—
—
Access this dialog at any time by typing "--help"
This map does not attempt to trace a classified system of music, nor claim to produce a true machinic reflection of high-dimensional latent space…nor conceptualize some citable vibe-machine…NOR offer an emancipatory blueprint to the "minds" of deep-learning models. Rather, it is a "renovated cartography" [1] – a field which constructs relational meaning from abstraction, alienation, and play. Its representation of embeddings is not exhaustive, yet still provides a "momentary ambience" [2] of their somatic experience.
Move through the world with WASD keys.
Click to look around you; press esc to regain your cursor.
Use number keys to dérive [1], détournée [2], or stroll [3].
Also use number keys to switch audio [4], lyrical [5], or combined [6] cartographical layers.
View your position (and where you've been) in the bottom-right minimap.
Audio: UMAP projection of 1024-dimensional audio embeddings.
Lyrical: UMAP projection of 384-dimensional semantic lyric embeddings.
Combined: UMAP projection of 512-dimensional 1:1 merged audio and lyric embeddings.
Hint: Be still for a moment after switching layers. Your song will come back to you.
Embedding coordinates: At the site of each embedding, you will find an abstracted cartographic marker – coordinates derived from UMAP dimension reduction, rounded to the first decimal. (see UMAP calculation)
Neighborhood coordinates: Located in the top-left corner of the HUD, find the mean coordinates of the nearest k embeddings between your positions of origin and arrival upon navigating.
p̄ = (1/k) · Σᵢ eᵢ → [x, z] = UMAP.transform(p̄)
Cosine similarity: Below neighborhood coordinates, observe the degree of angular (dis)similarity between your positions of origin and arrival upon navigating.
cos_sim(A, B) = (A · B) / (‖A‖ · ‖B‖)
(where cos_sim=1 is the most similar)
High dimensional embeddings have been projected onto a 2D UMAP field for human legibility. Parameters have been tuned to preserve as much raw machinic character as possible.
WORLD_SCALE=[-1000, 1000] The coordinate field is scaled to account for UMAP's small default float range:
[-2.36, 10.48] x [2.48, 10.77]
n_neighbors=10 for audio field
n_neighbors=20 for lyrical field
n_neighbors=10 for combined field
Tuned based on trustworthiness and continuity scores from n_neighbors=5→50, comparing UMAP output and raw embedding fidelity. (see full report)
The spectre of your habits, haunting you through temporal space. Shaped by your gestures: dwell time, navigation, and favored cartographies. Represented as the crosshairs in the bottom-right minimap.
It is calculated as a weighted centroid:
for i in visited_tracks:
w_i = dwell_i · Σ(α_k · sim_k(i))
pos_x = Σ(x_i·w_i) / Σ(w_i) = 123.45
pos_z = Σ(z_i·w_i) / Σ(w_i) = -67.89
year = Σ(yr_i·w_i) / Σ(w_i) = 2003.4
α_audio 0.82 α_lyric 0.61
α_year 0.44 α_dérive 1.00
α_détourn 0.50 α_stroll 0.25
In which your time spent dwelling (dwell_i) on similar embeddings (sim_k(i)) shifts the weights (α_k) of sonic versus semantic importance; importance of time; intentional versus aimless purpose. Dérive exerts an attraction towards the axis, so that drifting remains a practice of resisting one's established routes (or perhaps embracing them?).
[1] G. Debord, "Introduction to a Critique of Urban Geography," 1955. https://www.bopsecrets.org/SI/urbgeog.htm.
[2] G. Debord, "Report on the Construction of Situations and on the International Situationist Tendency's Conditions of Organization and Action," June, 1957. https://www.bopsecrets.org/SI/report.htm.
[3] G. Debord, "Theory of the Dérive," 1958. https://www.bopsecrets.org/SI/2.derive.htm.
[4] A. J. Elias, "Psychogeography, Détournement, Cyberspace," New Literary History, vol. 41, no. 4, pp. 821–845, Autumn 2010, doi: 10.1353/nlh.2010.0033.
[5] G. Debord and G. Wolman, "A User's Guide to Détournement," 1956. https://www.bopsecrets.org/SI/detourn.htm.
This map does not attempt to trace a classified system of music, nor claim to produce a true machinic reflection of high-dimensional latent space…nor conceptualize some citable vibe-machine…NOR offer an emancipatory blueprint to the “minds” of deep-learning models. Rather, it is a “renovated cartography” [1] – a field which constructs relational meaning from abstraction, alienation, and play. Its representation of embeddings is not exhaustive, yet still provides a “momentary ambience” [2] of their somatic experience.
Move through the world with WASD keys.
Click to look around you; press esc to regain your cursor.
Use number keys to dérive [1], détourn [2], or stroll [3].
Also use number keys to switch audio [4], lyrical [5], or combined [6] cartographical layers.
View your position (and where you've been) in the bottom-right minimap.
Audio: UMAP projection of 1024-dimensional audio embeddings.
Lyrical: UMAP projection of 384-dimensional semantic lyric embeddings.
Combined: UMAP projection of 512-dimensional 1:1 merged audio and lyric embeddings.
Hint: Be still for a moment after switching layers. Your song will come back to you.
Embedding coordinates: At the site of each embedding, you will find an abstracted cartographic marker – coordinates derived from UMAP dimension reduction, rounded to the first decimal. (see UMAP calculation)
Neighborhood coordinates: Located in the top-left corner of the HUD, find the mean coordinates of the nearest k embeddings between your positions of origin and arrival upon navigating.
p̄ = (1/k) · Σᵢ eᵢ → [x, z] = UMAP.transform(p̄)
Cosine similarity: Below neighborhood coordinates, observe the degree of angular (dis)similarity between your positions of origin and arrival upon navigating.
cos_sim(A, B) = (A · B) / (‖A‖ · ‖B‖)
(where cos_sim=1 is the most similar)
High dimensional embeddings have been projected onto a 2D UMAP field for human legibility. Parameters have been tuned to preserve as much raw machinic character as possible.
WORLD_SCALE=[-1000, 1000] The coordinate field is scaled to account for UMAP's small default float range:
[-2.36, 10.48] x [2.48, 10.77]
n_neighbors=10 for audio field
n_neighbors=20 for lyrical field
n_neighbors=10 for combined field
Tuned based on trustworthiness and continuity scores from n_neighbors=5→50, comparing UMAP output and raw embedding fidelity. (see full report)
The spectre of your habits, haunting you through temporal space. Shaped by your gestures: dwell time, navigation, and favored cartographies. Represented as the crosshairs in the bottom-right minimap.
It is calculated as a weighted centroid:
for i in visited_tracks:
w_i = dwell_i · Σ(α_k · sim_k(i))
pos_x = Σ(x_i·w_i) / Σ(w_i) = 123.45
pos_z = Σ(z_i·w_i) / Σ(w_i) = -67.89
year = Σ(yr_i·w_i) / Σ(w_i) = 2003.4
α_audio 0.82 α_lyric 0.61
α_year 0.44 α_dérive 1.00
α_détourn 0.50 α_stroll 0.25
In which your time spent dwelling (dwell_i) on similar embeddings (sim_k(i)) shifts the weights (α_k) of sonic versus semantic importance; importance of time; intentional versus aimless purpose. Dérive exerts an attraction towards the axis, so that drifting remains a practice of resisting one's established routes (or perhaps embracing them?).
[1] G. Debord, “Introduction to a Critique of Urban Geography,” 1955. https://www.bopsecrets.org/SI/urbgeog.htm.
[2] G. Debord, “Report on the Construction of Situations and on the International Situationist Tendency’s Conditions of Organization and Action,” June, 1957. https://www.bopsecrets.org/SI/report.htm.
[3] G. Debord, “Theory of the Dérive,” 1958. https://www.bopsecrets.org/SI/2.derive.htm.
[4] A. J. Elias, “Psychogeography, Détournement, Cyberspace,” New Literary History, vol. 41, no. 4, pp. 821–845, Autumn 2010, doi: 10.1353/nlh.2010.0033.
[5] G. Debord and G. Wolman, “A User’s Guide to Détournement,” 1956. https://www.bopsecrets.org/SI/detourn.htm.