A Fine Slice Of SQL Server
Tori Tompkins has a category for us:
Model drift refers to the decline of model performance due to changes in data and relationships. Most drift is caused by things entirely out of our control so while we can’t stop it from happening, we can identify and mitigate it.
Click through for four common explanations for model drift and what you can do about them.
Comments closedJust like the random forests, isolation forests are built using decision trees. They are implemented in an unsupervised fashion as there are no pre-defined labels. Isolation forests were designed with the idea that anomalies are “few and distinct” data points in a dataset.
Recall that decision trees are built using information criteria such as Gini index or entropy. The obviously different groups are separated at the root of the tree and deeper into the branches, the subtler distinctions are identified. Based on randomly picked characteristics, an isolation forest processes the randomly subsampled data in a tree structure. Samples that reach further into the tree and require more cuts to separate them have a very little probability that they are anomalies. Likewise, samples that are found on the shorter branches of the tree are more likely to be anomalies, since the tree found it simpler to distinguish them from the other data.
Click through for descriptions and the code.
Comments closedAntoine Soetewey has a handy chart for us:
Being a teaching assistant in statistics for students with diverse backgrounds, I have the chance to see what is globally not well understood by students.
I have realized that it is usually not a problem for students to do a specific statistical test when they are told which one to use (as long as they have good resources and they have been attentive during classes, of course). However, it appears that the task is much more difficult for them when they need to choose what test to do.
Click through for the chart, as well as a PDF version. H/T R-Bloggers.
Comments closedSebastian Sauer convinces me that many countries suffer a severe chocolate consumption shortage:
Following a study that reported a strong correlation between chocolate consumption and Nobel prices.
Severe.
Comments closedLuke Menzies and Gavita Regunath create a schedule:
Linear optimisation (often referred to as linear programming) is not cutting edge or new. It has been around for a very long time. It was first introduced within the field of operational research during World War II, where it was used to help minimise costings. The method proposed for solving these problems is known as the simplex method, and it hasn’t changed much today. Although this method hasn’t changed significantly, what has changed significantly is the computing power and accessibility of this technique, allowing these methods to be used on very complex scenarios with almost a click of a button. Convenient libraries have allowed the intricate complexities of setting these problems up on a computer to be simplified.
Read on for an example of linear programming. This is something I’ve always enjoyed, but haven’t had many places to use this technique in my professional career. That said, shout out to everyone who’s ever used LINGO.
Comments closedMichael Mayer has some interesting R and Python code for us:
On ML competition platforms like Kaggle, complex and unintuitively behaving models dominate. In this respect, reality is completely different. There, the majority of models do not serve as pure prediction machines but rather as fruitful source of information. Furthermore, even if used as prediction machine, the users of the models might expect a certain degree of consistency when “playing” with input values.
A classic example are statistical house appraisal models. An additional bathroom or an additional square foot of ground area is expected to raise the appraisal, everything else being fixed (ceteris paribus). The user might lose trust in the model if the opposite happens.
One way to enforce such consistency is to monitor the signs of coefficients of a linear regression model. Another useful strategy is to impose monotonicity constraints on selected model effects.
Certain types of regression algorithm make this easy, but random forest? Not so much. That’s where Michael steps in.
Comments closedFlorent Buisson has an interesting post on avoiding p-value calculations:
And indeed, I worked with highly-skilled data scientists who had a very sharp understanding of statistics. But after years of designing and analyzing experiments, I grew dissatisfied with the way we communicated results to decision-makers. I felt that the over-reliance on p-values led to sub-optimal decisions. After talking to colleagues in other companies, I realized that this was a broader problem, and I set up to write a guide to better data analysis. In this article, I’ll present one of the biggest recommendations of the book, which is to ditch p-values and use Bootstrap confidence intervals instead.
I’m a committed Bayesian (or at least a Bayesian who should be committed—depends on who you ask), so I’d consider this a big step forward.
Comments closedRoel Hogervorst thinks about data sizes in R and Python:
Your dataset becomes so big and unwieldy that operations take a long time. How long is too long? That depends on you, I get annoyed if I don’ t get feedback within 20 seconds (and I love it when a program shows me a progress bar at that point, at least I know how long it will take!), your boundary may lay at some other point. When you reach that point of annoyance or point of no longer being able to do your work. You should improve your workflow.
I will show you how to do some speedups by using other R packages, in python moving from pandas to polars, or leveraging databases. I see some hesitancy about moving to a database for analytical work, and that is too bad. Bad for two reasons, one: it is super simple, two it will save you a lot of time.
I definitely agree with Roel’s bottom line here. Granted, part of that is domain knowledge, but databases are extremely good at handling data and both languages have plenty of database accessibility.
One last tip, though: if you’re on the data science or data analytics track, learn SQL. Yes, libraries like dbplyr in R or ORMs in Python can cover up a lot, but that comes at a cost, typically in terms of performance. Building these skills will make your life considerably easier.
Eric Colson, et al, have a humbling thought for us:
The introduction of data science into the business world has contributed far more than recommendation algorithms; it has also taught us a lot about the efficacy with which we manage our businesses. Specifically, data science has introduced rigorous methods for measuring the outcomes of business ideas. These are the strategic ideas that we implement in order to achieve our business goals. For example, “We’ll lower prices to increase demand by 10%” and “we’ll implement a loyalty program to improve retention by 5%.” Many companies simply execute on their business ideas without measuring if they delivered the impact that was expected. But, science-based organizations are rigorously quantifying this impact and have learned some sobering lessons:
1. The vast majority of business ideas fail to generate a positive impact.
2. Most companies are unaware of this.
3. It is unlikely that companies will increase the success rate for their business ideas.
Read the whole thing. It gives a lot of perspective to a difficult problem: there aren’t as many “free wins” in a business as you might expect. To paraphrase Adam Smith, there is a lot of ruin in a company…but that doesn’t mean you know what exactly it is or how exactly to fix it. Coming in with appropriate humility and a flexible mind (by which I mean a willingness to see reality even when it doesn’t comport to the mental model you’ve built over time) can help improve those odds.
Comments closedBrendan Tierney performs a bit of document classification with scikit-learn and nltk:
Text mining is a popular topic for exploring what text you have in documents etc. Text mining and NLP can help you discover different patterns in the text like uncovering certain words or phases which are commonly used, to identifying certain patterns and linkages between different texts/documents. Combining this work on Text mining you can use Word Clouds, time-series analysis, etc to discover other aspects and patterns in the text. Check out my previous blog posts (post 1, post 2) on performing Text Mining on documents (manifestos from some of the political parties from the last two national government elections in Ireland). These two posts gives you a simple indication of what is possible.
We can build upon these Text Mining examples to include other machine learning algorithms like those for Classification. With Classification we want to predict or label a record or document to have a particular value. With Classification this could involve labeling a document as being positive or negative (movie or book reviews), or determining if a document is for a particular domain such as Technology, Sports, Entertainment, etc
Click through for a walkthrough of this process.
Comments closed