File size: 9,476 Bytes
ee107ce
 
 
24ec173
ee107ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24ec173
ee107ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24ec173
 
 
ee107ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
import streamlit as st
import pandas as pd
from langchain.prompts import FewShotPromptTemplate, PromptTemplate
from langchain.chat_models import ChatOpenAI
from langchain.callbacks import get_openai_callback
from langchain.chains import LLMChain


# Function to generate empty dataframe
def create_empty_dataframe(rows, columns):
    data = {column: [""] * rows for column in columns}
    return pd.DataFrame(data)


# Sidebar
st.sidebar.title("Configuration")
selected_model = st.sidebar.selectbox(
    "Select Model",
    [
        "chat-gpt-3.5-turbo",
    ],  # "gpt3-text-davinci-003"
)
selected_dataset = st.sidebar.selectbox(
    "Select Dataset", ["Regulatory Changes", "Dealer Risk", "Custom"]
)
openai_api_key = st.sidebar.text_input("OpenAI API Key", value="", type="password")


prompt_prefix = """You are a news recommender and your job is to look at an article and decide if I would find it useful or not.
{dataset_specific_prompt}
I'm going to pass in a few headlines and publishers. Can you just respond with "Relevant" or "Not Relevant" for the headlines I send you?"""

if selected_dataset == "Regulatory Changes":
    dataset_specific_prompt = """I work in a bank/credit card company and I'm looking for news articles that deal with changes in the regulatory environment. 
It could be articles about regulatory authorities like CFPB or OCC announcing changes in regulation or a new focus areas. 
Or articles like the president or senate or government announcing new regulations. 
Or peer financial institutions getting fined for violations. 
To reiterate I'm especially looking for articles in the finance, fin-tech, banking and credit card sector. 
**I don't need think pieces or opinion pieces or editorials that talk about the consequences of a regulation but just the ones that talk about objective changes in the regulatory environment that might impact my work**"""
    examples = [
        {
            "headline": "CFPB Launches Inquiry Into the Business Practices of Data Brokers",
            "publisher": "CFPB",
            "answer": "Relevant",
        },
        {
            "headline": "OCC Issues Prohibition Order, Fines Former Wells Fargo Executive $17 Million in Settlement",
            "publisher": "NYT",
            "answer": "Relevant",
        },
        {
            "headline": "CFPB Announces Appointments of New Advisory Committee Members",
            "publisher": "CFPB",
            "answer": "Not Relevant",
        },
    ]
    prediction_examples = [
        {
            "headline": "Ray Dalio Commentary: What I Think About the Silicon Valley Bank Situation",
            "publisher": "Yahoo Entertainment",
        },
        {
            "headline": "Financial Regulators, Black History and Epistemic Capital",
            "publisher": "Harvard School of Engineering",
        },
        {
            "headline": "Agencies Issue Joint Statement on Crypto-Asset Risks to Banking Organizations",
            "publisher": "OCC",
        },
        {
            "headline": "Innovating during a regulatory wave",
            "publisher": "Venture beat",
        },
        {
            "headline": "CFPB Proposes Rule to Rein in Excessive Credit Card Late Fees",
            "publisher": "CFPB",
        },
    ]

elif selected_dataset == "Dealer Risk":
    dataset_specific_prompt = """I work in a auto-loan company and I'm looking for news articles that talk about with automobile dealerships being in the news for committing fraud and crime.
I am not interested in just about any crime or fraud but only that which is related to dealerships. 
I am NOT concerned about articles that where the dealership is the victim of a crime. 
I am only interested where the dealerships are the perpetrators of said crime or accusations.
"""
    examples = [
        {
            "headline": "Kentucky Police arrest suspect following $600,000 multiple car heist that took less than 45 seconds",
            "publisher": "Business Insider",
            "answer": "Not Relevant",
        },
        {
            "headline": "Car Dealership workers kidnapped, tortured in Armed Robbery",
            "publisher": "ABC7 Chicago",
            "answer": "Not Relevant",
        },
        {
            "headline": "Auto Dealership Owner Sentenced to 10 Years in Prison for $1.5 Million Fraud Scheme",
            "publisher": "Long Island Business News",
            "answer": "Relevant",
        },
    ]
    prediction_examples = [
        {
            "headline": "New York City Pharmacist charged with running healthcare fraud scheme",
            "publisher": "amnycom",
        },
        {
            "headline": "California man arrested for 3.6 million paycheck protection program fraud",
            "publisher": "DoJ",
        },
        {
            "headline": "Former Moline Businessman sentenced to 41 months in connection with scam to rollback odometeres at used car dealerships",
            "publisher": "DoJ",
        },
        {
            "headline": "Kentucky car dealer convicted in truck warranty fraud scheme",
            "publisher": "apnews.com",
        },
        {
            "headline": "Owner of car dealership pleads guilty to defraduing financing company",
            "publisher": "Automotive News",
        },
    ]

elif selected_dataset == "Custom":
    dataset_specific_prompt = """I work in a XYZ company and I'm looking for news articles that deal with XYZ"""
    examples = [{k: "" for k in ["headline", "publisher", "answer"]} for i in range(3)]
    prediction_examples = [{k: "" for k in ["headline", "publisher"]} for i in range(5)]


prompt_prefix = prompt_prefix.format(
    dataset_specific_prompt=dataset_specific_prompt
).replace("\n", " ")


# Main panel
st.title("News Topic Subscription")

st.header("Dataset")
st.write(f"Prompt data for {selected_dataset}:")

prompt_data = st.text_area(
    "Prompt Prefix", prompt_prefix, placeholder="Enter dataset prompt here", height=300
)

st.header("Few Shot Examples")

columns = ["headline", "publisher", "answer"]
data = pd.DataFrame({k: v for k, v in example.items()} for example in examples)
data.columns = columns
# data["Relevance"] = [""] * 3

for index, row in data.iterrows():
    row["headline"] = st.text_input(
        f"Headline", row["headline"], key=f"headline_{index}"
    )
    row["publisher"] = st.text_input(
        f"Publisher", row["publisher"], key=f"publisher_{index}"
    )
    choice_index = (
        ["Relevant", "Not Relevant"].index(row["answer"]) if row["answer"] != "" else 0
    )
    row["answer"] = st.radio(
        f"Relevance {index + 1}",
        index=choice_index,
        options=["Relevant", "Not Relevant"],
        key=f"answer_{index}",
    )
    st.markdown("---")

st.header("Predictions")

predictions_data = pd.DataFrame(
    {k: v for k, v in example.items()} for example in prediction_examples
)
predictions_data["answer"] = [""] * 5
prediction_labels = []

for index, row in predictions_data.iterrows():
    row["headline"] = st.text_input(
        f"headline", row["headline"], key=f"prediction_headline_{index}"
    )
    row["publisher"] = st.text_input(
        f"publisher", row["publisher"], key=f"prediction_publisher_{index}"
    )
    prediction_labels.append(st.empty())
    st.markdown("---")


llm = ChatOpenAI(
    temperature=0.0, openai_api_key=openai_api_key if openai_api_key != "" else " "
)
example_prompt = PromptTemplate(
    input_variables=["headline", "publisher", "answer"],
    template="Headline: {headline}\nPublisher: {publisher}\nAnswer: {answer}",
)


def update_relevance(predictions_data):

    prompt = FewShotPromptTemplate(
        prefix=prompt_data,
        examples=data.to_dict("records"),
        example_prompt=example_prompt,
        suffix="Headline: {headline}\nPublisher: {publisher}\nAnswer: ",
        input_variables=["headline", "publisher"],
    )
    print(prompt.format(headline="", publisher=""))
    print(data)
    news_predictor_chain = LLMChain(prompt=prompt, llm=llm, verbose=True)
    token_counts = []

    for index, row in predictions_data.iterrows():
        if row["headline"] != "":
            with get_openai_callback() as cb:
                predictions_data.loc[index, "answer"] = news_predictor_chain.run(
                    {
                        "headline": row["headline"],
                        "publisher": row["publisher"]
                        if row["publisher"] != ""
                        else "Unknown",
                    }
                )
                token_counts.append(cb.total_tokens)

        else:
            predictions_data.loc[index, "answer"] = ""
    st.text(
        f"{sum(token_counts)} tokens used in this run for {len(token_counts)} articles. This costed ${sum(token_counts) * 0.002/1000 } @$0.002/1K tokens."
    )
    st.text(
        f"For predicting 1000 articles, it would cost ${sum(token_counts)/len(token_counts) * 0.002 } at the current rate. "
    )
    return predictions_data


if st.button("Run Model", type="primary"):
    predictions_data = update_relevance(predictions_data)
    for index, row in predictions_data.iterrows():
        if row["answer"] != "":
            with prediction_labels[index].container():
                if row["answer"] == "Relevant":
                    st.markdown("This article is **Relevant** :thumbsup:")
                else:
                    st.markdown("This article is **Not Relevant** :thumbsdown:")