From 46814cdb51578567b66bd6daec25359c15a43ddb Mon Sep 17 00:00:00 2001
From: Jaco DuToit <jacowp357@users.noreply.github.com>
Date: Tue, 3 Oct 2023 11:33:43 +0200
Subject: [PATCH] added web traffic dataset and info (#1326)

* added web traffic dataset and info

* removed CSV file and updated the url
---
 river/datasets/web_traffic.py | 48 +++++++++++++++++++++++++++++++++++
 1 file changed, 48 insertions(+)
 create mode 100644 river/datasets/web_traffic.py

diff --git a/river/datasets/web_traffic.py b/river/datasets/web_traffic.py
new file mode 100644
index 0000000000..9a253f8923
--- /dev/null
+++ b/river/datasets/web_traffic.py
@@ -0,0 +1,48 @@
+from __future__ import annotations
+
+from river import stream
+
+from . import base
+
+
+class WebTraffic(base.RemoteDataset):
+    """Web sessions information from an events company based in South Africa.
+
+    The goal is to predict the number of web sessions in 4 different regions in South Africa.
+
+    The data consists of 15 minute interval traffic values between '2023-06-16 00:00:00' and '2023-09-15 23:45:00' for each region. 
+    Two types of sessions are captured `sessionsA` and `sessionsB`. The `isMissing` flag is equal to 1 if any of the servers failed 
+    to capture sessions, otherwise if all servers functioned properly this flag is equal to 0. Things to consider:
+
+    * region `R5` captures sessions in backup mode. Strictly speaking, `R5` is not necessary to predict.
+    * Can `sessionsA` and `sessionsB` events be predicted accurately for each region over the next day (next 96 intervals)? 
+    * What is the best way to deal with the missing values?
+    * How can model selection be used (a multi-model approach)?
+    * Can dependence (correlation) between regions be utilised for more accurate predictions?
+    * Can both `sessionA` and `sessionB` be predicted simultaneously with one model?
+
+    This dataset is well suited for time series forecasting models, as well as anomaly detection
+    methods. Ideally, the goal is to build a time series forecasting model that is robust to the
+    anomalous events and generalise well on normal operating conditions.
+
+    """
+
+    def __init__(self):
+        super().__init__(
+            url="https://maxhalford.github.io/files/datasets/web-traffic.csv.zip",
+            filename="web-traffic.csv",
+            task=base.REG,
+            n_features=2,
+            n_samples=44_160,
+        )
+
+    def _iter(self):
+        return stream.iter_csv(
+            self.path,
+            target="sessionsA",
+            converters={
+                "region": str,
+                "isMissing": int,
+            },
+            parse_dates={"dateTime": "%Y-%m-%d %H:%M:%S"},
+        )