Press "Enter" to skip to content

Category: Data Science

Analyzing Federal Reserve Data With Ordinary Least Squares

Published 2018-06-27 by Kevin Feasel

Sam Shum has a tutorial walking us through extracting and analyzing data from the St. Louis Federal Reserve’s FRED economic database:

Download specific macroeconomic data from FRED St. Louis economic databases and ETL the data. Many other data series can be found at the FRED’s website.

# get unemployment data time series from FRED St. Louis
dfunrate <- get_fred_series("UNRATE", "unrate", observation_start = startdate, observation_end = enddate)

# get University of Michigan consumer sentiment index data time series from FRED St. Louis
dfumcsent <- get_fred_series("UMCSENT", "umcsent", observation_start = startdate, observation_end = enddate)

# combine the two time series data into one data frame
dfall <- cbind(dfunrate,dfumcsent)

# strip or remove redundant month field from data downloaded from FRED St. Louis
dfall <- dfall[,c(1,2,4)]

# obtain the number of data points in the dataframe
mdx <- (1:nrow(dfall))  

# convert FRED date field from string to R's date type
dfall$date <- as.Date(dfall$date)

There’s a nice chart builder on the FRED website too, but it’s good to be able to grab the data on your own.

Comments closed

Neural Topic Models On Amazon SageMaker

Published 2018-06-26 by Kevin Feasel

David Ping, et al, show off topic modeling on Amazon SageMaker:

Topic Modeling is used to organize a corpus of documents into “topics” which is a grouping based on a statistical distribution of words within the documents themselves. Amazon Comprehend, our fully managed text analytics service, provides a pre-configured topic modeling API that is best suited for the most popular use cases like organizing customer feedback, support incidents or workgroup documents. Amazon Comprehend is the suggested topic modeling choice for customers as it removes a lot of the most routine steps associated with topic modeling like tokenization, training a model and adjusting parameters. Amazon SageMaker’s Neural Topic Model (NTM) caters to the use cases where a finer control of the training, optimization, and/or hosting of a topic model is required, such as training models on text corpus of particular writing style or domain, or hosting topic models as part of a web application. While Amazon SageMaker NTM provides a starting point of state-of-the-art topic modeling, customers have the flexibility to modify the network architecture as well as hyperparameters to accommodate the idiosyncrasies of their data sets as well as to tune the trade-off between a multitude of metrics such as document modeling accuracy, human interpretability and granularity of the learned topics, based on their applications. In addition, Amazon SageMaker NTM leverages the full power of the Amazon SageMaker platform: easily configurable training and hosting infrastructure, automatic hyperparameter optimization, and fully-managed hosting with auto-scaling.

They walk through the entire topic modeling process, so check it out.

Comments closed

Comparing Keras In Python Versus R

Published 2018-06-20 by Kevin Feasel

Dmitry Kisler performs image classification using Keras in both Python and R:

From the plots above, one can see that:

  • the accuracy of your model doesn’t depend on the language you use to build and train it (the plot shows only train accuracy, but the model doesn’t have high variance and the bias accuracy is around 99% as well).

  • even though 10 measurements may be not convincing, but Python would reduce (by up to 15%) the time required to train your CNN model. This is somewhat expected because R uses Python under the hood when executes Keras functions.

This is just one example, but the results are about what I’d expect.

Comments closed

Auto-Encoders And KernelML

Published 2018-06-20 by Kevin Feasel

Rohan Kotwani gives us an example where KernelML might be better than TensorFlow or PyTorch:

So what’s the point of using KernelML?

1. The parameters in each layer can be non-linear
2. Each parameter can be sampled from a different random distribution
3. The parameters can be transformed to meet certain constraints
4. Network combinations are defined in terms of numpy operations
5. Parameters are probabilistically updated
6. Each parameter update samples the loss function around a local or global minima

KerneML Specs

KernelMLis brute force optimizer that can be used to train machine learning algorithms. The package uses a combination of a machine learning and monte carlo simulations to optimize a parameter vector with a user defined loss function. Using kernelml creates a high computational cost for large complex networks because it samples the loss function using a subspace for each parameter in the parameter vector which requires many random simulations. The computational cost was reduced by enabling parallel computations with the ipyparallel. The decision to use this package was made because it effectively utilizes the cores on a machine.

It’s an interesting use case, though I would have liked to have seen a direct comparison to other frameworks.

Comments closed

Calculating TF-IDF Using Apache Spark

Published 2018-06-18 by Kevin Feasel

Arseniy Tashoyan shows us how to calculate Term Frequency-Inverse Document Frequency using Apache Spark:

TF-IDF is used in a large variety of applications. Typical use cases include:

  • Document search.
  • Document tagging.
  • Text preprocessing and feature vector engineering for Machine Learning algorithms.

There is a vast number of resources on the web explaining the concept itself and the calculation algorithm. This article does not repeat the information in these other Internet resources, it just illustrates TF-IDF calculation with help of Apache Spark. Emml Asimadi, in his excellent article Understanding TF-IDF, shares an approach based on the old Spark RDD and the Python language. This article, on the other hand, uses the modern Spark SQL API and Scala language.

Although Spark MLlib has an API to calculate TF-IDF, this API is not convenient to learn the concept. MLlib tools are intended to generate feature vectors for ML algorithms. There is no way to figure out the weight for a particular term in a particular document. Well, let’s make it from scratch, this will sharpen our skills.

Read on for the solution.  It seems that there tend to be better options today than TF-IDF for natural language problems, but it’s an easy algorithm to understand, so it’s useful as a first go.

Comments closed

Using The Azure Data Science VM With GPUs

Published 2018-06-15 by Kevin Feasel

Jennifer Marsman has some tips and tricks around using the Azure Data Science Virtual Machine on an instance running with GPU support:

To get GPU support, you need both hardware with GPUs in a datacenter, as well as the right software – namely, a virtual machine image that includes GPU drivers so you can use the GPU.

The biggest tip is to use the Deep Learning Virtual Machine!  The provisioning experience has been optimized to filter to the options that support GPU (the NC series – see below), which make it easier to set it up correctly.

Read on for the rest of the advice.

Comments closed

Visualizing Model Input Effects

Published 2018-06-14 by Kevin Feasel

Ilknur Kaynar Kabul shows us how to use partial dependence plots and individual conditional expectation plots to view the specific effect of an input variable on a model:

A partial dependence (PD) plot depicts the functional relationship between a small number of input variables and predictions. They show how the predictions partially depend on values of the input variables of interest.  For example, a PD plot can show whether the probability of flu increases linearly with fever. It can show whether high energy level will decrease the probability of having flu. PD can also show the type of relationship, such as a step function, curvilinear, linear and so on.

The simplest PD plots are 1-way plots, which show how a model’s predictions depend on a single input. The plot below shows the relationship (according the model that we trained) between price (target) and number of bathrooms. Here, we see that house prices increase as we increase the number of bathroom up to 4. After that it does not change the house price.

These types of plots are helpful for understanding the mechanics behind a model.

Comments closed

Pareto Efficiency And Mario Kart

Published 2018-06-11 by Kevin Feasel

The folks at Civis Analytics answer one of the more important questions in life:

Mario Kart was a staple of my childhood — my friends and I would spend hours after school as Mario, Luigi, and other characters from the Nintendo universe racing around cartoonish tracks and lobbing pixelated bananas at each other. One thing that always vexed our little group of would-be speedsters was the question of which character was best. Some people swore by zippy Yoshi, others argued that big, heavy Bowser was the best option. Back then there were only eight options to choose from; fast forward to the current iteration of the Mario Kart franchise and the question is even more complicated because you can select different karts and tires to go with your character. My Mario Kart reflexes aren’t what they used to be, but I am better at data science than I was as a fourth grader, so in this post I’ll use data to finally answer the question “Who is the best character in Mario Kart?”

This post also acts as a primer on Pareto Efficiency, an important concept in economics.

Comments closed

Stack Overflow Developer Survey Data Available

Published 2018-06-01 by Kevin Feasel

Julia Silge has a post looking at the Stack Overflow 2018 developer survey:

Starting today, you can access the public data release for Stack Overflow’s 2018 Developer Survey. Over 100,000 developers from around the world shared their opinions about everything from their favorite technologies to job preferences, and this data is now available for you to analyze yourself. This year, we are partnering with Kaggle to publish and highlight this dataset. This means you can access the data both here on our site and on Kaggle Datasets, and that on Kaggle, you can explore the dataset using Kernels. Kaggle is awarding two $1,000 awards over the next two weeks to authors of top Kernels on the Stack Overflow dataset.

Looks like an interesting data set.

Comments closed

Probabilities And Poker

Published 2018-05-25 by Kevin Feasel

Steve Miller has a notebook on 5-card draw probabilities:

The population of 5 card draw hands, consisting of 52 choose 5 or 2598960 elements, is pretty straightforward both mathematically and statistically.

So of course ever the geek, I just had to attempt to show her how probability and statistics converge. In addition to explaining the “combinatorics” of the counts and probabilities, I undertook two computational exercises. The first was to delineate all possible combinations of 5 card draws from a 52 card deck, counting occurrences of relevant combinations such as 2 pair, a straight, or nothing in a cell loop.

Steve has made his notebook available for us.

Comments closed