Hello, I\'m a language model,', num_generation=3, length=10)
|
8 |
+
# printj.purple(story_till_now)
|
9 |
+
# printj.yellow(emotion)
|
10 |
+
|
11 |
+
|
12 |
+
# %%
|
13 |
+
gen.get_stats(story_till_now="For myriad of eons i’ve forgotten who I really was, harvesting the essence of all existence.",
|
14 |
+
length=10, num_generation=3, num_tests=50)
|
15 |
+
|
16 |
+
# %%
|
17 |
+
gen.save_stats('/home/jitesh/haru/ist/results/a.xlsx')
|
18 |
+
|
19 |
+
|
20 |
+
|
21 |
+
|
22 |
+
# %%
|
23 |
+
data=gen.stats_df[gen.stats_df.sentence_no==3]
|
24 |
+
import seaborn as sns
|
25 |
+
sns.set_theme(style="whitegrid")
|
26 |
+
# ax = sns.violinplot(x="day", y="total_bill", data=tips)
|
27 |
+
ax = sns.violinplot(x="reaction_weight", y="num_reactions", data=data).set_title('Analysing ProbabilityEmote (Max reactions=3)')
|
28 |
+
# %%
|
29 |
+
|
30 |
+
gen.stats_df[gen.stats_df.sentence_no==3]
|
31 |
+
# %%
|
32 |
+
import re
|
33 |
+
len(re.findall(r'\w+', 'line ive '))
|
34 |
+
# %%
|