- Controllable Parameter Combinations: 23,328
- Environmental Parameter Combinations: 9
- Sweep Combinations: 209,952
- MC Runs: 3
- Total Trajectories: 629,856
- Timesteps: 1,096
- Total State Measurements: 690,322,176
- Execution Time on a Mac M1, 4 Jobs: 274,000 measurements on 250 trajectories in 232 seconds
- 1,181 M/s
- 295 M/(J*s)
- Under the current numbers, the full simulation would take 162 hours to execute.
- If executed with 100 Jobs, it would take 6.5 hours to execute.
- Execution Time on a Mac M1, 4 Jobs: 438,400 measurements on 400 trajectories in 419 seconds
- 1,047 M/s
- 262 M/(J*s)
| Machine | -d (Days) | -s (Monte Carlo) | -sw (Sweep Samples) | N_jobs (number of processes) | N_t (total timesteps) | N_sweeps (sweeps per process) | N_mc | N_trajectories | N_measurements | Duration(s) | M/s | M/(J*S) | Dataset Compressed | Dataset in Memory |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Danilo | 3*365 | 3 | 83 | 4 | 1096 | 10 | 3 | 250 | 274,000 | 232 | 1,181 | 295 | x | x |
| YGG | 3*365 | 3 | 230 | 23 | 1096 | 5 | 3 | 690 | 756,240 | 126 | 6,001.88 | 260.95 | 137MB | 0.69GB |
| YGG | 3*365 | 3 | 2300 | 23 | 1096 | 5 | 3 | 6900 | 7,562,400 | 1,261.71 | 5,993.78 | 260.60 | 1.4GB | 6.9GB |
What can be achieved in 4 hours of compute?
- YGG is getting 6K measurements / second with 23 jobs.
- How many samples can we sweep in 4 hours?
- X * 1096 * 3 / 6000 = 4*60*60
- X = 4*60*60 * 6000 / (1096 * 3) = 26277 sweep samples.
- Running now with
RETURN_SIM_DF: bool = Falseto eliminate memory constraint in combining the parts. - Calculating trajectory_tensor batch-wise