report.rmd

---
title: "Google Analytics Anomaly Detection"
output:
  html_notebook: 
    fig_caption: yes
    highlight: zenburn
    theme: flatly
    toc: yes
    toc_depth: 1
  pdf_document: default
  word_document: default
  html_document:
    fig_caption: yes
    highlight: pygments
    number_sections: yes
    theme: united
    toc: yes
    toc_depth: 1
editor_options:
  chunk_output_type: inline
---

```{r init stuff , message=FALSE, warning=FALSE, include=FALSE}
## Assuming your working directory is sample-r-project's below
# Required packages
source("../requirements.R")

## Functions needed
source("../Functions/functions.R")

## Theme(s) for ggthemr
# source("../Configuration/custom_ggthemr.R")

## Project settings
source("../Configuration/project_settings.R")

## Authentication with googleapis -----------------------------------
options(
  googleAuthR.scopes.selected =
    c(
      # "https://www.googleapis.com/auth/webmasters",
      "https://www.googleapis.com/auth/analytics",
      "https://www.googleapis.com/auth/analytics.readonly",
      "https://www.googleapis.com/auth/tagmanager.readonly"
      # "https://www.googleapis.com/auth/devstorage.full_control",
      # "https://www.googleapis.com/auth/cloud-platform",
      # "https://www.googleapis.com/auth/bigquery",
      # "https://www.googleapis.com/auth/bigquery.insertdata"
    )
)

googleAuthR::gar_auth(".httr-oauth")
# acc.list <- google_analytics_account_list()
```
# Getting data
Let's get some data to start working with. At this point the data are coming from Google Analytics so in the following we need to setup the basic parameters


```{r parameters setup, message=FALSE, warning=FALSE, include=FALSE}
# Date range to fetch
start <- floor_date(Sys.Date() - months(6), "month")
end <- date(Sys.time()) - 1

## Getting the GA data ------------------------------------------
## Define the ID of the VIEW we need to fetch
id <- "YOUR_VIEW_ID" # this is for the internal/legacy/YOU_NAME_IT...

## Build the event list we are interested
## in monitoring for the V1.0
events_category <- c(
  # YOUR_EVENTS_LIST
)

## Dimensions for breakdown
dimensions <- c(
  # YOUR_DIMENSIONS_LIST
)
# ## Parameters for the custom functions
# direction <- "neg"
# anoms <- 0.02
```

### Get the data

Now, we are pulling the data from Google Analytics API using purrr's `map_df()`, which is awesome.

```{r munge the data, echo=TRUE, fig.height=12, fig.width=15, message=FALSE, warning=FALSE, cache=TRUE}
## Get the data from GA
ga_data <- events_category %>%
  map_df(~ get_ga_data(id, start, end, .x, breakdown_dimensions = dimensions))
```

In order to check that we actually got data we can get a summary of the `ga_data`.

```{r inspect data from ga, echo=TRUE}
# Summary of what we got from GA API
# Look for strange things in the 'n_unique' column of dimensions
# and 5-num summary of metrics (ie totalEvents)
ga_data %>%
  skimr::skim_to_wide()
```

Below we filter stuff we are *actually* interested in. Note that we have to do our own sanity check of inputs to the data that we pass to prophet object! This is out of the scope of the current implementation. So use the section below for passing over the constrains you'd like to.
```{r filter out tablet, fig.height=12, fig.width=15, message=FALSE, warning=FALSE,echo=FALSE}
data <- ga_data %>%
  filter(deviceCategory != "tablet")

## Let's keep the most important stuff
channel_groups <- c("Direct", "Non Brand SEO", "Brand SEO", "SEM Brand", "SEM Non Brand")
landing_groups <- c(
  # YOUR_LANDING_PAGE_GROUP_LIST
  )
```

## Run the prophet!
Now we can move on with applying iteratively the prophet method to our own time series of data.
```{r run the prophet, echo=TRUE, fig.height=12, fig.width=15, message=FALSE, warning=FALSE}
## Apply the prophet prediction to each group
prophet_data <- data %>%
  filter(channelGrouping %in% channel_groups &
           landingContentGroup1 %in% landing_groups) %>%
  group_by_if(is.character) %>% # group by all dimensions present to `data`
  nest() %>%
  mutate(prophet_range = map_chr(data, ~ suppressWarnings(
    get_prophet_prediction(.x[["totalEvents"]], start_date = start, verbose = TRUE)
  ))) %>%
  mutate(last_day = map_dbl(data, ~ last(.x[["totalEvents"]]))) %>% # this is the last day ; we'll compare against it
  separate(prophet_range,
           into = c("min", "estimate", "max"),
           sep = ",") %>%
  mutate(
    prophet_lower_range = as.numeric(min),
    prophet_estimate_point = as.numeric(estimate),
    prophet_upper_range = as.numeric(max)
  )
```

# Inspect predictions
Let's check a random 10 rows of prediction along their actual value on the last day of the run.
```{r inspect data, echo=FALSE, fig.height=12, fig.width=15, message=FALSE, warning=FALSE}
prophet_data %>%
  dplyr::select(-min, -max, -estimate, -data) %>%
  mutate_at(vars(starts_with("prophet_")), funs(round(., digits = 2))) %>%
  filter(prophet_lower_range > 0) %>% 
  dplyr::select(-prophet_lower_range, -prophet_upper_range) %>%
  sample_n(10)
```

# Get Alert
Next, we pull all the deviating cases.    
(*NOTE* : If this section is empty then we have no anomalous case)

```{r plot data of interest, echo=TRUE, fig.height=12, fig.width=15, message=FALSE, warning=FALSE}
## Apply the prophet prediction to each group
alert_data <- prophet_data %>%
  rowwise() %>%
  filter(prophet_lower_range > 0) %>%
  mutate(flag = if_else(
    between(last_day, prophet_lower_range, prophet_upper_range),
    0,
    1
  )) %>%
  filter(flag > 0) %>%
  dplyr::select(-min, -max, -estimate, -data) %>%
  mutate_at(vars(starts_with("prophet_")), funs(round(., digits = 2)))
```

If there is a alerting case then we can get the plot of the recent history to get a better understanding of what is going on.

```{r plot the alerting events, echo=TRUE, fig.height=12, fig.width=15, message=FALSE, warning=FALSE}
if (length(alert_data) > 0) {
  alert_data %>%
    knitr::kable()
  ## We can create now the graphs
  alert_graph <- prophet_data %>%
    rowwise() %>%
    filter(prophet_lower_range > 0) %>%
    mutate(flag = if_else(
      between(last_day, prophet_lower_range, prophet_upper_range),
      0,
      1
    )) %>%
    filter(flag > 0) %>%
    dplyr::select(-min, -max, -estimate) %>%
    ungroup() %>%
    mutate(prophet_gg = map(
      data,
      ~ get_prophet_prediction_graph(
        .$"totalEvents",
        start_date = start
      )
    ))

  # Summary stats
  alert_graph %>%
    skimr::skim_to_wide()

  # Plot the alert evolution
  alert_graph %$%
    walk(prophet_gg, plot)
}
```