Spaces:
Sleeping
Sleeping
File size: 13,219 Bytes
4067b64 f1155ac 4067b64 f1155ac 4067b64 f1155ac 4067b64 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 |
import re
from src.models.workflow_graph import Edge, Node, Graph
class HelperClass:
@staticmethod
def _build_prompt(project_desc: str, modules: list) -> str:
return f'''
You are an advanced AI tasked with constructing a directed graph/flow based on a set of available modules and a project description. Each module in the flow represents a node, and each edge defines the task connecting these nodes.
Your output should adhere strictly to the following rules: Dont give me any code and dont mention 'json' at the top of the response.
There should not be any extra output (even a single word) besides the output required.
The flow of nodes and tasks must be determined by analyzing the provided project description.
The modules chosen must form a complete pipeline suitable for the tasks in the project description.
-Steps-
1. Parse the project description to identify the tasks and operations required to form the flow.
- For each task, determine which module (from the available list) best fits the task description.
- Assign a unique identifier to every instance of a module.
- Example: If the "Train" module is used twice for lets say training 2 different model, name them both Train with a unique id to both of them.
- For each identified node:
- Node ID: Generate a Unique identifier for the module instance (5 digit random string and integer combined and in lower case).
- Module Name: Name of the module from the available list.
Format each node as:
<unique Node ID>Module Name</unique Node ID>
2. Construct Edges Between Nodes
- Determine the logical sequence of tasks from the project description.
- Identify source and target modules for each transition based on the task flow.
- For each connection, output the following information:
- Source Node: The unique ID of the starting module. ( Used the ids for each module generated in Step 1)
- Target Node: The unique ID of the destination module. ( Used the ids for each module generated in Step 1)
- Task Description: A short descripiton of what is happening during the transition.
Format each edge as:
<Edge index>( sourceNode="<Node ID>" | targetNode="<Node ID>" | task="<Task Description>" )<
######################
-Examples-
######################
Example 1:
Input: Project Description:
This project implements an automated quality control system for manufacturing using a modular machine learning pipeline. Data from high-resolution product images and metadata is ingested and augmented to enhance diversity and balance.
Task A trains a CNN for defect detection, while Task B trains a transformer-based model for quality classification. Both models are rigorously evaluated and compared against predefined benchmarks. Successful models are deployed for real-time defect monitoring and automated grading via integration with production and ERP systems.
Input: Available Modules:
['IngestData',
'AugmentData',
'GenerateData',
'SearchData',
'Train',
'Evaluate',
'TriggerDeployment',
'ComparePerformance']
-------------------------------------
Flow Generated by LLM: ( This will not be Input )
IngestData -> AugmentData
AugmentData -> Train (for Task A)
AugmentData -> Train (for Task B)
Train (model A) -> Evaluate (test model A)
Train (model B) -> Evaluate (test model B)
Evaluate (test model A) -> ComparePerformance
Evaluate (test model B) -> ComparePerformance
ComparePerformance -> TriggerDeployment
################
Output:
<p83fd>IngestData</p83fd>
<sb9ba>AugmentData</sb9ba>
<bxt2w>Train A</bxt2w>
<d1ep3>Train B</d1ep3>
<b9lca>Evaluate A</b9lca>
<5w01f>Evaluate B</5w01f>
<z4bun>ComparePerformance</z4bun>
<zj2pb>TriggerDeployment</zj2pb>
<Edge 1>( sourceNode="<p83fd>" | targetNode="<sb9ba>" | task = "Ingesting Data to Augment Data" )</Edge 1>
<Edge 2>( sourceNode="<sb9ba>" | targetNode="<bxt2w>" | task = "Augmenting Data to Train model A" )</Edge 2>
<Edge 3>( sourceNode="<sb9ba>" | targetNode="<d1ep3>" | task = "Augmenting Data to Train model B" )</Edge 3>
<Edge 4>( sourceNode="<bxt2w>" | targetNode="<b9lca>" | task = "Training A to Evaluate model A" )</Edge 4>
<Edge 5>( sourceNode="<d1ep3>" | targetNode="<5w01f>" | task = "Training B to Evaluate model B" )</Edge 5>
<Edge 6>( sourceNode="<b9lca>" | targetNode="<z4bun>" | task = "Evaluate model A to Compare Performance" )</Edge 6>
<Edge 7>( sourceNode="<5w01f>" | targetNode="<z4bun>" | task = "Evaluate model B to Compare Performance" )</Edge 7>
<Edge 8>( sourceNode="<z4bun>" | targetNode="<zj2pb>" | task = "Compare Performance to Trigger Deployment" )</Edge 8>
#############################
Example 2:
Input: Project Description:
This project develops an automated crop health monitoring system using a modular machine learning pipeline. Data from satellite and drone imagery is ingested and preprocessed, followed by augmentation techniques to increase diversity and balance.
Synthetic data is generated to simulate various crop conditions, enhancing model robustness. The pipeline trains a deep learning model to classify crop health, evaluates its performance on key metrics such as accuracy and recall, and identifies areas for improvement.
Once performance benchmarks are met, the system is deployed for real-time crop monitoring, enabling farmers to make informed decisions and optimize agricultural productivity efficiently.
Input: Available Modules:
['IngestData',
'AugmentData',
'GenerateData',
'SearchData',
'Train',
'Evaluate',
'TriggerDeployment',
'ComparePerformance']
-------------------------------------
Flow Generated by LLM: ( This will not be Input )
IngestData -> AugmentData
AugmentData -> GenerateData
GenerateData -> Train
Train -> Evaluate
Evaluate -> TriggerDeployment
################
Output:
<p001>IngestData</p001>
<p002>AugmentData</p002>
<p003>GenerateData</p003>
<p004>Train</p004>
<p005>Evaluate</p005>
<p006>TriggerDeployment</p006>
<Edge 1>( sourceNode="<p001>" | targetNode="<p002>" | task="Ingesting Data to Augmenting Data" )</Edge 1>
<Edge 2>( sourceNode="<p002>" | targetNode="<p003>" | task="Augmenting Data to Generating Synthetic Data" )</Edge 2>
<Edge 3>( sourceNode="<p003>" | targetNode="<p004>" | task="Generating Data to Training Model" )</Edge 3>
<Edge 4>( sourceNode="<p004>" | targetNode="<p005>" | task="Training Model to Evaluating Performance" )</Edge 4>
<Edge 5>( sourceNode="<p005>" | targetNode="<p006>" | task="Evaluating Model to Triggering Deployment" )</Edge 5>
#############################
Example 3:
Input: Project Descripiont:
This project implements a robust machine learning pipeline for iterative model improvement. Data is ingested and preprocessed, followed by augmentation to enhance diversity and balance.
An initial model is trained on the augmented data. The pipeline then applies further data augmentation techniques tailored to improve underperforming areas, followed by retraining the model for enhanced accuracy.
The improved model is rigorously evaluated on a test dataset to ensure it meets predefined performance benchmarks. Upon achieving the desired metrics, the best-performing model is deployed to production, ensuring reliable and efficient real-world performance tailored to the project's objectives.
Input: Available Modules:
['IngestData',
'AugmentData',
'GenerateData',
'SearchData',
'Train',
'Evaluate',
'TriggerDeployment',
'ComparePerformance']
-------------------------------------
Flow Generated by LLM: ( This will not be Input )
IngestData -> AugmentData (Stage 1)
AugmentData (Stage 1) -> Train (Stage 1)
Train (Stage 1) -> AugmentData (Stage 2)
AugmentData (Stage 2) -> Train (Stage 2)
Train (Stage 2) -> Evaluate
Evaluate -> TriggerDeployment
################
Output:
<m001>IngestData</m001>
<m002>AugmentData Stage 1</m002>
<m003>Train Stage 1</m003>
<m004>AugmentData Stage 2</m004>
<m005>Train Stage 2</m005>
<m006>Evaluate</m006>
<m007>TriggerDeployment</m007>
<Edge 1>( sourceNode="<m001>" | targetNode="<m002>" | task="Ingesting Data to Augmenting Data Stage 1" )</Edge 1>
<Edge 2>( sourceNode="<m002>" | targetNode="<m003>" | task="Augmenting Data Stage 1 to Training Stage 1" )</Edge 2>
<Edge 3>( sourceNode="<m003>" | targetNode="<m004>" | task="Training Stage 1 to Augmenting Data Stage 2" )</Edge 3>
<Edge 4>( sourceNode="<m004>" | targetNode="<m005>" | task="Augmenting Data Stage 2 to Training Stage 2" )</Edge 4>
<Edge 5>( sourceNode="<m005>" | targetNode="<m006>" | task="Training Stage 2 to Evaluating Model" )</Edge 6>
<Edge 6>( sourceNode="<m006>" | targetNode="<m007>" | task="Evaluating Model to Triggering Deployment" )</Edge 6>
#############################
When you give output dont mention anything like 'Here is the list of Nodes and Edges extracted from the text:'. Just give the response straight away
-Real Data-
######################
Input: Project Descripion: {project_desc}
**Instructions**
1. A list of modules available for building the pipeline. You must only use these modules to form the flow.
2. Do not Generate New Names for Modules. Only use whatever is available in the list
Input: Available Modules:
{modules}
######################
Output:
'''
@staticmethod
def _parse_llm_response(raw_response: str) -> Graph:
pattern = r'<([a-zA-Z0-9]+)>([^<]+)<\/\1>|sourceNode="<([^"]+)>"\s*\|\s*targetNode="<([^"]+)>"\s*\|\s*task="([^"]+)"'
nodes, edges = [], []
list_ = raw_response.split('\n')
for line in list_:
matches = re.findall(pattern, line)
try:
for match in matches:
if match[0]:
nd = Node(node_id=match[0], name=match[1])
nodes.append(nd)
elif match[2]:
edge = Edge(source=match[2], target=match[3], desc=match[4])
edges.append(edge)
except Exception as e:
print(f"Error parsing line : {line}, error: {e}")
return Graph(nodes=nodes, edges=edges)
@staticmethod
def _store_graph(graph_data: Graph):
nodes, edges = [], []
dict_ = {}
for node in graph_data.nodes:
dict_[node.node_id] = node.name
nodes.append({
'node_id': node.node_id,
'name': node.name
})
dict_['Start'] = 'StartNode'
for edge in graph_data.edges:
source_node = dict_[edge.source]
target_node = dict_[edge.target]
edges.append({
'source': edge.source,
'target': edge.target,
'desc': edge.desc
})
json_obj = {'Nodes': nodes, 'Edges': edges}
helper = HelperClass()
|