Can we get 99.9% accuracy? Just add ONE more feature before launch The server is suspiciously slow today Ground truth isn’t actually true Stakeholder ghosted after week 2 Server rebooted overnight. RIP my results. Loss went to NaN on step 1 Pet photobombed the video call Works perfectly on my machine This notebook is 2000 cells long The business need has… evolved Requirements changed… again OOM killed at epoch 99 of 100 CUDA out of memory Restart kernel, run all, pray Actually, let’s pivot the entire approach We need more data The labels are completely wrong Forgot to shuffle the dataset Who left training running ALL weekend? SSH connection died mid- experiment Internet died at demo time When will the model be production- ready? Can we make it real-time? Nice pajama top on the call Server crashed during the live demo Jupyter kernel died mysteriously Extreme class imbalance strikes again The data is too noisy to use Dataset has 47 samples. Total. I’ll document this later Someone hogged all GPUs This should be simple, right? Test data leaked into training set “Quick 5- min sync” = 45 minutes As per my last email… Free! Wait… which cell did I run last? Forgot to kill my process… sorry team! A meeting could have been an email Who annotated this?! Can we get 99.9% accuracy? Just add ONE more feature before launch The server is suspiciously slow today Ground truth isn’t actually true Stakeholder ghosted after week 2 Server rebooted overnight. RIP my results. Loss went to NaN on step 1 Pet photobombed the video call Works perfectly on my machine This notebook is 2000 cells long The business need has… evolved Requirements changed… again OOM killed at epoch 99 of 100 CUDA out of memory Restart kernel, run all, pray Actually, let’s pivot the entire approach We need more data The labels are completely wrong Forgot to shuffle the dataset Who left training running ALL weekend? SSH connection died mid- experiment Internet died at demo time When will the model be production- ready? Can we make it real-time? Nice pajama top on the call Server crashed during the live demo Jupyter kernel died mysteriously Extreme class imbalance strikes again The data is too noisy to use Dataset has 47 samples. Total. I’ll document this later Someone hogged all GPUs This should be simple, right? Test data leaked into training set “Quick 5- min sync” = 45 minutes As per my last email… Free! Wait… which cell did I run last? Forgot to kill my process… sorry team! A meeting could have been an email Who annotated this?!
(Print)
Can we get 99.9% accuracy?
Just add ONE more feature before launch
The server is suspiciously slow today
Ground truth isn’t actually true
Stakeholder ghosted after week 2
Server rebooted overnight. RIP my results.
Loss went to NaN on step 1
Pet photobombed the video call
Works perfectly on my machine
This notebook is 2000 cells long
The business need has… evolved
Requirements changed… again
OOM killed at epoch 99 of 100
CUDA out of memory
Restart kernel, run all, pray
Actually, let’s pivot the entire approach
We need more data
The labels are completely wrong
Forgot to shuffle the dataset
Who left training running ALL weekend?
SSH connection died mid-experiment
Internet died at demo time
When will the model be production-ready?
Can we make it real-time?
Nice pajama top on the call
Server crashed during the live demo
Jupyter kernel died mysteriously
Extreme class imbalance strikes again
The data is too noisy to use
Dataset has 47 samples. Total.
I’ll document this later
Someone hogged all GPUs
This should be simple, right?
Test data leaked into training set
“Quick 5-min sync” = 45 minutes
As per my last email…
Free!
Wait… which cell did I run last?
Forgot to kill my process… sorry team!
A meeting could have been an email
Who annotated this?!
