cmi-detect-behavior-with-sensor-data

[amber apex]

finally linked my discord to kaggle...

[hushed compass]

cross-posting fomr the discussions tab, anyone found a use for the non-BFRB samples?

[smoky wyvern]

I dont get the sequence counter column what is it for since the imus data are themselves sequential ?

[shadow root]

Wdym? Sequence counter is for which time step in one unique sequence it is

[icy roost]

Hey everyone! I’m just starting out in machine learning and would love to collaborate with others. If you're also learning or willing to help a beginner grow, feel free to DM me. Thankssssss

[last heron]

why is tof data given in the form of image?

[tacit loom]

Hi u may check this useful data analysis on kaggle: https://www.kaggle.com/code/tarundirector/sensor-pulse-viz-eda-for-bfrb-detection

3. Time-of-Flight (ToF) Sensors (tof_1_v0 … tof_5_v63)¶
   ⏩ What They Are
   Five 8×8 pixel Time-of-Flight (VL53L7CX) arrays integrated on the watch.
   Measure distance (0–254 units, with –1 meaning “no echo”) from each pixel to the nearest object.
   ⏩ Data Columns
   For each sensor i (1–5), 64 columns: • tof_1_v0, tof_1_v1, …, tof_1_v63 (and similarly for tof_2_…, etc.) • Each v0–v63 represents one pixel’s raw distance reading.
   ⏩ How to Interpret
   Value ≈ 0–50 → object (e.g., skin) is very close (few centimeters).
   Value ≈ 100–200 → object is farther away (10–20 cm).
   –1 → no reflection detected (empty space).
   Spatial pattern: An 8×8 grid of low values forms an “image” of a curved surface (e.g., cheek).
   Temporal change: A cluster of pixels transitioning from –1 → low distance marks the instant of contact.
   Use Case: • Gesture signature: “Cheek-pinch” produces a localized 8×8 hot spot near the face; “drink from cup” shows a different shape. • Phase detection: Before contact, most pixels read –1; on contact, a contiguous block of low values appears.

Hi all! 👋
I’ve got a PhD in ontology and logic, with basic Python and ML/DL skills.
This is my first Kaggle competition, so I’m here to learn, but very motivated to work. I find the topic really fascinating...
If anyone’s looking for teammates, I’d love to join!

[weary trail]

could you please analyze my paper? I recently got onto Ontology from Pure math and physics.. it's the basis of my framework ... in addition to Epistemic Physics

I'm doing another kaggle competition rn, but I love the field... if you find a team and need my skillset hmu 🙂

[exotic shard]

Anyone know if the TOF sensors are transposed to the same orientation as some sort of preprocessing done on the data? I can see from the device they are in two orientations but the data page lists reading off the values in rows of 0-7, 8-15, etc for the 8x8 grid. Wondering if they were already adjusted so that the tof data corresponds with 0 being in the top left orientation across each sensor.

[cyan haven]

Hi everyone, glad to see you all

[subtle dew]

Hi, do we know what's the sampling rate for imu device used in the competition?

[steady talon]

Couple of discussions on this:
https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/discussion/582475
https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/discussion/583676
Consensus seems to be the supplied data is 10Hz, though the IMU itself samples at 1kHz

[subtle dew]

Couldn't find the answer there, thought someone might know here

[steady talon]

That is the answer, though, no? As far as models in the competition are concerned, 10Hz

[somber nymph]

hello guys this is stuck for a long time what could be the reason for this?

[tacit loom]

Based on this user’s comment, I think 10 Hz is a good approximation. In this respect, the mean of the timesteps for each sequence is around 70. If you divide 70 by 6 seconds, you get about 11.67 Hz. So I think that 10 Hz is a good try.

"The videos of the gestures are about 5–6 seconds each, which would imply about 10 Hz sample frequency?"

https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/discussion/582475#3218088

Other users also refer to the technical details of the IMU sensor, but then each sensor could have its own sampling rate.

https://fcon_1000.projects.nitrc.org/indi/cmi_healthy_brain_network/Competitions/Helios2025/IMU_Sensor.pdf

[real jackal]

Looking for a teammate who is just getting started with this problem statement. Intersted ppl dm me.

[sick matrix]

surprising that all the current gold medal spots on LB are solo

[glad light]

this message is a scam

[rare idol]

Hello everyone,
I’m new to Kaggle competitions and a bit confused about how to submit my notebook. Can anyone please share the correct submission code and explain the steps for submitting predictions? I’d really appreciate your guidance.

Thanks in advance! 🙏

[regal hollow]

Hello everyone,
I'm experiencing "Notebook Inference Server Never Started" error when trying to submit my notebook.

My notebook runs successfully locally (completes in 6-8 minutes), but fails during submission with this error. I've tried:

• Reducing model complexity (n_estimators from 400 to 150)
• Simplifying feature extraction
• Using only basic libraries

The notebook completes successfully and generates submission.csv, but the server never starts during submission.

Any suggestions on how to resolve this issue? Thank you for your help!