Interacting with and Analyzing Numerai Network Growth

class: center, middle, inverse, title-slide

# Interacting with and Analyzing <br> Numerai Network Growth
## with GraphQL and <code>ggplot2</code><br> <br><br><br>
### Omni Analytics Group

---

## Background

Numerai is a modern-day hedge fund where a large pool of anonymous data scientists submit stock market predictions in a weekly tournament. These predictions are then crowd-sourced into an AI-based ensemble model that performs trades on behalf of Numerai. Over the last 3 years, the Numerai network has seen explosive growth and this case study will teach you how to analyze that growth using R's graphQL capabilities.

![](images/numerai.png)

---

## GraphQL

Numerai's API uses GraphQL as a backend, which is a structured query language much like SQL itself.

- Developed initially as an internal project by Facebook.
- Provides a method for development of APIs much like REST.
- Flexible and rich compared to REST and therefore may be less suitable for more simple web APIs.
- Where REST APIs are organized as a collection of endpoints, GraphQL is organized as a collection of types and fields with their associated datatype specification.

Let's take a look at how to interact with GraphQL in R...

---

## GraphQL in R

There are two primary methods of accessing the Numerai data:

- Directly, using the `ghql` R package.
- Indirectly, by downloading the parsed JSON data from the Numerai API and reading it into R.

We will use `ghql` to keep the steps reproducible.

</p>
---

## `ghql`

Let's begin by installing `ghql`:

```r
install.packages("ghql")
```

Next, we connect to the Numerai API:

```r
library(ghql)

con <- GraphqlClient$new(
  url = "https://api-tournament.numer.ai/"
)
```

This connection object maintains the GraphQL client connection to the Numerai API server. Note that it is an <a href="https://adv-r.hadley.nz/r6.html">R6-style object</a> and hence is initialized with the GraphqlClient’s new() function

---

## Making a Leaderboard Query

We can perform a query with the following:

```r
qry <- Query$new() #create a new query

qry$query('leaderboard', '{
  v2Leaderboard {
    username
    corrRep
    mmcRep
    return_52Weeks
    return_13Weeks
  }
}')

result <- con$exec(qry$queries$leaderboard) #execute query
```

Note that we begin with the initialization of a new instance of the R6 `Query` class, and then call the `query()` method, passing in two arguments:

- The name of the resulting object.
- The raw GraphQL query that is to be executed.

Here 'leaderboard' indicates the name of the query.

---

## Viewing the Results <img src="stickers/book1.png"  width="70px" height="70px">

A quick peak at the raw values returned shows JSON data that we need to parse using the `fromJSON()` function, in order to create a data frame:

```r
print(paste0(substring(result, 1, 50), "..."))
```

```
## [1] "{\"data\":{\"v2Leaderboard\":[{\"corrRep\":0.06688481132..."
```

And the parsed data:

```r
nmr <- fromJSON(result)[[1]]$v2Leaderboard

head(nmr, 3) %>%
  kable()
```

<table>
 <thead>
  <tr>
   <th style="text-align:right;"> corrRep </th>
   <th style="text-align:right;"> mmcRep </th>
   <th style="text-align:right;"> return_13Weeks </th>
   <th style="text-align:right;"> return_52Weeks </th>
   <th style="text-align:left;"> username </th>
  </tr>
 </thead>
<tbody>
  <tr>
   <td style="text-align:right;"> 0.0668848 </td>
   <td style="text-align:right;"> 0.0269104 </td>
   <td style="text-align:right;"> 99.52905 </td>
   <td style="text-align:right;"> NA </td>
   <td style="text-align:left;"> mdl3 </td>
  </tr>
  <tr>
   <td style="text-align:right;"> 0.0651065 </td>
   <td style="text-align:right;"> 0.0270910 </td>
   <td style="text-align:right;"> 101.81269 </td>
   <td style="text-align:right;"> 430.3838 </td>
   <td style="text-align:left;"> benben11 </td>
  </tr>
  <tr>
   <td style="text-align:right;"> 0.0631207 </td>
   <td style="text-align:right;"> 0.0194339 </td>
   <td style="text-align:right;"> NA </td>
   <td style="text-align:right;"> NA </td>
   <td style="text-align:left;"> anna13 </td>
  </tr>
</tbody>
</table>

---

## 52 Week Returns

Now we can use `ggplot()` to visualize aspects of the data!

```r
ggplot(data = nmr, aes(x = return_52Weeks)) +
  geom_histogram(fill = "blue", colour = "black") +
  labs(title = "52 Week Returns on Staked NMR in the 'Classic' Tournament", x = "52 Week Returns")
```

---

## 13 Weekly Returns

We can use `ggplot()` to visualize weekly returns.

```r
ggplot(data = nmr, aes(x = return_13Weeks)) +
  geom_histogram(fill = "blue", colour = "black") +
  labs(title = "13 Week Returns on Staked NMR in the 'Classic' Tournament", x = "13 Week Returns")
```

---

## 52 Week Returns vs 13 Week Returns

```r
ggplot(data = nmr, aes(x = return_13Weeks, y = return_52Weeks)) +
  geom_point() +
  geom_smooth(method = "lm") +
  labs(title = "52 Week Returns vs 13 Week Returns")
```

---

## Correlation vs MMC

```r
ggplot(data = nmr, aes(x = corrRep, y = mmcRep)) +
  geom_point() +
  geom_smooth(method = "lm") +
  labs(title = "Correlation vs MMC")
```

---

## Conclusion

You can now do the following using `ghql` and `jsonlite` R packages:

* Connect to Numerai API
* Create a Query
* Execute a query
* Parse the data

Try it yourself!

</p>