interacting-with-gpt-3.Rmd

---
title: "OpenAI API in R"
output: html_notebook
---
## Sample notebook

This R Markdown notebook shows how you can interface with the [OpenAI API](https://beta.openai.com/docs/api-reference/introduction).

### Libraries
```{r}
pacman::p_load(tidyverse, ggplot2, httr)
```

### Variables for the OpenAI API

See the list of possible model URLs to query from [here](https://beta.openai.com/docs/api-reference/models/list).

```{r}
# Change this link if you want to use another model (see above)
api_url <- "https://api.openai.com/v1/engines/davinci/completions"

# This is the secret API key from your OpenAI API
apikey <- "sk-*"

# Your prompt is what the mdoel sees to continue
prompt <- "This is a test"

# Tokens decides how large the output is allowed to be (higher = more time)
tokens <- 5

# Generate the completion
output <- POST(api_url, body = list(prompt = prompt, max_tokens = tokens), add_headers(Authorization = paste("Bearer", apikey)), encode = "json")
```

### Output

```{r}
content(output)$choices[[1]]$text
# Example output: [1] " of the ship to see"
```

## Example use

Here, we use the model to generate endings to Danish news article titles' first 5 letters.

```{r}
# Use the Sentida package to analyze sentiment of outputs
pacman::p_load(devtools)
devtools::install_github("Guscode/Sentida")
library(Sentida)

# Dataset generation
df <- tibble(
  title = c(
    "Ukraine om gaslæk i Østersøen:",
    "Næsten 100 procent i flere",
    "Russere flygter i tusindvis: Betaler",
    "Verdenspremiere på Rigshospitalet: Ny behandling",
    "Irans fodboldlandshold fik mundkurv på"
  )
)

tokens <- 15

# Run the API on the titles
df_gpt <- df %>%
  mutate(
    output = map(title, function(x){
      POST(api_url, body = list(prompt = title, max_tokens = tokens), add_headers(Authorization = paste("Bearer", apikey)), encode = "json")
    }),
    data = map(output, content, as="text")
  ) %>% 
  mutate(
    text = as.character(map(data, function(x){jsonlite::fromJSON(x)$choices[[1]][1]})),
    new_title = paste0(title, text),
    sentiment = as.numeric(map(new_title, sentida)),
  )
```

### Inspect the results
```{r}
df_gpt %>% 
  select(sentiment, text, everything())

df_gpt %>%
  mutate(
    new_title_trunc = str_trunc(paste0(title, text), 30)
  ) %>% 
  ggplot() +
  aes(y = new_title_trunc, x = sentiment, fill=sentiment) +
  geom_col() +
  theme_minimal() +
  theme(legend.position="none") +
  labs(title="Sentiment of titles finished with GPT-3 davinci")
```