Графические вероятностные модели для принятия решений в проектном управлении
Ольга Татаринцева (Data Scientist at Eleks)
Как часто вам приходится принимать решения, используя знания в определенной предметной области? На сколько хороши такие решения? А теперь представьте, что вы собрали знания лучших экспертов в предметной области. Похоже, что ваши решения, основанные на этих знаниях, будут куда более взвешенными, не так ли? Мы будем говорить о системе ProjectHealth, которая была построена на основе опыта лучших экспертов в проектном управлении в компании Eleks. Для реализации поставленной задачи была использована графовая вероятностная модель, а именно байесовская сеть, имплементированная на Python. За время работы над проектом мы прошли шаги от извлечения требований, поиска данных и построения модели с нуля до реализации BI дашборда с возможностью углубиться в детали, доходя до сырых данных. Сейчас ProjectHealth экономит большое количество времени для топ менеджмента и ресурсов компании, так как мониторит состояние бизнеса в малейших деталях ежедневно и делает это как настоящий эксперт.
Все материалы: http://datascience.in.ua/report2017
Report
Share
Report
Share
1 of 32
Download to read offline
More Related Content
DataScience Lab 2017_Графические вероятностные модели для принятия решений в проектном управлении_Ольга Татаринцева
2. • Neural networks
• Deep learning
• Natural language processing
• Image processing
• Big data
• Bioinformatics
Today we are NOT about
3. • Neural networks
• Deep learning
• Natural language processing
• Image processing
• Big data
• Bioinformatics
• Project based business
• Company organization
• Project management and
project success
• Applying the DS methods to
business problems
• Small example
• Results we gained
Today we are NOT about We are about
4. Chief Technical officer
... Delivery Director
... PM
... Architects
Business
Analysts
Developers QA
Data
scientists
...
...
...
Simplified structure of delivery organization
5. Chief Technical officer
... Delivery Director
... PM
... Architects
Business
Analysts
Developers QA
Data
scientists
...
...
...
Simplified structure of delivery organization
PROJECT
7. ProjectTeam
Direct contacts with the
customer
Requirements
management
Solution design
Product development
Project budget planning
Working on improvements
and up-sale opportunities
Management of internal
activities
DeliveryDirector
Summarized projects’
budget and cash flow
results
Calculated CSAT
Calculated ESAT
General view of project
activities
Information from reports
built by project team
CTO
Budget flow for vertical
Tendencies of overall CSAT
Tendencies of overall ESAT
Information from reports
built by DDs
Information transfer
8. Chief Technical officer
Delivery director
PM PM …
Delivery Director
PM PM …
...
PM ...
1000+ employees
Simplified structure of delivery organization
11. • Declarative representation of our understanding of the world
• Identifies the variables and their interaction with each other
• Sources:
• experts’ knowledge
• historical data
• Representation, inference, and learning
• Handles uncertainty
Probabilistic graphical models
12. «As far as the laws of mathematics refer to reality, they are not certain,
as far as they are certain, they do not refer to reality»
Albert Einstein, 1921
Uncertainty
21. Bayesian Model with pgmpy
res = BeliefPropagation(pr_model).query(
variables=["Project health"],
evidence={'Customer budget':1, 'Customer maturity':0})
print res["Project health"]
+------------------+-----------------------+
| Project health | phi(Project health) |
|------------------+-----------------------|
| success | 0.7382 |
| fail | 0.2618 |
+------------------+-----------------------+
22. Observations:
Customer maturity: mature
Customer budget: large
Resource availability: available
Project complexity: small
Project importance: very important
Project health:
success = 85.73% (↑11.91%)
23. Observations:
Customer maturity: mature
Customer budget: large
Resource availability: available
Project complexity: complex
Project importance: very important
Project health:
success = 79.45% (↓ 6.28%)
27. Cooperation model: fixed bid
Project stage: stable
Cycle time: decreasing
Number of the opened and reopened
bugs: decreasing
Finance: fit the company’s KPI values
Environment: tools are set and in-use
Soft- and hardware: no blocking requests
Human resources: no open vacancies
Customer satisfaction index: high
Number of tasks in the in-progress state:
increasing
Predicted release date: out of the schedule
Hypothetic Project observations
29. Latest date:
2017-02-14
Project health:
success = 53.18%
Major degradation happened on:
2017-02-10
Reason:
Outage in Quality
Reason:
Descending of Project Effectiveness
Reason:
Cumulating of tasks in Progress
Real project
30. • New communication tool for all levels of the company which is
already in use on every day basis
• For C-level provides understanding:
• Of the department
• Of each project in particular
• For Project Manager it is an instrument:
• For project organization and control
• For the Client:
• Fair presentation of the project work
Benefits
31. • Improve metrics for measuring project performance
• Build even more intuitive dashboard
• Introduce the model to our customers
• Improve the reflection of the domain by fitting the model to data
Future work