Context In 2007, I wrote my masters thesis on indoor location determination using Wi-Fi received signal strength indicator (RSSI). As a part of my research, I gathered 120K RSSI samples from 4 access points on floor 1 and 2 of the building of my faculty. I wrote a custom software to do this (no reliable open source software for this at the time) and sampling was a long and painful process. With my limited ML learning skills at the time, I came up with a simple algorithm that was able to find the location of a Wi-Fi device in that building with 85% accuracy. I imagine this is a solved problem now and state of the art indoor positioning systems are way more accurate today. Yet, I decided to find and publish this data set because it's small and simple enough for practice and at the same time, it has some peculiarity for advanced data science and ML fun. Content The samples were taken in a two-level building. Each row in the dataset is a single RSSI sample from one of the 4 access points. Access points are identified by one of the letters A, B, C, or D. Physical coordinates of the location where each sample was taken is identified by X,Y, Z coordinates with Z being the floor (1 or 2). At each coordinate, multiple samples are taken, where sample number is identified by a field name *sequence*. The reason for taking multiple samples is that signal strength fluctuates due to things like scattering and reflection, specially in buildings with moving objects and people. So, to have reliable measurements, one needs 10s of samples from each access points to make for the variability. Having- said this, each row of the sample set has the format: **ap ,signal, sequence ,x,y,z** where: **ap:** Access point identifier. one of A, B, C or D **signal:** Signal strength from access point *ap*. **Note:** RSSI values in the table are negated. That is, smaller values in this cell mean stronger signal reception from the access point. **sequence:** The sequence of sample from this particular access point at this particular coordinate **x,y,z:**: Coordinates where this sample was taken Note: do not assume all locations have the same number of samples from all access points. For example, at location (x=1,y=1,z=1) we might have *a* samples from *A*, *b* samples from *B*, *c* samples from *C* and *d* samples from *D* and *a != b != c!= d*. Why? That's because in some areas we might have poor reception (or complete lack thereof) from an access point and so we end up with fewer or no samples. Here's the floor plan of the building where sampling took place  These are the locations of the access points: {"A": (23, 17, 2), "B": (23, 41, 2), "C" : (1, 15, 2), "D": (1, 41, 2)} You can also find the location of access points on the map. By poking at this data and comparing it with the floor plan, you'll learn interesting things about Wi-Fi radio signals (in 2.4 GHz frequency) and how they behave in indoors. One challenge is to come up with a ML model that given RSSI from the 4 access points finds the (x,y,z) coordinates of the location.