Debian Codegraph @

docs/options.md

raw
1 ---
2 id: options
3 title: List of options
4 ---
5
6 Invoke a command without arguments to list available arguments and their default values:
7
8 ```bash
9 $ ./fasttext supervised
10 Empty input or output path.
11
12 The following arguments are mandatory:
13 -input training file path
14 -output output file path
15
16 The following arguments are optional:
17 -verbose verbosity level [2]
18
19 The following arguments for the dictionary are optional:
20 -minCount minimal number of word occurrences [1]
21 -minCountLabel minimal number of label occurrences [0]
22 -wordNgrams max length of word ngram [1]
23 -bucket number of buckets [2000000]
24 -minn min length of char ngram [0]
25 -maxn max length of char ngram [0]
26 -t sampling threshold [0.0001]
27 -label labels prefix [__label__]
28
29 The following arguments for training are optional:
30 -lr learning rate [0.1]
31 -lrUpdateRate change the rate of updates for the learning rate [100]
32 -dim size of word vectors [100]
33 -ws size of the context window [5]
34 -epoch number of epochs [5]
35 -neg number of negatives sampled [5]
36 -loss loss function {ns, hs, softmax} [softmax]
37 -thread number of threads [12]
38 -pretrainedVectors pretrained word vectors for supervised learning []
39 -saveOutput whether output params should be saved [0]
40
41 The following arguments for quantization are optional:
42 -cutoff number of words and ngrams to retain [0]
43 -retrain finetune embeddings if a cutoff is applied [0]
44 -qnorm quantizing the norm separately [0]
45 -qout quantizing the classifier [0]
46 -dsub size of each sub-vector [2]
47 ```
48
49 Defaults may vary by mode. (Word-representation modes `skipgram` and `cbow` use a default `-minCount` of 5.)
50
51
52 Hyperparameter optimization (autotune) is activated when you provide a validation file with `-autotune-validation` argument.
53 ```text
54 The following arguments are for autotune:
55 -autotune-validation validation file to be used for evaluation
56 -autotune-metric metric objective {f1, f1:labelname} [f1]
57 -autotune-predictions number of predictions used for evaluation [1]
58 -autotune-duration maximum duration in seconds [300]
59 -autotune-modelsize constraint model file size [] (empty = do not quantize)
60 ```