Abstract

This dissertation evaluates the extent to which material culture networks, constructed from similarities in artifacts, can serve as effective proxies for past social networks. While commonly assumed in archaeological network studies, this link has rarely been systematically tested. To address this, I integrate agent-based modeling, geometric morphometrics, machine learning, and multilayer network analysis with a case study from the central U.S. Southwest.

The ArchMatNet agent-based model simulates social interaction and material production among hunter-gatherers and small-scale horticulturalists. Key variables such as population size, interaction frequency, learning strategy, and trait visibility can be controlled. Results demonstrate that material culture networks align with social networks under specific conditions, particularly when traits differ between groups and are transmitted through a mix of prestige and conformism.

These findings are tested against an empirical case study. A heuristic projectile point typology was developed using geometric morphometrics and machine learning, based on over 3,000 points from 81 sites in the U.S. Southwest. This typology forms the basis for a regional network analysis, which suggests that projectile points reflect broad-scale interaction but differ in structure from ceramic networks. A multilayer analysis of ceramic, architectural, and lithic data from Tonto Basin further shows that different artifact classes produce distinct network structures. These variations likely reflect diverse social processes, including gendered patterns of production and exchange.

Overall, the results demonstrate that material culture networks can offer meaningful insights into social interaction, but their reliability depends on several factors: trait distinctiveness, transmission dynamics, trait visibility, and data quality. This work contributes to archaeological network analysis by combining experimental and empirical methods and by emphasizing the need to critically assess assumptions behind similarity-based models.

The approaches presented here are broadly applicable, but their interpretive power depends on context. Under the right conditions, material culture can reflect social networks in analytically useful ways, but only if archaeologists remain attentive to its constraints.

NetLogo

NetLogo is a popular tool for building agent based models. A helpful group has built a nice R package for interfacing with NetLogo called NLRX. One of the functions that is really fun to play with is the nldoc_network function. It builds an igraph object using the NetLogo code that graphs the procedure calls in the model.

We’ll use the Bacterial Infection model as an example.

We need to first find the directory it is in. This will vary by operating system and version. Then we can call the function and get an igraph network.

library(nlrx, quietly = T, warn.conflicts = F)
dir = "C:\\Program Files\\NetLogo 6.2.2\\app\\models\\Sample Models\\Biology\\Evolution"
model = file.path(dir,"Bacterial Infection.nlogo")
nw = nldoc_network(model)
nw

We can plot it with igraph.

plot(nw)

Flowcharts

Typically, we don’t present a graph like this when we’re writing about our models. We need something like a flowchart. A flowchart is a kind of graph, but it has a bit more structure. How can we turn this graph into a flowchart?

The diagrammeR package has all of the necessary tools, but the syntax has to be written out in a form like this:

library(DiagrammeR, quietly = T, warn.conflicts = F)
grViz("
digraph boxes_and_circles {

  # a 'graph' statement
  graph [overlap = true, fontsize = 10]

  # several 'node' statements
  node [shape = box,
        fontname = Helvetica]
  A; B; C; D;

  node [shape = circle,
        fixedsize = true,
        width = 0.9] // sets as circles
  1; 2; 3;

  # several 'edge' statements
  A->1 B->2 B->3 C->A
  1->D C->B
}
")

Writing it out like this kind of kills the fun in automatically building our flowchart from the code. But really this is just a character string, so we can paste or glue together what we need without having to type it out.

We can extract a dataframe from our igraph object easily enough. Note: in this model the procedure names use a hyphen, which is common in NetLogo (although I don’t recommend it). The hyphen is a special character in GraphViz (the underlying software we are using in diagrammeR), so we will just remove it.

library(igraph, quietly = T, warn.conflicts = F)
library(dplyr, quietly = T, warn.conflicts = F)
library(stringr, quietly = T, warn.conflicts = F)
edges = igraph::as_data_frame(nw) %>% 
  mutate_all(str_remove_all,pattern = "-")
nodes = V(nw)$name %>% str_remove_all("-")

We will then use a combination of paste and map (the glue package caused a conflict with the github pages build on my website) to format our edges and nodes into the proper shape.

library(purrr, quietly = T, warn.conflicts = F)
diagram = paste("
  digraph {
      graph []
      
      node []
  ",paste(nodes,collapse = '; '),"
      
  ",paste(map2_chr(edges$from,edges$to,function(x,y)paste0(c(x,y), collapse = '->')),collapse = ' '),"
      }"
)
grViz(diagram)

Already this looks like a much better graph visualization than what we had before.

But we can always tweak it and make some improvements. How about adding some color, changing the font, and changing the ellipses to boxes?

diagram = paste(
  "
  digraph {
      graph [fontsize = 10, fontname = Helvetica]
      
      node [shape = box, fillcolor = Beige, style = filled]
  ",paste(nodes,collapse = '; '),"
      
  ",paste(map2_chr(edges$from,edges$to,function(x,y)paste0(c(x,y), collapse = '->')),collapse = ' '),"
      }"
)
grViz(diagram)

The last change we might want to make is to highlight certain nodes. Let’s make it simple and just highlight setup and go. These two nodes are the only procedures directly called by the user, so let’s make them stand out. Let’s also make the arrows straight.

diagram = paste(
  "
  digraph {
      graph [fontsize = 10, fontname = Helvetica, splines = ortho]
      
      node [shape = triangle, fillcolor = Honeydew, style = filled]
    go; setup
      
      node [shape = box, fillcolor = Beige, style = filled]
  ",paste(nodes[which(!nodes %in% c('go','setup'))],collapse = '; '),"
      
  ",paste(map2_chr(edges$from,edges$to,function(x,y)paste0(c(x,y), collapse = '->')),collapse = ' '),"
      }"
)
grViz(diagram)

And that is a much improved flowchart from the initial igraph plot and this code is easily adaptable to any NetLogo model. There are a lot of options that you can use to customize the flowchart even further, but we still need to save the flowchart. We can use the DiagrammeRsvg for that and save to file or save as a png using the magick package as well.

library(DiagrammeRsvg, quietly = T, warn.conflicts = F)
grViz(diagram) %>% export_svg() %>% writeLines(con = "flowChart.svg")
library(magick, quietly = T, warn.conflicts = F)
grViz(diagram) %>% export_svg() %>% image_read_svg() %>% image_write("flowChart.png")

And there you have an automatically generated flowchart, and for most NetLogo models all you would have to do is change the file name and path.

Conclusion

I like that GitHub pages is free, you can be up and running in minutes, and it is endlessly customizable. The drawback is customizing it is not as user friendly as some other sites, but that can be part of the fun as you learn new things.

Selecting an object

Factors to consider when selecting an object are object geometry, surface reflection/transparency, sharp/fine edges, and moving/flowing parts. You need to be able to capture all aspects of the object from multiple camera angles. This can be challenging with small orifices or complicated geometry. For example, the Clay Pipe model below was challenging because I could not capture very far into the interior of the pipe. The size of the object makes a big difference but only because of the difficulty photographing it. If you can take a picture of it then then you can make a 3D model, but you may need special equipment for very small objects. A macro lens is a good investment.

Transparent and reflective images will not work for photogrammetry due to the light distortion, but you can coat the surface with something non-reflective. I’ve tried talcum powder and it worked, but it wasn’t my favorite (see Clovis point below for my talcum example–note that the obsidian looking surface in the 3D model was made using Blender. This is a good study comparing different options.

Another challenge is fine particles like hair or fur or sharp edges. The sharp edges of the obsidian Clovis point took several attempts to capture, but I found that taking lots of pictures helped. Anything that moves freely like hair or fur will be a challenge if it is not in the same relative position from photo to photo. This cradle figurine is a good example of how fur is difficult to capture. You can see that the edges are much fuzzier than other parts of the model. There are also holes where it was difficult to photograph the object due to the angles. I also didn’t want to set the basket upside down to photograph the back so that part is missing. To make the model seem nicer I added a board using Blender to make it look like it was sitting on something.

To start with, I recommend choosing an object with rough texture and relatively simple geometry. It should have some distinguishing features though. I’ve found clay figures work very well. These are pictures of Pillings Figurine No. 3 which is on display at the USU Eastern Prehistoric Museum. While choosing a good artifact is helpful for learning, don’t give up! I have been able to make 3D models work in some pretty bad situations–I even wrote a post about it.

Photography

This is the most important step! It doesn’t matter how good your software is if your photos are garbage or you don’t have enough of them. I’ll be brief in a section that could use a lengthy discussion, and I still have a lot to learn myself.

Masks

I like to use photoshop to make my masks. You can even automate the process. What I do first is copy the photos into a new folder labeled masks. I then edit each photo in the masks folder and save it as is. It may help to add an _mask tag to each filename, but it’s up to you. The mask process is simple. Essentially you just need to select the object using your favorite selection tool. Use the fill tool to make the object white. Invert your selection. Then use the fill tool to make everything else black. Check my post on automating the process to see more details.

Import images and masks

Now we can open metashape. If you don’t have it, then download it on a free trial basis to start. You can follow along with your own photos, or contact me if you want the reduced-resolution sample set used in this tutorial. Click on the Workflow drop down menu and select Add Photos. Select all of your photos and import them. Next, we need to add the masks. You’ll see a folder labeled Cameras under Chunk 1 in the workspace on the left. Click the arrow next to the Cameras folder to see all of the images. Right click on one (pick a photo, any photo) and find the option labeled Masks. Click Import Masks. You will see a menu like the one below. Your method should be From File. The operation should be Replacement. If you are using the test dataset then the filename should look like {filename}.jpg. What the filename structure means is that the software will look for a file with the same name as the images we already imported. If your image is A1.jpg and your mask is called A1_mask.jpg then the Filename template should look like {filename}_mask.jpg. Tolerance doesn’t apply to this operation and you want to make sure the All cameras option is selected in the Apply to section. Press ok, and you can then select the folder where the masks are stored. You should now see the non-object areas of your images grayed out.

Align cameras

Go back to the Workflow menu and choose Align Photos. You’ll see a menu like the one below. You haven’t had to make any real decisions in the workflow up to this point, but now we have a lot of options. Aligning the photos is the most important step in the 3D modeling process. If this step works well then you will get a good 3D model, otherwise you’ll need new photos. Don’t get discouraged if your photos don’t align on the first try though! It can take a lot of playing with the settings if you have a difficult object, but you can get good results with bad photographs (see this article for archaeological examples). While you want your photos to align nicely the first try, higher settings shouldn’t be your first option and you can sometimes get better results with lower settings. The Accuracy option controls whether the image is reduced in size (your original images are not modified) before matching begins. I would try medium first and then work to higher settings if you get bad results. Higher settings can significantly increase the time it takes to align photos. Each of the following steps can take seconds to hours depending on the number of images, masks, and settings. Each of the steps below took less than a minute on my machine, but it may take much longer. You can check the manual for descriptions of every option, as I’ll only point out the ones that you need to change or double check. For this step, just choose your accuracy and make sure that the Apply masks to option reads Key points in the Advanced section. Hit ok and let the software run the calculations.

Click the show cameras option on the menu bar (looks like a camera) if you don’t see the positions of your camera next to the tie points that were just generated. The tie points should generally resemble your object, but don’t worry if some points are scattered about in the wrong positions. You can also check the positions of the cameras around the object. If one looks out of place then you can remove it, disable it, or (best option) try to realign it by right clicking on the photo and using the options listed there. If not all cameras are aligned, the first step I recommend is to try aligning again with higher settings.

Build dense cloud

If all your cameras seem aligned then build your dense cloud by going back to the Workflow menu and choosing Build Dense Cloud. I would start with high quality in this case, as the photos are already reduced. I also usually choose Aggressive for the Depth filtering shown in the Advanced section. Press ok and you should get a result (this part will take the longest) that looks a lot like what you are trying to model.

Build mesh

As you might guess, the next step is to go back to the Workflow menu and choose the next option–Build Mesh. I always choose a high Face count at this step. I leave everything else at its default. For some objects you will need to disable Interpolation in the Advanced section. This will attempt to fill holes which is often necessary, but sometimes it can distort the model and create an innacurate representation. If the model is meant for precise geometric morphometrics or other analysis then interpolation may cause problems. You should now have a 3D model that should even be 3D printable.

Add texture

The last step in the photogrammetry process is to add a texture. This does nothing to the model from a 3D standpoint, but it does add a nice, photorealistic touch that significantly improves the look of your model. Select the Build Texture option from the Workflow menu, and leave everything at the default. In some cases, you may want to increase the texture size for more detail. You should notice much clearer looking details on the model now.

This is what your model should look like if you used the example files (except this model was made on full-resolution photos):

Export

There are other options to manipulate the model in Metashape that can be useful. 3D models are often huge files and your computer may struggle to render it. You can decimate the mesh (reduce the number of triangles) by going to Tools > Mesh > Decimate Mesh. Choose a target face count and the model size will be reduced. The great thing about photogrammetry is that as long as you have the original photos you can reproduce your workflow and create higher resolution models.

Once you are satisified with the model you can export it to a number of formats. Choose File > Export > Export Model. My default format is OBJ as it allows you to export the model texture. I usually use the defaults, except I like to choose png for the texture resolution. I will then use Blender to import the model and resize it or clean up bad parts of the model. Meshlab and Cloud Compare are other, open-sourced software that have a lot of great tools for working with 3D models. The easiest to view and share option is to save as an Adobe pdf. If you want to 3D print it, then export to STL.

Next steps

Now that you have a 3D model what do you do? Share it, print it, and/or analyze it. Sketchfab is probably the easiest way to share your model, although they’ve made it harder to do anything for free lately. 3D printers are a lot cheaper than they used to be, but some libraries now have 3D printing services and there are many online options for 3D printing things. The field of geometric morphometrics has a lot of exciting developments for analyzing shapes and is the next step to explore if you want to start analyzing your 3D models.

If you have any questions, comments, or tips to share, please let me know in the comments or via email.

Create Photoshop Action

The first thing to do is to make a copy of all of the photos you’ll be using and place them in a new folder labeled masks. Open one of these images in Photoshop. Access the Actions menu by pressing F9, and press the Create new action button.

I named my action Automatic Mask. Press the Record button and now Photoshop will keep track of everything we do. Press w to access the Quick Selection Tool and you should see the Select Subject button at the top of the screen.

Press that button and you should see the most prominent object selected. The selection worked fairly well. It’s not surprising the out of focus ground was removed, but I wasn’t sure that my hand would be left out of the selection.

Now go to Edit>Fill and make sure the Content is white, the Blending Mode is Normal, and the Opacity is 100%. Press Ok and the object should be white. Now invert the selection by pressing Ctrl+Shift+I, and repeat the Edit>Fill process except this time choose black. You should now have a black and white mask.

Just save the image and then close it. Make sure those two actions are separate as we are recording them and for some reason closing and saving at the same time doesn’t work well. We are still recording, so open the image again, access the Actions menu and press the Stop recording button. You’ll then have to delete the Open step. If you cannot see the steps in the action then click the dropdown arrow next to the action name. Select Open and press the Delete button. You should see six steps once you’re done.

1. Select Subject
2. Fill
3. Inverse
4. Fill
5. Save
6. Close

Now close the image again.

Run the Action

Photoshop allows you to run the action on each photo in a folder automatically. Go to File>Automate>Batch. The last action you made should automatically be selected, but select your action if it is not already. Choose the folder you saved the images in (make sure it is the folder for masks, not the original images (make sure you have backups just in case). The Destination should be Save and Close, and the Override Action “Save As” Commands box should be checked. Press Ok and it will repeat this action for every image.

Now each image should be a black and white mask. You can see that not every one of these images was perfectly masked, but you can see from the model I made that the masks worked well enough, and this process saved me a lot of time.

The Model

It’s not a perfect model, but I’ve never tried to make a model from an object held in my hand before, so I’ll call this a good result for the circumstances. Feel free to comment if you found this useful or if you have some tips on improvements or a better process.

Step 1 - Remove cube

If you have the default setup for Blender, then the first step is to remove the default cube. This is done by first selecting the object. Right click on the cube and it will be outlined in orange. Then, with the mouse inside the 3D View window (keyboard shortcuts only work if the mouse is hovering in the right area), press X or the delete key. A menu will appear, click delete.

Step 2 - Import 3D model

The next step is to import the 3D model (this tutorial uses OBJ, but is similar for other file types). Select the following from the menu: File→Import→Wavefront (.obj).

Warning! it make take some time to load if the model is large (you may consider decimating the model in the program used to create it or by using Blender itself).

Step 3 - Adjust orientation

Once the object is loaded it should appear in the view screen. Occasionally the object will not be visible, either due to its small size or that the 3D model is located far from the origin. The easiest way to find the model is to press . on the numpad (note that only the numpad can be used for navigation unless changed in the Blender settings). Another method is to use the toolbar on the bottom of the 3D View window (the lowest screen is defaulted to the timeline window and selections from this toolbar will not affect the 3D view; see a general Blender tutorial for more details). Select View→Align View→View selected to view the object. This only works on the selected object. Another way to select an object is to use the Outliner window located in the top right of the default window. Any object in this window can be selected by left-clicking it.

If the imported object looks correct, then next we adjust the origin to make the object easier to manage (assuming the 3D model isn’t tied to a meaningful origin point [it usually isn’t]). The orientation of the selected object can be changed using the Tool Shelf. This panel can be viewed/hidden by pressing T on the keyboard. The first tab on the Tool Shelf is labeled Tools. Select set origin and a menu will appear. Select the first option (Geometry to Origin). This will move the object to the center (use . on the numpad to reorient the 3D Viewer).

Next we will rotate the object. This step is critical as the orientation of the object will be used for sizing the object. R on the keyboard is used to rotate the object. I find it easiest to rotate one axis at a time. Limit the rotation by pressing R and then X, Y, or Z. There is no need to hold down keys when executing sequences on the keyboard. Simply press R once, then press X once. Changing the view will help with orienting the object. First change the view to ortho by pressing 5 on the numpad (while I will be using keyboard shortcuts the corresponding navigation options are under the View menu at the bottom of the 3D View screen). Next, move to Front view by pressing 1 on the numpad. Rotate the object using the keyboard shortcuts above until the objects looks as it should when viewed from the front. Move the mouse in a circle to rotate it and use the shift key to make fine adjustments. Repeat this step for the side view (3 on the numpad) and top view (7 on the numpad). You can view the opposite side by holding down Ctrl. For example, press Ctrl-T to see the bottom view. You may need to move between the views several times to properly orient the object. Remember to limit the rotation to one axis to avoid changing previously oriented axes. We will scale the object along its longest axis. To do this the longest axis must be aligned directly to the X, Y, or Z axis. I use the Y axis in this tutorial.

In the above screenshot the edge of the object farthest above and below the X axis are directly aligned. The green line is the Y axis running through these points. The screenshot shows the top view (7 on the numpad). The Y axis is aligned with the location of the measurement taken from the physical object. The axis lines are a good way to align the model. Drag the arrows in the direction you want to move the object so the longest axis is aligned with the chosen axis (you can also move the object free-form by pressing G and then left-clicking to place the object).

The last step in orienting the object is to apply the changes. Failing to apply the changes will mess up the next step. Press Ctrl-A and a menu will appear. Select Rotation & Scale. I also like to apply the Location as well, although this is not necessary.

Step 4 - Scaling

Blender does not use real world measurements by default. To change this, first select Scene on the properties window located by default on the far right (the icon has a sun, globe, and cylinder). In the Units section select Unit Presets (as shown in the screenshot above) and select Millimeters. Any unit can be used if desired, but millimeter is the default unit for STLs, which is the most commonly used file type for 3D printing. After changing the scene unit the dimensions will now have mm.

In the screenshot above the dimensions are shown in the properties panel. This is different from the properties window on the right (I know it’s confusing). If you do not see this tab press N. This panel has a lot of useful information about the selected object, and most of these attributes can be directly modified on this panel. The Rotation values should all be 0° and the scale values should all be 1. If this is not the case, then please see the last part of Step 3.

To scale the object simply enter the measurement value in millimeters on the appropriate axis. In my example I have aligned the model so that my physical measurement is directly along the Y axis. The method I demonstrate only works if this is the maximum Y length. In my example, the Y axis is currently showing as 5.551 mm. If part of the model extends further along the Y axis then what I measured the scaling will be off.

In the above screenshot, I entered 24mm into the Dimensions box, which is the physical measurement I made. Note that this distorts the model. The Scale now reads 4.323 in the Y axis box. Copy and paste the Y scale to the X and Z boxes using the keyboard shortcuts Ctrl-C and Ctrl-V. You will see that the model is no longer distorted.

The final step is to apply these changes. Press Ctrl_A and apply the Scale. The values in the Scale box will now read 1. Now any measurements made on this model will be correct.

Next you may want to measure something. Here’s a link to get you started — Ruler & Protractor.

Thanks for reading the tutorial and feel free to comment below or email me if you have any questions, or if you have any suggestions on improving this process.

Links to articles discussing the Pilling Figurines:

https://en.wikipedia.org/wiki/Pilling_Figurines

http://etv10news.com/bud-pilling-shares-story-of-rare-discovery-of-fremont-artifacts/

http://archive.sltrib.com/article.php?id=53879624&itype=CMSID

3D Models

<iframe width="640" height="480" src="https://sketchfab.com/models/45e036d22eb0430c840751b5ae8deeb7/embed" frameborder="0" allow="autoplay; fullscreen; vr" mozallowfullscreen="true"
        webkitallowfullscreen="true">
</iframe>
<p style="font-size: 13px; font-weight: normal; margin: 5px; color: #4A4A4A;">
    <a href="https://sketchfab.com/models/45e036d22eb0430c840751b5ae8deeb7?utm_medium=embed&utm_source=website&utm_campain=share-popup"
       target="_blank"
       style="font-weight: bold; color: #1CAAD9;">
        Pilling Figurine No. 1
    </a>
    by
    <a href="https://sketchfab.com/rbischoff?utm_medium=embed&utm_source=website&utm_campain=share-popup"
       target="_blank"
       style="font-weight: bold; color: #1CAAD9;">
        Robert Bischoff
    </a>
    on
    <a href="https://sketchfab.com?utm_medium=embed&utm_source=website&utm_campain=share-popup"
       target="_blank"
       style="font-weight: bold; color: #1CAAD9;">
        Sketchfab
    </a>
</p>