Layout: FastText Baseline

Created on February 3|Last edited on February 4
Comment
﻿
Optimal Learning Rate
﻿
﻿
﻿
Loss by Learning Rate
Loss by Learning Rate
020406080Step46810121416
Optimal learning rate: 0.5-0.6Optimal learning rate appears to be around 0.5-0.6 (yellows). Lower values show higher loss (red) and higher values also show slightly higher loss (blue).
Vary LR8
Name8 visualized
lr test lr1.00 e25 t5
lr test lr1.00 e25 t5
lr test lr0.90 e25 t5
lr test lr0.90 e25 t5
lr test lr0.80 e25 t5
lr test lr0.80 e25 t5
lr test lr0.60 e25 t5
lr test lr0.60 e25 t5
lr test lr0.50 e25 t5
lr test lr0.50 e25 t5
lr test lr0.30 e25 t5
lr test lr0.30 e25 t5
lr test lr0.20 e25 t5
lr test lr0.20 e25 t5
lr test lr0.10 e25 t5
lr test lr0.10 e25 t5
State
Notes
User
Tags
Created
Runtime
Sweep
epochs
lr
threads
word_vec_dim
ngrams
loss
Finished
stacey
LR
6y ago
13s
-
25
1
5
100
-
4.13772
Finished
stacey
LR
6y ago
14s
-
25
0.9
5
100
-
3.82115
Finished
stacey
LR
6y ago
14s
-
25
0.8
5
100
-
3.61619
Finished
stacey
LR
6y ago
14s
-
25
0.6
5
100
-
3.52926
Finished
stacey
LR
6y ago
14s
-
25
0.5
5
100
-
3.65292
Finished
stacey
LR
6y ago
14s
-
25
0.3
5
100
-
4.55334
Finished
stacey
LR
6y ago
14s
-
25
0.2
5
100
-
5.66532
Finished
stacey
LR
6y ago
14s
-
25
0.1
5
100
-
7.66276
1-8
of 8
﻿
Word Embedding and Ngram Size
﻿
﻿
﻿
Word Dim (green)3
Ngrams (violet)3
﻿
﻿
Add a comment