# Useful properties of ROC curves, AUC scoring, and Gini Coefficients

Receiver Operating Characteristic (ROC) curves and AUC values are often used to score binary classification models in Kaggle and in papers. However, for a long time I found them fairly unintuitive and confusing. In this blog post, I will explain some basic properties of ROC curves that are useful to know for Kaggle competitions, and how you should interpret them.

Above: Example of a ROC curve

First, the definitions. A ROC curve plots the performance of a binary classifier under various threshold settings; this is measured by true positive rate and false positive rate. If your classifier predicts “true” more often, it will have more true positives (good) but also more false positives (bad). If your classifier is more conservative, predicting “true” less often, it will have fewer false positives but fewer true positives as well. The ROC curve is a graphical representation of this tradeoff.

A perfect classifier has a 100% true positive rate and 0% false positive rate, so its ROC curve passes through the upper left corner of the square. A completely random classifier (ie: predicting “true” with probability p and “false” with probability 1-p for all inputs) will by random chance correctly classify proportion p of the actual true values and incorrectly classify proportion p of the false values, so its true and false positive rates are both p. Therefore, a completely random classifier’s ROC curve is a straight line through the diagonal of the plot.

The AUC (Area Under Curve) is the area enclosed by the ROC curve. A perfect classifier has AUC = 1 and a completely random classifier has AUC = 0.5. Usually, your model will score somewhere in between. The range of possible AUC values is [0, 1]. However, if your AUC is below 0.5, that means you can invert all the outputs of your classifier and get a better score, so you did something wrong.

The Gini Coefficient is 2*AUC – 1, and its purpose is to normalize the AUC so that a random classifier scores 0, and a perfect classifier scores 1. The range of possible Gini coefficient scores is [-1, 1]. If you search for “Gini Coefficient” on Google, you will find a closely related concept from economics that measures wealth inequality within a country.

Why do we care about AUC, why not just score by percentage accuracy?

AUC is good for classification problems with a class imbalance. Suppose the task is to detect dementia from speech, and 99% of people don’t have dementia and only 1% do. Then you can submit a classifier that always outputs “no dementia”, and that would achieve 99% accuracy. It would seem like your 99% accurate classifier is pretty good, when in fact it is completely useless. Using AUC scoring, your classifier would score 0.5.

In many classification problems, the cost of a false positive is different from the cost of a false negative. For example, it is worse to falsely imprison an innocent person than to let a guilty criminal get away, which is why our justice system assumes you’re innocent until proven guilty, and not the other way around. In a classification system, we would use a threshold rule, where everything above a certain probability is treated as 1, and everything below is treated as 0. However, deciding on where to draw the line requires weighing the cost of a false positive versus a false negative — this depends on external factors and has nothing to do with the classification problem.

AUC scoring lets us evaluate models independently of the threshold. This is why AUC is so popular in Kaggle: it enables competitors to focus on developing a good classifier without worrying about choosing the threshold, and let the organizers choose the threshold later.

(Note: This isn’t quite true — a classifier can sometimes be better at certain thresholds and worse at other thresholds. Sometimes it’s necessary to combine classifiers to get the best one for a particular threshold. Details in the paper linked at the end of this post.)

Next, here’s a mix of useful properties to know when working with ROC curves and AUC scoring.

AUC is not directly comparable to accuracy, precision, recall, or F1-score. If your model is achieving 0.65 AUC, it’s incorrect to interpret that as “65% accurate”. The reason is that AUC exists independently of a threshold and is immune to class imbalance, whereas accuracy / precision / recall / F1-score do require you picking a threshold, so you’re measuring two different things.

Only relative order matters for AUC score. When computing ROC AUC, we predict a probability for each data point, sort the points by predicted probability, and evaluate how close is it from a perfect ordering of the points. Therefore, AUC is invariant under scaling, or any transformation that preserves relative order. For example, predicting [0.03, 0.99, 0.05, 0.06] is the same as predicting [0.15, 0.92, 0.89, 0.91] because the relative ordering for the 4 items is the same in both cases.

A corollary of this is we can’t treat outputs of an AUC-optimized model as the likelihood that it’s true. Some models may be poorly calibrated (eg: its output is always between 0.3 and 0.32) but still achieve a good AUC score because its relative ordering is correct. This is something to look out for when blending together predictions of different models.

That’s my summary of the most important properties to know about ROC curves. There’s more that I haven’t talked about, like how to compute AUC score. If you’d like to learn more, I’d recommend reading “An introduction to ROC analysis” by Tom Fawcett.

# I trained a neural network to describe pictures and it’s hilariously bad

This month, I’ve been working on a neural network to describe in a sentence what’s happening in a picture, otherwise known as image captioning. My model roughly follows the architecture outlined in the paper “Show and Tell: A Neural Image Caption Generator” by Vinyals et al., 2014.

A high level overview: the neural network first uses a convolutional neural network to turn the picture into an abstract representation. Then, it uses this representation as the initial hidden state of a recurrent neural network or LSTM, which generates a natural language sentence. This type of neural network is called an encoder-decoder network and is commonly used for a lot of NLP tasks like machine translation.

Above: Encoder-decoder image captioning neural network (Figure 1 of paper)

When I first encountered LSTMs, I was really confused about how they worked, and how to train them. If your output is a sequence of words, what is your loss function and how do you backpropagate it? In fact, the training and inference passes of an LSTM are quite different. In this blog post, I’ll try to explain this difference.

Above: Training procedure for caption LSTM, given known image and caption

During training mode, we train the neural network to minimize perplexity of the image-caption pair. Perplexity measures how the likelihood that the neural network would generate the given caption when it sees the given image. If we’re training it to output the caption “a cute cat”, the perplexity is:

P(“a” | image) *

P(“cute” | image, “a”) *

P(“cat” | image, “a”, “cute”) *

(Note: for numerical stability reasons, we typically work with sums of negative log likelihoods rather than products of likelihood probabilities, so perplexity is actually the negative log of that whole thing)

After passing the whole sequence through the LSTM one word at a time, we get a single value, the perplexity, which we can minimize using backpropagation and gradient descent. As perplexity gets lower and lower, the LSTM is more likely to produce similar captions to the ground truth when it sees a similar image. This is how the network learns to caption images.

Above: Inference procedure for caption LSTM, given only the image but no caption

During inference mode, we repeatedly sample the neural network, one word at a time, to produce a sentence. On each step, the LSTM outputs a probability distribution for the next word, over the entire vocabulary. We pick the highest probability word, add it to the caption, and feed it back into the LSTM. This is repeated until the LSTM generates the end marker. Hopefully, if we trained it properly, the resulting sentence will actually describe what’s happening in the picture.

This is the main idea of the paper, and I omitted a lot details. I encourage you to read the paper for the finer points.

I implemented the model using PyTorch and trained it using the MS COCO dataset, which contains about 80,000 images of common objects and situations, and each image is human annotated with 5 captions.

To speed up training, I used a pretrained VGG16 convnet, and pretrained GloVe word embeddings from SpaCy. Using lots of batching, the Adam optimizer, and a Titan X GPU, the neural network trains in about 4 hours. It’s one thing to understand how it works on paper, but watching it actually spit out captions for real images felt like magic.

Above: How I felt when I got this working

How are the results? For some of the images, the neural network does great:

“A train is on the tracks at a station”

“A woman is holding a cat in her arms”

Other times the neural network gets confused, with amusing results:

“A little girl holding a stuffed animal in her hand”

“A baby laying on a bed with a stuffed animal”

“A dog is running with a frisbee in its mouth”

I’d say we needn’t worry about the AI singularity anytime soon 🙂

The original paper has some more examples of correct and incorrect captions that might be generated. Newer models also made improvements to generate more accurate captions: for example, adding a visual attention mechanism improved the results a bit. However, the state-of-the-art models still fall short on human performance; they often make mistakes when describing pictures with objects in unusual configurations.

This is a work in progress; source code is on Github here.

# Publishing Negative Results in Machine Learning is like Proving Dragons don’t Exist

I’ve been reading a lot of machine learning papers lately, and one thing I’ve noticed is that the vast majority of papers report positive results — “we used method X on problem Y, and beat the state-of-the-art results”. Very rarely do you see a paper that reports that something doesn’t work.

The result is publication bias — if we only publish the results of experiments that succeed, even statistically significant results could be due to random chance, rather than anything actually significant happening. Many areas of science are facing a replication crisis, where published research cannot be replicated.

There is some community discussion of encouraging more negative paper submissions, but as of now, negative results are rarely publishable. If you attempt an experiment but don’t get the results you expected, your best hope is to try a bunch of variations of the experiment until you get some positive result (perhaps on a special case of the problem), after which you pretend the failed experiments never happened. With few exceptions, any positive result is better than a negative result, like “we tried method X on problem Y, and it didn’t work”.

## Why publication bias is not so bad

I just described a cynical view of academia, but actually, there’s a good reason why the community prefers positive results. Negative results are simply not very useful, and contribute very little to human knowledge.

Now why is that? When a new paper beats the state-of-the-art results on a popular benchmark, that’s definite proof that the method works. The converse is not true. If your model fails to produce good results, it could be due to a number of reasons:

• Your dataset is too small / too noisy
• You’re using the wrong batch size / activation function / regularization
• You’re using the wrong loss function / wrong optimizer
• You have a bug in your code

Above: Only when everything is correct will you get positive results; many things can cause a model to fail. (Source)

So if you try method X on problem Y and it doesn’t work, you gain very little information. In particular, you haven’t proved that method X cannot work. Sure, you found that your specific setup didn’t work, but have you tried making modification Z? Negative results in machine learning are rare because you can’t possibly anticipate all possible variations of your method and convince people that all of them won’t work.

## Searching for dragons

Suppose we’re scientists attending the International Conference of Flying Creatures (ICFC). Somebody mentioned it would be nice if we had dragons. Dragons are useful. You could do all sorts of cool stuff with a dragon, like ride it into battle.

“But wait!” you exclaim: “Dragons don’t exist!”

I glance at you questioningly: “How come? We haven’t found one yet, but we’ll probably find one soon.”

Your intuition tells you dragons shouldn’t exist, but you can’t articulate a convincing argument why. So you go home, and you and your team of grad students labor for a few years and publish a series of papers:

• “We looked for dragons in China and we didn’t find any”
• “We looked for dragons in Europe and we didn’t find any”
• “We looked for dragons in North America and we didn’t find any”

Eventually, the community is satisfied that dragons probably don’t exist, for if they did, someone would have found one by now. But a few scientists still harbor the possibility that there may be dragons lying around in a remote jungle somewhere. We just don’t know for sure.

This remains the state of things for a few years until a colleague publishes a breakthrough result:

• “Here’s a calculation that shows that any dragon with a wing span longer than 5 meters will collapse under its own weight”

You read the paper, and indeed, the logic is impeccable. This settles the matter once and for all: dragons don’t exist (or at least the large, flying sort of dragons).

## When negative results are actually publishable

The research community dislikes negative results because they don’t prove a whole lot — you can have a lot of negative results and still not be sure that the task is impossible. In order for a negative result to be valuable, it needs to present a convincing argument why the task is impossible, and not just a list of experiments that you tried that failed.

This is difficult, but it can be done. Let me give an example from computational linguistics. Recurrent neural networks (RNNs) can, in theory, compute any function defined over a sequence. In practice, however, they had difficulty remembering long-term dependencies. Attempts to train RNNs using gradient descent ran into numerical difficulties known as the vanishing / exploding gradient problem.

Then, Bengio et al. (1994) formulated a mathematical model of an RNN as an iteratively applied function. Using ideas from dynamical systems theory, they showed that as the input sequence gets longer and longer, the result is more and more sensitive to noise. The details are technical, but the gist of it is that under some reasonable assumptions, training RNNs using gradient descent is impossible. This is a rare example of a negative result in machine learning — it’s an excellent paper and I’d recommend reading it.

Above: A Long Short Term Memory (LSTM) network handles long term dependencies by adding a memory cell (Source)

Soon after the vanishing gradient problem was understood, researchers invented the LSTM (Hochreiter and Schmidhuber, 1997). Since training RNNs with gradient descent was hopeless, they added a ‘latching’ mechanism that allows state to persist through many iterations, thus avoiding the vanishing gradient problem. Unlike plain RNNs, LSTMs can handle long term dependencies and can be trained with gradient descent; they are among the most ubiquitous deep learning architectures in NLP today.

After reading the breakthrough dragon paper, you pace around your office, thinking. Large, flying dragons can’t exist after all, as they would collapse under their own weight — but what about smaller, non-flying dragons? Maybe we’ve been looking for the wrong type of dragons all along? Armed with new knowledge, you embark on a new search…

Above: Komodo Dragon, Indonesia

…and sure enough, you find one 🙂

# XGBoost learns the Canadian Flag

XGBoost is a machine learning library that’s great for classification tasks. It’s often seen in Kaggle competitions, and usually beats other classifiers like logistic regression, random forests, SVMs, and shallow neural networks. One day, I was feeling slightly patriotic, and wondered: can XGBoost learn the Canadian flag?

Above: Our home and native land

Let’s find out!

## Preparing the dataset

The task is to classify each pixel of the Canadian flag as either red or white, given limited data points. First, we read in the image with R and take the red channel:

library(png)
library(ggplot2)
library(xgboost)

red <- img[,,2]

HEIGHT <- dim(red)[1]
WIDTH <- dim(red)[2]


Next, we sample 7500 random points for training. Also, to make it more interesting, each point has a probability 0.05 of flipping to the opposite color.

ERROR_RATE <- 0.05

get_data_points <- function(N) {
x <- sample(1:WIDTH, N, replace = T)
y <- sample(1:HEIGHT, N, replace = T)
p <- red[cbind(y, x)]
p <- round(p)
flips <- sample(c(0, 1), N, replace = T,
prob = c(ERROR_RATE, 1 - ERROR_RATE))
p[flips == 1] <- 1 - p[flips == 1]
data.frame(x=as.numeric(x), y=as.numeric(y), p=p)
}

data <- get_data_points(7500)


This is what our classifier sees:

Alright, let’s start training.

## Quick introduction to XGBoost

XGBoost implements gradient boosted decision trees, which were first proposed by Friedman in 1999.

Above: XGBoost learns an ensemble of short decision trees

The output of XGBoost is an ensemble of decision trees. Each individual tree by itself is not very powerful, containing only a few branches. But through gradient boosting, each subsequent tree tries to correct for the mistakes of all the trees before it, and makes the model better. After many iterations, we get a set of decision trees; the sum of the all their outputs is our final prediction.

For more technical details of how this works, refer to this tutorial or the XGBoost paper.

## Experiments

Fitting an XGBoost model is very easy using R. For this experiment, we use decision trees of height 3, but you can play with the hyperparameters.

fit <- xgboost(data = matrix(c(data$x, data$y), ncol = 2), label = data$p, nrounds = 1, max_depth = 3)  We also need a way of visualizing the results. To do this, we run every pixel through the classifier and display the result: plot_canada <- function(dataplot) { dataplot$y <- -dataplot$y dataplot$p <- as.factor(dataplot$p) ggplot(dataplot, aes(x = x, y = y, color = p)) + geom_point(size = 1) + scale_x_continuous(limits = c(0, 240)) + scale_y_continuous(limits = c(-120, 0)) + theme_minimal() + theme(panel.background = element_rect(fill='black')) + theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank()) + scale_color_manual(values = c("white", "red")) } fullimg <- expand.grid(x = as.numeric(1:WIDTH), y = as.numeric(1:HEIGHT)) fullimg$p <- predict(fit, newdata = matrix(c(fullimg$x, fullimg$y), ncol = 2))
fullimg$p <- as.numeric(fullimg$p > 0.5)



In the first iteration, XGBoost immediately learns the two red bands at the sides:

After a few more iterations, the maple leaf starts to take form:

By iteration 60, it learns a pretty recognizable maple leaf. Note that the decision trees split on x or y coordinates, so XGBoost can’t learn diagonal decision boundaries, only approximate them with horizontal and vertical lines.

If we run it for too long, then it starts to overfit and capture the random noise in the training data. In practice, we would use cross validation to detect when this is happening. But why cross-validate when you can just eyeball it?

That was fun. If you liked this, check out this post which explores various classifiers using a flag of Australia.

The source code for this blog post is posted here. Feel free to experiment with it.

# Kaggle Speech Recognition Challenge

For the past few weeks, I’ve been working on the TensorFlow Speech Recognition Challenge on Kaggle. The task is to recognize a one-second audio clip, where the clip contains one of a small number of words, like “yes”, “no”, “stop”, “go”, “left”, and “right”.

In general, speech recognition is a difficult problem, but it’s much easier when the vocabulary is limited to a handful of words. We don’t need to use complicated language models to detect phonemes, and then string the phonemes into words, like Kaldi does for speech recognition. Instead, a convolutional neural network works quite well.

## First Steps

The dataset consists of about 64000 audio files which have already been split into training / validation / testing sets. You are then asked to make predictions on about 150000 audio files for which the labels are unknown.

Actually, this dataset had already been published in academic literature, and people published code to solve the same problem. I started with GCommandPytorch by Yossi Adi, which implements a speech recognition CNN in Pytorch.

The first step that it does is convert the audio file into a spectrogram, which is an image representation of sound. This is easily done using LibRosa.

Above: Sample spectrograms of “yes” and “no”

Now we’ve converted the problem to an image classification problem, which is well studied. To an untrained human observer, all the spectrograms may look the same, but neural networks can learn things that humans can’t. Convolutional neural networks work very well for classifying images, for example VGG16:

Above: A Convolutional Neural Network (LeNet). VGG16 is similar, but has even more layers.

## Voice Activity Detection

You might ask: if somebody already implemented this, then what’s there left to do other than run their code? Well, the test data contains “silence” samples, which contain background noise but no human speech. It also has words outside the set we care about, which we need to label as “unknown”. The Pytorch CNN produces about 95% validation accuracy by itself, but the accuracy is much lower when we add these two additional requirements.

For silence detection, I first tried the simplest thing I could think of: taking the maximum absolute value of the waveform and decide it’s “silence” if the value is below a threshold. When combined with VGG16, this gets accuracy 0.78 on the leaderboard. This is a crude metric because sufficiently loud noise would be considered speech.

Next, I tried running openSMILE, which I use in my research to extract various acoustic features from audio. It implements an LSTM for voice activity detection: every 0.05 seconds, it outputs a probability that someone is talking. Combining the openSMILE output with the VGG16 prediction gave a score of 0.81.

## More improvements

I tried a bunch of things to improve my score:

1. Fiddled around with the neural network hyperparameters which boosted my score to 0.85. Each epoch took about 10 minutes on a GPU, and the whole model takes about 2 hours to train. Somehow, Adam didn’t produce good results, and SGD with momentum worked better.
2. Took 100% of the data for training and used the public LB for validation (don’t do this in real life lol). This improved my score to 0.86.
3. Trained an ensemble 3 versions of the same neural network with same hyperparameters but different randomly initialized weights and took a majority vote to do prediction. This improved the score to 0.87. I would’ve liked to train more, but other people in my research group needed to use the GPUs.

In the end, the top scoring model had a score of 0.91, which beat my model by 4 percentage points. Although not enough to win a Kaggle medal, my model was in the top 15% of all submissions. Not bad!

My source code for the contest is available here.

# What if math contests were scored using Principal Component Analysis?

In many math competitions, all problems are weighted equally, even though the problems have very different difficulties. Sometimes, the harder problems are weighted more. But how should we assign weights to each problem?

Usually, the organizers make up weights based on how difficult they believe the problems are. However, they sometimes misjudge the difficulty of problems. Wouldn’t it be better if the weightings were determined from data?

Let’s try Principal Component Analysis!

Principal Component Analysis (PCA) is a statistical procedure that finds a transformation of the data that maximizes the variance. In our case, the first principal component gives a relative weighting of the problems that maximizes the variance of the total scores. This makes sense because we want to separate the good and bad students in a math contest.

## IMO 2017 Data

The International Mathematics Olympiad (IMO) is an annual math competition for top high school students around the world. It consists of six problems, divided between two days: on each day, contestants are given 4.5 hours to solve three problems.

Here are the 2017 problems, if you want to try them.

Above: Score distribution for IMO 2017

This year, 615 students wrote the IMO. Problems 1 and 4 were the easiest, with the majority of contestants receiving full scores. Problems 3 and 6 were the hardest: only 2 students solved the third problem. Problems 2 and 5 were somewhere in between.

This is a good dataset to play with, because the individual results show what each student scored for every problem.

## Derivation of PCA for the 1-dimensional case

Let $X$ be a matrix containing all the data, where each column represents one problem. There are 615 contestants and 6 problems so $X$ has 615 rows and 6 columns.

We wish to find a weight vector $\vec u \in \mathbb{R}^{6 \times 1}$ such that the variance of $X \vec u$ is maximized. Of course, scaling up $\vec u$ by a constant factor also increases the variance, so we need the constraint that $| \vec u | = 1$.

First, PCA requires that we center $X$ so that the mean for each of the problems is 0, so we subtract each column by its mean. This transformation shifts the total score by a constant, and doesn’t affect the relative weights of the problems.

Now, $X \vec u$ is a vector containing the total scores of all the contestants; its variance is the sum of squares of its elements, or $| X \vec u |^2$.

To maximize $|X \vec u |^2$ subject to $|\vec u| = 1$, we take the singular value decomposition of $X = U \Sigma V^T$. Then, the leftmost column of $V$ (corresponding to the largest singular value) gives $\vec u$ that maximizes $| X \vec u|^2$. This gives the first principal axis, and we are done.

## Experiments

Running PCA on the IMO 2017 data produced interesting results. After re-scaling the weights so that the minimum possible score is 0 and the maximum possible score is 42 (to match IMO’s scoring), PCA recommends the following weights:

• Problem 1: 9.15 points
• Problem 2: 9.73 points
• Problem 3: 0.15 points
• Problem 4: 15.34 points
• Problem 5: 5.59 points
• Problem 6: 2.05 points

This is the weighting that produces the highest variance. That’s right, solving the hardest problem in the history of the IMO would get you a fraction of 1 point. P4 had the highest variance of the six problems, so PCA gave it the highest weight.

The scores and rankings produced by the PCA scheme are reasonably well-correlated with the original scores. Students that did well still did well, and students that did poorly still did poorly. The top students that solved the harder problems (2, 3, 5, 6) usually also solved the easier problems (1 and 2). The students that would be the unhappiest with this scheme are a small number of people who solved P3 or P6, but failed to solve P4.

Here’s a comparison of score distributions with the original and PCA scheme. There is a lot less separation between the best of the best students and the middle of the pack. It is easy to check that PCA does indeed produce higher variance than weighing all six problems equally.

Now, let me comment on the strange results.

It’s clearly absurd to give 0.15 points to the hardest problem on the IMO, and make P4, a much easier problem, be worth 100 times more. But it makes sense from PCA’s perspective. About 99% of the students scored zero on P3, so its variance is very low. Given that PCA has a limited amount of weight to “spend” to increase the total variance, it would be wasteful to use much of it on P3.

The PCA score distribution has less separation between the good students and the best students. However, by giving a lot of weight to P1 and P4, it clearly separates mediocre students that solve one problem from the ones who couldn’t solve anything at all.

In summary, scoring math contests using PCA doesn’t work very well. Although it maximizes overall variance, math contests are asymmetrical as we care about differentiating between the students on the top end of the spectrum.

## Source Code

If you want to play with the data, I uploaded it as a Kaggle dataset.

The code for this analysis is available here.

# Simple models in Kaggle competitions

This week I participated in the Porto Seguro Kaggle competition. Basically, you’re asked to predict a binary variable — whether or not an insurance claim will be filed — based on a bunch of numerical and categorical variables.

With over 5000 teams entering the competition, it was the largest Kaggle competition ever. I guess this is because it’s a fairly well-understood problem (binary classification) with a reasonably sized dataset, making it accessible to beginning data scientists.

Kaggle is a machine learning competition platform filled with thousands of smart data scientists, machine learning experts, and statistics PhDs, and I am not one of them. Still, I was curious to see how my relatively simple tools would fare against the sophisticated techniques on the leaderboard.

The first thing I tried was logistic regression. All you had to do was load the data into memory, invoke the glm() function in R, and output the predictions. Initially my logistic regression wasn’t working properly and I got a negative score. It took a day or so to figure out how to do logistic regression properly, which got me a score of 0.259 on the public leaderboard.

Next, I tried gradient boosted decision trees, which I had learned about in a stats class but never actually used before. In R, this is simple — I just needed to change the glm() call to gbm() and fit the model again. This improved my score to 0.265. It was near the end of the competition so I stopped here.

At this point, the top submission had a score of 0.291, and 0.288 was enough to get a gold medal. Yet despite being within 10% of the top submission in overall accuracy, I was still in the bottom half of the leaderboard, ranking in the 30th percentile.

The public leaderboard looked like this:

Above: Public leaderboard of the Porto Seguro Kaggle competition two days before the deadline. Line in green is my submission, scoring 0.265.

This graph illustrates the nature of this competition. At first, progress is easy, and pretty much anyone who submitted anything that was not “predict all zeros” got over 0.200. From there, you make steady, incremental progress until about 0.280 or so, but afterwards, any further improvements is limited.

The top of the leaderboard is very crowded, with over 1000 teams having the score of 0.287. Many teams used ensembles of XGBoost and LightGBM models with elaborate feature engineering. In the final battle for the private leaderboard, score differences of less than 0.001 translated to hundreds of places on the leaderboard and spelled the difference between victory and defeat.

Above: To run 90% as fast as Usain Bolt, you need to run 100 meters in 10.5 seconds. To get 90% of the winning score in Kaggle, you just need to call glm().

This pattern is common in Kaggle and machine learning — often, a simple model can do quite well, at least the same order of magnitude as a highly optimized solution. It’s quite remarkable that you can get a decent solution with a day or two of work, and then, 5000 smart people working for 2 months can only improve it by 10%. Perhaps this is obvious to someone doing machine learning long enough, but we should look back and consider how rare this is. The same does not apply to most activities. You cannot play piano for two days and become 90% as good as a concert pianist. Likewise, you cannot train for two days and run 90% as fast as Usain Bolt.

Simple models won’t win you Kaggle competitions, but we shouldn’t understate their effectiveness. Not only are they quick to develop, but they are also easier to interpret, and can be trained in a few seconds rather than hours. It’s comforting to see how far you can get with simple solutions — the gap between the best and the rest isn’t so big after all.

Read further discussion of this post on the Kaggle forums!