數(shù)據(jù)分布策略

Many data science projects do not go into production, why is that? There is no doubt in my mind that data science is an efficient tool with impressive performances. However, a successful data project is also about effectiveness: doing the right things as Russell Ackoff would write in “A systemic view of transformational leadership”.

許多數(shù)據(jù)科學(xué)項目沒有投入生產(chǎn) ，為什么呢？ 毫無疑問，數(shù)據(jù)科學(xué)是一種具有出色性能的有效工具。 但是，一個成功的數(shù)據(jù)項目也與有效性有關(guān)：如羅素·阿科夫(Russell Ackoff)在“ 變革型領(lǐng)導(dǎo)的系統(tǒng)觀點 ”中所寫， 做正確的事 。

Successful problem solving requires finding the right solution to the right problem. We fail more often because we solve the wrong problem than because we get the wrong solution to the right problem — Russell L. Ackoff (1974)

成功的問題解決需要找到正確問題的正確解決方案。 我們失敗的原因更多是因為我們解決了錯誤的問題，而不是因為我們沒有解決正確的問題— Russell L. Ackoff(1974)

How do you focus on your projects and make sure they will bring value to the company? Are you strategically thinking about how to bring your project to fruition?

您如何專注于您的項目并確保它們將為公司帶來價值？ 您是否在戰(zhàn)略上考慮如何使您的項目實現(xiàn)？

NB: I will use golf — a strategic sport — as an illustrative analogy here.

注意：在這里，我將使用高爾夫這一具有戰(zhàn)略意義的運動作為比喻。

OKR：設(shè)定您致力于實現(xiàn)的目標 (OKR: Setting objectives that you commit to achieve)

Objectives and Key Results (OKR) have been adopted in successful organisations to drive tremendous growth (Intel, Google, …). They were initially introduced by John Doerr to increase focus that produces value.

目標和關(guān)鍵結(jié)果(OKR)已通過 成功的組織來推動巨大的增長(英特爾，谷歌等)。 它們最初是由約翰·多爾(John Doerr)引入的，以增加對創(chuàng)造價值的關(guān)注。

The general idea is to set Objectives that motivate you. Imagine you are passionate about golf and next Friday there is a big competition. In the last few years, nobody won it performing well on more than 15 holes out of the 18 on the course. Setting yourself to win it is a good objective — it is specific, ambitious, and happens at a given time. You then set Key Results that can measure how you are doing on this objective. In this golf example, they could be:

一般的想法是設(shè)定激勵您的目標 。 想象一下您對高爾夫充滿熱情，下周五將進行一場激烈的比賽。 在過去的幾年中，沒有人贏得過比賽中18個洞中超過15個洞的出色表現(xiàn)。 讓自己贏得勝利是一個很好的目標-這是特定的，雄心勃勃的，并且在特定時間發(fā)生 。 然后，您可以設(shè)置關(guān)鍵結(jié)果 ，以衡量您在此目標上的表現(xiàn)。 在這個高爾夫示例中，它們可能是：

• Hit a par (ideal number of shots to get into a hole) on at least 16 out of the 18 holes.
  在18個洞中的至少16個洞上擊出標準桿(理想的開槍次數(shù))。
• Avoid dropping the ball in a sand trap more than three times — because you know that you are bad at getting out of them.
  避免將球掉入沙坑中超過3次-因為您知道自己很難擺脫掉它們。
• Go for a 20 min practice session before the competition — as you usually make a few bad shots with cold muscles.
  比賽前進行20分鐘的練習(xí)-因為您通常會因肌肉冰冷而做出一些不好的動作。

Checking all the key results are then a good indicator that you could win.

檢查所有關(guān)鍵結(jié)果便是您可能會獲勝的良好指示。

In another scenario, working for a large bank, picture you are tasked to build a loan risk model with 80% accuracy. Here are some possible key results:

在另一種情況下(為一家大型銀行工作)，您需要負責建立準確性為80％的貸款風(fēng)險模型。 以下是一些可能的關(guān)鍵結(jié)果：

• Get 80% client repayment behaviour data by XX/YY/ZZZZ.
  通過XX / YY / ZZZZ獲取80％的客戶還款行為數(shù)據(jù)。
• Test three explainable model types by AA/BB/CCCC.
  通過AA / BB / CCCC測試三種可解釋的模型類型。
• Define and track four metrics to follow the model’s performances and understand where the model is wrong.
  定義并跟蹤四個指標，以跟蹤模型的性能并了解該模型在哪里出錯。

OKRs can be used to drive focus on anything. I find them useful to define my goals on a project: building a model or an application, when will it be good enough? Aiming for the key results brings clarity. Failing becomes a learning experience that stimulates better OKRs definitions and work. On the other hand, success is then crystal clear, and you should enjoy it.

OKR可用于推動對任何事物的關(guān)注。 我發(fā)現(xiàn)它們對于定義項目目標很有用：建立模型或應(yīng)用程序，什么時候足夠好？ 瞄準關(guān)鍵結(jié)果會帶來清晰度。 失敗成為一種學(xué)習(xí)經(jīng)驗，可以激發(fā)更好的OKR定義和工作。 另一方面，成功是顯而易見的，您應(yīng)該享受成功。

Must read on the topic: Measure what matters by John Doerr.

必須閱讀以下主題： 衡量 約翰·杜爾的重要性。

傳動系統(tǒng)方法 (The Drivetrain Approach)

A drivetrain approach is a comprehensive strategy to data products definition. The following diagram shows its essential steps:

傳動系統(tǒng)方法是數(shù)據(jù)產(chǎn)品定義的綜合策略。 下圖顯示了其基本步驟：

iStock under license to M. Koutero.iStock的元素已獲M. Koutero許可。

In a new project we might ask ourselves:

在一個新項目中，我們可能會問自己：

• Objectives
  目標

Setting objectives includes answering questions such as: Does it add value to the business? Is it aligned with the current roadmap? When should it be done? Is it opening new perspectives?

設(shè)定目標包括回答以下問題：是否能為企業(yè)增加價值？ 它與當前路線圖一致嗎？ 什么時候應(yīng)該做？ 它開辟了新的視角嗎？

• Levers
  杠桿

What elements in the final product are under my control? Can I change the price of the product? The ranking on the recommendation page? …

我可以控制最終產(chǎn)品中的哪些元素？ 我可以更改產(chǎn)品的價格嗎？ 推薦頁面上的排名？ …

• Data
  數(shù)據(jù)

Given objectives and levers, what kind of data could I use? What are the compliance issues?

給定目標和杠桿，我可以使用哪種數(shù)據(jù)？ 有哪些合規(guī)性問題？

• Model / Simulation
  模型/模擬

Simulations should indicate if there is enough information in your data combined with your levers to get to your objectives. Could you drive more sales with fewer risks in the loan model example?

模擬應(yīng)表明您的數(shù)據(jù)中是否有足夠的信息與您的杠桿相結(jié)合以實現(xiàn)目標。 在貸款模型示例中，您能否以更少的風(fēng)險推動更多的銷售？

Every step is also an exit point. If you can’t find a solution alone or collectively, it might be an indication that it is not worth your time and should focus on something else.

每一步也是一個出口點。 如果您不能單獨或集體找到解決方案，則可能表明它不值得您花時間，而應(yīng)專注于其他方面。

Must read on the topic: Designing great data products by Jeremy Howard, Margit Zwemer and Mike Loukides.

必須閱讀的題目是： 設(shè)計大數(shù)據(jù)產(chǎn)品 由 杰里米·霍華德 ， 瑪吉特池維謀 和 麥克Loukides 。

決策智能 (Decision intelligence)

Decision intelligence is a more general discipline that tackles how to build a strategy given objectives in complex situations. The general process integrates notions such as external causes, multiple causal links, and feedback loops. Teams creating causal diagrams can then rationally decide upon a strategy with a clear perception of the problem at hand. One might understand decision intelligence as an extended merger between OKRs and the drive train approach.

決策智能是一門比較通用的學(xué)科，致力于解決復(fù)雜情況下給定目標的戰(zhàn)略制定方法。 常規(guī)過程集成了諸如外部原因，多個因果鏈接和反饋循環(huán)之類的概念。 然后，創(chuàng)建因果圖的團隊可以合理地決定策略，并清楚地了解當前的問題。 人們可能將決策智能理解為OKR與動力傳動系統(tǒng)方法之間的擴展合并。

iStock under license to M. Koutero.iStock的元素，并已獲得M. Koutero的許可。

In the small example above, once you select a club, whether the ball will fly high (and hopefully far) or stay rolling on the ground means the wind is more or less likely to affect. Staying on the ground might be safer, but making only small shots, you will need more of them. Having a good strategy means you will find a reasonable equilibrium to achieve your objectives and goals.

在上面的小示例中，一旦選擇了一個球桿，球會飛高(并希望遠飛)還是保持在地面上滾動，這意味著或多或少會影響風(fēng)。 留在地面上可能會更安全，但只拍攝一點，您將需要更多。 擁有良好的策略意味著您將找到一個合理的平衡點來實現(xiàn)自己的目標。

In the OKR example about the risk loan model, we would make here deeper inquiries. Would having a loan model that makes mistakes on certain types of customers be a hazard on equity? Is it possible that employees in charge of validating loans would only rely on the model, become less critical thinkers and be less likely to adjust their behaviour when delicate cases occur? Causal diagrams enable you to understand the indirect consequences of your decisions. If you consider that getting the right clean data and building a model ready for production is often a task that takes months, is it not worth spending some time on the reasons you are doing it?

在有關(guān)風(fēng)險貸款模型的OKR示例中，我們將在這里進行更深入的查詢。 具有在某些類型的客戶上犯錯誤的貸款模型會危害股本嗎？ 負責發(fā)生貸款問題的員工是否可能僅依靠模型，變得不那么批判性的思想家并且在發(fā)生細微情況時不太可能調(diào)整其行為？ 因果圖使您能夠理解決策的間接后果。 如果您認為獲取正確的干凈數(shù)據(jù)并為生產(chǎn)做好準備的模型構(gòu)建通常需要花費數(shù)月的時間，難道不應(yīng)該花一些時間在做這些事情的原因上嗎？

For engineers and scientists, it is not extremely different from specifying a classical digital product with its constraints and target performances but broadening the perspective. What is interesting to me, is the focus on the decision making (“should I build this product and how?”) putting both business and technical people together to make sure that at the scale of a whole ecosystem, the next move is the right one.

對于工程師和科學(xué)家而言，它與指定具有約束條件和目標性能的經(jīng)典數(shù)字產(chǎn)品并沒有什么不同，但是拓寬了視野。 對我而言，有趣的是將重點放在決策上(“我應(yīng)該制造這種產(chǎn)品以及如何制造嗎？”)，將業(yè)務(wù)人員和技術(shù)人員放在一起，以確保在整個生態(tài)系統(tǒng)的規(guī)模上，下一步行動是正確的之一。

Must read on the topic: Link by Lorien Pratt

必須閱讀以下主題： Lorien Pratt的 鏈接

Strategy is not limited to a top/down practice falling under the umbrella of leaders, managers, product managers, etc. I think it is part of any job to meet halfway and have some strategical thinking under the hood. Maybe these frameworks are sometimes too elaborated, but at its core, they start with a simple question that we can ask ourselves: why should I do this project?

戰(zhàn)略不僅限于領(lǐng)導(dǎo)，經(jīng)理，產(chǎn)品經(jīng)理等領(lǐng)導(dǎo)下的自上而下的實踐。我認為，中途開會并有一些戰(zhàn)略思想是任何工作的一部分。 也許這些框架有時過于復(fù)雜，但從根本上講，它們以一個簡單的問題開始，我們可以問自己：我為什么要進行這個項目？

As a field, growing beyond the AI hype, we cannot stay in an isolated system, extending our level of specialisation without clearly showing its value. Missing middle professionals are likely to be of importance in this task (Paul R. Daugherty — CTIO at Accenture and Lorien Pratt). Whether they will be decision intelligence specialists, data strategists or data product managers will be a matter of semantics and establishing new practices in the field.

作為一個超越AI炒作的領(lǐng)域，我們不能停留在孤立的系統(tǒng)中，無法在沒有清楚顯示其價值的情況下擴展專業(yè)化水平。 缺少中層專業(yè)人員可能對這項任務(wù)很重要( Paul R. Daugherty-埃森哲公司的CTIO和Lorien Pratt)。 他們將是決策情報專家 ， 數(shù)據(jù)戰(zhàn)略家還是數(shù)據(jù)產(chǎn)品經(jīng)理，將取決于語義并確定該領(lǐng)域的新實踐。

___________

___________

References :

參考文獻：

• Ackoff, R, L: 1998, A Systemic View of Transformational Leadership (Systemic Practice and Action Research).

  Ackoff，R，L：1998年， 《變革型領(lǐng)導(dǎo)的系統(tǒng)觀點》 (系統(tǒng)實踐與行動研究)。

• Ackoff, R. L.: 1974, Redesigning the Future: A Systems Approach to Societal Problems (John Wiley & Sons).

  RL，阿科夫(Ackoff)，1974年，《 重新設(shè)計未來：社會問題的系統(tǒng)方法》 (約翰·威利父子)。

• Doerr, J: 2018, Measure what matters: How Google, Bono, and the Gates Foundation rock the world with OKRs (Portfolio Penguin).

  Doerr，J：2018， 衡量重要的事情：Google，Bono和蓋茨基金會如何利用OKR (Portfolio Penguin) 震撼整個世界 。

• Pratt, L: 2019, Link (Emerald Publishing Limited).

  普拉特，L：2019， Link (Emerald Publishing Limited)。

翻譯自: https://towardsdatascience.com/three-strategies-towards-effective-data-projects-eed29ad05ded

數(shù)據(jù)分布策略

總結(jié)

以上是生活随笔為你收集整理的数据分布策略_有效数据项目的三种策略的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。