Spaces:
Running
Running
DrishtiSharma
commited on
Update interim.py
Browse files- interim.py +145 -46
interim.py
CHANGED
@@ -4,12 +4,13 @@ import sqlite3
|
|
4 |
import os
|
5 |
import json
|
6 |
from pathlib import Path
|
|
|
7 |
from datetime import datetime, timezone
|
8 |
from crewai import Agent, Crew, Process, Task
|
9 |
-
from
|
10 |
from langchain_groq import ChatGroq
|
|
|
11 |
from langchain.schema.output import LLMResult
|
12 |
-
from langchain_core.callbacks.base import BaseCallbackHandler
|
13 |
from langchain_community.tools.sql_database.tool import (
|
14 |
InfoSQLDatabaseTool,
|
15 |
ListSQLDatabaseTool,
|
@@ -20,39 +21,41 @@ from langchain_community.utilities.sql_database import SQLDatabase
|
|
20 |
from datasets import load_dataset
|
21 |
import tempfile
|
22 |
|
23 |
-
|
24 |
-
|
25 |
|
26 |
# Initialize LLM
|
27 |
-
|
28 |
-
def __init__(self, log_path: Path):
|
29 |
-
self.log_path = log_path
|
30 |
-
|
31 |
-
def on_llm_start(self, serialized, prompts, **kwargs):
|
32 |
-
with self.log_path.open("a", encoding="utf-8") as file:
|
33 |
-
file.write(json.dumps({"event": "llm_start", "text": prompts[0], "timestamp": datetime.now().isoformat()}) + "\n")
|
34 |
-
|
35 |
-
def on_llm_end(self, response: LLMResult, **kwargs):
|
36 |
-
generation = response.generations[-1][-1].message.content
|
37 |
-
with self.log_path.open("a", encoding="utf-8") as file:
|
38 |
-
file.write(json.dumps({"event": "llm_end", "text": generation, "timestamp": datetime.now().isoformat()}) + "\n")
|
39 |
-
|
40 |
-
llm = ChatGroq(
|
41 |
-
temperature=0,
|
42 |
-
model_name="groq/llama-3.3-70b-versatile",
|
43 |
-
max_tokens=1024,
|
44 |
-
callbacks=[LLMCallbackHandler(Path("prompts.jsonl"))],
|
45 |
-
)
|
46 |
|
47 |
-
|
48 |
-
st.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
|
50 |
# Initialize session state for data persistence
|
51 |
if "df" not in st.session_state:
|
52 |
st.session_state.df = None
|
|
|
|
|
53 |
|
54 |
# Dataset Input
|
55 |
input_option = st.radio("Select Dataset Input:", ["Use Hugging Face Dataset", "Upload CSV File"])
|
|
|
56 |
if input_option == "Use Hugging Face Dataset":
|
57 |
dataset_name = st.text_input("Enter Hugging Face Dataset Name:", value="Einstellung/demo-salaries")
|
58 |
if st.button("Load Dataset"):
|
@@ -60,16 +63,25 @@ if input_option == "Use Hugging Face Dataset":
|
|
60 |
with st.spinner("Loading dataset..."):
|
61 |
dataset = load_dataset(dataset_name, split="train")
|
62 |
st.session_state.df = pd.DataFrame(dataset)
|
|
|
63 |
st.success(f"Dataset '{dataset_name}' loaded successfully!")
|
64 |
-
st.dataframe(st.session_state.df.head())
|
65 |
except Exception as e:
|
66 |
st.error(f"Error: {e}")
|
|
|
67 |
elif input_option == "Upload CSV File":
|
68 |
uploaded_file = st.file_uploader("Upload CSV File:", type=["csv"])
|
69 |
if uploaded_file:
|
70 |
-
|
71 |
-
|
72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
# SQL-RAG Analysis
|
75 |
if st.session_state.df is not None:
|
@@ -86,19 +98,20 @@ if st.session_state.df is not None:
|
|
86 |
|
87 |
@tool("tables_schema")
|
88 |
def tables_schema(tables: str) -> str:
|
89 |
-
"""Get schema and sample rows for
|
90 |
return InfoSQLDatabaseTool(db=db).invoke(tables)
|
91 |
|
92 |
@tool("execute_sql")
|
93 |
def execute_sql(sql_query: str) -> str:
|
94 |
-
"""Execute a SQL query against the database."""
|
95 |
return QuerySQLDataBaseTool(db=db).invoke(sql_query)
|
96 |
|
97 |
@tool("check_sql")
|
98 |
def check_sql(sql_query: str) -> str:
|
99 |
-
"""
|
100 |
return QuerySQLCheckerTool(db=db, llm=llm).invoke({"query": sql_query})
|
101 |
|
|
|
102 |
sql_dev = Agent(
|
103 |
role="Senior Database Developer",
|
104 |
goal="Extract data using optimized SQL queries.",
|
@@ -116,11 +129,19 @@ if st.session_state.df is not None:
|
|
116 |
|
117 |
report_writer = Agent(
|
118 |
role="Technical Report Writer",
|
119 |
-
goal="
|
120 |
-
backstory="
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
121 |
llm=llm,
|
122 |
)
|
123 |
|
|
|
124 |
extract_data = Task(
|
125 |
description="Extract data based on the query: {query}.",
|
126 |
expected_output="Database results matching the query.",
|
@@ -129,33 +150,111 @@ if st.session_state.df is not None:
|
|
129 |
|
130 |
analyze_data = Task(
|
131 |
description="Analyze the extracted data for query: {query}.",
|
132 |
-
expected_output="Analysis
|
133 |
agent=data_analyst,
|
134 |
context=[extract_data],
|
135 |
)
|
136 |
|
137 |
write_report = Task(
|
138 |
-
description="
|
139 |
-
expected_output="Markdown report
|
140 |
agent=report_writer,
|
141 |
context=[analyze_data],
|
142 |
)
|
143 |
|
144 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
145 |
agents=[sql_dev, data_analyst, report_writer],
|
146 |
tasks=[extract_data, analyze_data, write_report],
|
147 |
process=Process.sequential,
|
148 |
verbose=True,
|
149 |
)
|
150 |
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
158 |
|
159 |
temp_dir.cleanup()
|
160 |
else:
|
161 |
-
st.info("Please load a dataset to proceed.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
import os
|
5 |
import json
|
6 |
from pathlib import Path
|
7 |
+
import plotly.express as px
|
8 |
from datetime import datetime, timezone
|
9 |
from crewai import Agent, Crew, Process, Task
|
10 |
+
from crewai.tools import tool
|
11 |
from langchain_groq import ChatGroq
|
12 |
+
from langchain_openai import ChatOpenAI
|
13 |
from langchain.schema.output import LLMResult
|
|
|
14 |
from langchain_community.tools.sql_database.tool import (
|
15 |
InfoSQLDatabaseTool,
|
16 |
ListSQLDatabaseTool,
|
|
|
21 |
from datasets import load_dataset
|
22 |
import tempfile
|
23 |
|
24 |
+
st.title("SQL-RAG Using CrewAI π")
|
25 |
+
st.write("Analyze datasets using natural language queries powered by SQL and CrewAI.")
|
26 |
|
27 |
# Initialize LLM
|
28 |
+
llm = None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
|
30 |
+
# Model Selection
|
31 |
+
model_choice = st.radio("Select LLM", ["GPT-4o", "llama-3.3-70b"], index=0, horizontal=True)
|
32 |
+
|
33 |
+
# API Key Validation and LLM Initialization
|
34 |
+
groq_api_key = os.getenv("GROQ_API_KEY")
|
35 |
+
openai_api_key = os.getenv("OPENAI_API_KEY")
|
36 |
+
|
37 |
+
if model_choice == "llama-3.3-70b":
|
38 |
+
if not groq_api_key:
|
39 |
+
st.error("Groq API key is missing. Please set the GROQ_API_KEY environment variable.")
|
40 |
+
llm = None
|
41 |
+
else:
|
42 |
+
llm = ChatGroq(groq_api_key=groq_api_key, model="groq/llama-3.3-70b-versatile")
|
43 |
+
elif model_choice == "GPT-4o":
|
44 |
+
if not openai_api_key:
|
45 |
+
st.error("OpenAI API key is missing. Please set the OPENAI_API_KEY environment variable.")
|
46 |
+
llm = None
|
47 |
+
else:
|
48 |
+
llm = ChatOpenAI(api_key=openai_api_key, model="gpt-4o")
|
49 |
|
50 |
# Initialize session state for data persistence
|
51 |
if "df" not in st.session_state:
|
52 |
st.session_state.df = None
|
53 |
+
if "show_preview" not in st.session_state:
|
54 |
+
st.session_state.show_preview = False
|
55 |
|
56 |
# Dataset Input
|
57 |
input_option = st.radio("Select Dataset Input:", ["Use Hugging Face Dataset", "Upload CSV File"])
|
58 |
+
|
59 |
if input_option == "Use Hugging Face Dataset":
|
60 |
dataset_name = st.text_input("Enter Hugging Face Dataset Name:", value="Einstellung/demo-salaries")
|
61 |
if st.button("Load Dataset"):
|
|
|
63 |
with st.spinner("Loading dataset..."):
|
64 |
dataset = load_dataset(dataset_name, split="train")
|
65 |
st.session_state.df = pd.DataFrame(dataset)
|
66 |
+
st.session_state.show_preview = True # Show preview after loading
|
67 |
st.success(f"Dataset '{dataset_name}' loaded successfully!")
|
|
|
68 |
except Exception as e:
|
69 |
st.error(f"Error: {e}")
|
70 |
+
|
71 |
elif input_option == "Upload CSV File":
|
72 |
uploaded_file = st.file_uploader("Upload CSV File:", type=["csv"])
|
73 |
if uploaded_file:
|
74 |
+
try:
|
75 |
+
st.session_state.df = pd.read_csv(uploaded_file)
|
76 |
+
st.session_state.show_preview = True # Show preview after loading
|
77 |
+
st.success("File uploaded successfully!")
|
78 |
+
except Exception as e:
|
79 |
+
st.error(f"Error loading file: {e}")
|
80 |
+
|
81 |
+
# Show Dataset Preview Only After Loading
|
82 |
+
if st.session_state.df is not None and st.session_state.show_preview:
|
83 |
+
st.subheader("π Dataset Preview")
|
84 |
+
st.dataframe(st.session_state.df.head())
|
85 |
|
86 |
# SQL-RAG Analysis
|
87 |
if st.session_state.df is not None:
|
|
|
98 |
|
99 |
@tool("tables_schema")
|
100 |
def tables_schema(tables: str) -> str:
|
101 |
+
"""Get the schema and sample rows for the specified tables."""
|
102 |
return InfoSQLDatabaseTool(db=db).invoke(tables)
|
103 |
|
104 |
@tool("execute_sql")
|
105 |
def execute_sql(sql_query: str) -> str:
|
106 |
+
"""Execute a SQL query against the database and return the results."""
|
107 |
return QuerySQLDataBaseTool(db=db).invoke(sql_query)
|
108 |
|
109 |
@tool("check_sql")
|
110 |
def check_sql(sql_query: str) -> str:
|
111 |
+
"""Validate the SQL query syntax and structure before execution."""
|
112 |
return QuerySQLCheckerTool(db=db, llm=llm).invoke({"query": sql_query})
|
113 |
|
114 |
+
# Agents for SQL data extraction and analysis
|
115 |
sql_dev = Agent(
|
116 |
role="Senior Database Developer",
|
117 |
goal="Extract data using optimized SQL queries.",
|
|
|
129 |
|
130 |
report_writer = Agent(
|
131 |
role="Technical Report Writer",
|
132 |
+
goal="Write a structured report with Key Insights and Analysis. DO NOT include Introduction or Conclusion.",
|
133 |
+
backstory="Specializes in detailed analytical reports without conclusions.",
|
134 |
+
llm=llm,
|
135 |
+
)
|
136 |
+
|
137 |
+
conclusion_writer = Agent(
|
138 |
+
role="Conclusion Specialist",
|
139 |
+
goal="Summarize findings into a clear and concise 3-5 line Conclusion highlighting only the most important insights.",
|
140 |
+
backstory="An expert in crafting impactful and clear conclusions.",
|
141 |
llm=llm,
|
142 |
)
|
143 |
|
144 |
+
# Define tasks for report and conclusion
|
145 |
extract_data = Task(
|
146 |
description="Extract data based on the query: {query}.",
|
147 |
expected_output="Database results matching the query.",
|
|
|
150 |
|
151 |
analyze_data = Task(
|
152 |
description="Analyze the extracted data for query: {query}.",
|
153 |
+
expected_output="Key Insights and Analysis without any Introduction or Conclusion.",
|
154 |
agent=data_analyst,
|
155 |
context=[extract_data],
|
156 |
)
|
157 |
|
158 |
write_report = Task(
|
159 |
+
description="Write the analysis report with Key Insights. DO NOT include a Conclusion.",
|
160 |
+
expected_output="Markdown-formatted report excluding Conclusion.",
|
161 |
agent=report_writer,
|
162 |
context=[analyze_data],
|
163 |
)
|
164 |
|
165 |
+
write_conclusion = Task(
|
166 |
+
description="Write a brief and impactful 3-5 line Conclusion summarizing only the most important insights.",
|
167 |
+
expected_output="Markdown-formatted concise Conclusion section.",
|
168 |
+
agent=conclusion_writer,
|
169 |
+
context=[analyze_data],
|
170 |
+
)
|
171 |
+
|
172 |
+
# Separate Crews for report and conclusion
|
173 |
+
crew_report = Crew(
|
174 |
agents=[sql_dev, data_analyst, report_writer],
|
175 |
tasks=[extract_data, analyze_data, write_report],
|
176 |
process=Process.sequential,
|
177 |
verbose=True,
|
178 |
)
|
179 |
|
180 |
+
crew_conclusion = Crew(
|
181 |
+
agents=[data_analyst, conclusion_writer],
|
182 |
+
tasks=[write_conclusion],
|
183 |
+
process=Process.sequential,
|
184 |
+
verbose=True,
|
185 |
+
)
|
186 |
+
|
187 |
+
# Tabs for Query Results and Visualizations
|
188 |
+
tab1, tab2 = st.tabs(["π Query Insights + Viz", "π Full Data Viz"])
|
189 |
+
|
190 |
+
# Query Insights + Visualization
|
191 |
+
with tab1:
|
192 |
+
query = st.text_area("Enter Query:", value="Provide insights into the salary of a Principal Data Scientist.")
|
193 |
+
if st.button("Submit Query"):
|
194 |
+
with st.spinner("Processing query..."):
|
195 |
+
# Step 1: Generate the analysis report
|
196 |
+
report_inputs = {"query": query + " Provide detailed analysis but DO NOT include Conclusion."}
|
197 |
+
report_result = crew_report.kickoff(inputs=report_inputs)
|
198 |
+
|
199 |
+
# Step 2: Generate only the concise conclusion
|
200 |
+
conclusion_inputs = {"query": query + " Provide ONLY the most important insights in 3-5 concise lines."}
|
201 |
+
conclusion_result = crew_conclusion.kickoff(inputs=conclusion_inputs)
|
202 |
+
|
203 |
+
# Step 3: Display the report
|
204 |
+
st.markdown("### Analysis Report:")
|
205 |
+
st.markdown(report_result if report_result else "β οΈ No Report Generated.")
|
206 |
+
|
207 |
+
# Step 4: Generate Visualizations
|
208 |
+
visualizations = []
|
209 |
+
|
210 |
+
fig_salary = px.box(st.session_state.df, x="job_title", y="salary_in_usd",
|
211 |
+
title="Salary Distribution by Job Title")
|
212 |
+
visualizations.append(fig_salary)
|
213 |
+
|
214 |
+
fig_experience = px.bar(
|
215 |
+
st.session_state.df.groupby("experience_level")["salary_in_usd"].mean().reset_index(),
|
216 |
+
x="experience_level", y="salary_in_usd",
|
217 |
+
title="Average Salary by Experience Level"
|
218 |
+
)
|
219 |
+
visualizations.append(fig_experience)
|
220 |
+
|
221 |
+
fig_employment = px.box(st.session_state.df, x="employment_type", y="salary_in_usd",
|
222 |
+
title="Salary Distribution by Employment Type")
|
223 |
+
visualizations.append(fig_employment)
|
224 |
+
|
225 |
+
# Step 5: Insert Visual Insights
|
226 |
+
st.markdown("## π Visual Insights")
|
227 |
+
for fig in visualizations:
|
228 |
+
st.plotly_chart(fig, use_container_width=True)
|
229 |
+
|
230 |
+
# Step 6: Display Concise Conclusion
|
231 |
+
st.markdown("## Conclusion")
|
232 |
+
st.markdown(conclusion_result if conclusion_result else "β οΈ No Conclusion Generated.")
|
233 |
+
|
234 |
+
# Full Data Visualization Tab
|
235 |
+
with tab2:
|
236 |
+
st.subheader("π Comprehensive Data Visualizations")
|
237 |
+
|
238 |
+
fig1 = px.histogram(st.session_state.df, x="job_title", title="Job Title Frequency")
|
239 |
+
st.plotly_chart(fig1)
|
240 |
+
|
241 |
+
fig2 = px.bar(
|
242 |
+
st.session_state.df.groupby("experience_level")["salary_in_usd"].mean().reset_index(),
|
243 |
+
x="experience_level", y="salary_in_usd",
|
244 |
+
title="Average Salary by Experience Level"
|
245 |
+
)
|
246 |
+
st.plotly_chart(fig2)
|
247 |
+
|
248 |
+
fig3 = px.box(st.session_state.df, x="employment_type", y="salary_in_usd",
|
249 |
+
title="Salary Distribution by Employment Type")
|
250 |
+
st.plotly_chart(fig3)
|
251 |
|
252 |
temp_dir.cleanup()
|
253 |
else:
|
254 |
+
st.info("Please load a dataset to proceed.")
|
255 |
+
|
256 |
+
|
257 |
+
# Sidebar Reference
|
258 |
+
with st.sidebar:
|
259 |
+
st.header("π Reference:")
|
260 |
+
st.markdown("[SQL Agents w CrewAI & Llama 3 - Plaban Nayak](https://github.com/plaban1981/Agents/blob/main/SQL_Agents_with_CrewAI_and_Llama_3.ipynb)")
|