Update README.md

JuanFMontesinos · Oct 5, 2020 · 8734345 · 8734345
1 parent 75bcfb7
commit 8734345
Showing 1 changed file with 7 additions and 1 deletion.
diff --git a/README.md b/README.md
@@ -11,13 +11,19 @@ Pytorch implementation of [mir_eval](https://craffel.github.io/mir_eval/).
 Not backpropagable.
 Algorithm is ~ 5 times faster  
 Tests carried out using torch.Float64, Nvidia Quadro P6000, ADM Threadripper 1920X in a single run
-Sources vary across tests 
+
+**Note: `float64` is preferred to achieve high tolerance during the computation (~1e-15). Using `float32` implies a tolerance ~1e-5 and result may diverge from original's `mir_eval`**  
+
+
+
 
 ```
 bss_eval_sources test...	Compute permutation: False	CPU: 4.324	torch-CPU: 5.121
 .bss_eval_sources test...	Compute permutation: False	CPU: 9.711	GPU: 2.250
 ..bss_eval_sources test...	compute_permutation: True	CPU: 38.775	torch-CPU: 37.519
 bss_eval_sources test...	compute_permutation: True	CPU: 54.358	GPU: 7.257
+*Sources vary across tests  
+** Tolerance <= 1e-3
 ```