1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
|
.TH ML_CONFIG "5" "September 2024" "ml 0.5.0" "Configuration file"
.SH NAME
ml_config - ml configuration file
.SH DESCRIPTION
.PP
ml configuration file format consists of lines of key-value options grouped in
multiple sections as is shown below in the demo configuration file:
.PP
.EX
.RS 4
; comments
[net]
loss = square ; options (square)
epochs = 500 ; comment
batch = 32
alpha = 1
weights_path = utils/weights.bin
inputs = sepal_length,sepal_width,petal_length,petal_width
labels = species
[categorical_fields]
species=setosa,versicolor,virginica
[preprocessing]
onehot=species
[layer]
neurons=10
; options (relu, sigmoid, softplus, leaky_relu, linear, tanh)
activation=sigmoid
[outlayer]
activation = sigmoid
.RE
.EE
.SS [net]
.PP
In this section you describe general network configurations more specifically
its training parameter, weights filepath and input and output columns.
.TS
box nospaces center tab(|);
L L L
Lb L L.
Key | Description | Type
_
alpha | learning rate | decimal
loss | loss function | option (string)
epochs | training epochs | integer
batch | batch size | integer
weights_path | weights filepath | string
inputs | input fields | list (string)
labels | label fields | list (string)
.TE
.SS [preprocessing]
Indicate preprocessing operations for input or label fields
.TS
box nospaces center tab(|);
L L L
Lb L L.
Key | Description | Type
_
onehot | onehot columns | list (string)
.TE
.PP
Categorical preprocessing like operations
.B onehot
require to have their fields specified in
.B [categorical_fields].
otherwise the program will prompt you an error message.
.SS [categorical_fields]
.PP
This section is kind of special, here you must set the possible values a field
can take. taking the first example as reference:
.EX
.RS
[categorical_fields]
species=setosa,versicolor,virginica
.RE
.EE
the column
.B species
can only take the values
.B setosa,versicolor
and
.B virginica,
if the program encounters a different value it will stop its execution.
.SS [layer]
.SS [outlayer]
.PP
Create a layer component.
.B [outlayer]
section does not require
.B neurons
parameter.
.TS
box nospaces center tab(:);
L L L
Lb L L.
Key : Description : Type
_
neurons : number of neurons : integer
activation : activation function : option (string)
.TE
.PP
The topology of the network depends of the order in which
.B [layer]
, is put on the file, the program will load each
.B [layer]
section sequentially. For instance the following example:
.PP
.EX
.RS
[layer]
neurons=10
activation=relu
[layer]
neurons=20
activation=sigmoid
[outlayer]
activation = sigmoid
.RE
.EE
will produce a network with the following topology
.TS
nospaces center tab(:);
Lb | Lb
L | L.
Activation : Dimension
_
ReLu : in \[mu] 10
Sigmoid : 10 \[mu] 20
outlayers : 20 \[mu] out
.TE
.SH FILES
.PP
The configuration file is searched in the following order:
.PP
.PD 0
.IP \(bu 4
CLI filepath
.IP \(bu 4
ML_CONFIG_PATH env variable
.IP \(bu 4
$HOME/.\&config/ml/ml.cfg
.PP
.SH AUTHOR
Written by jvech
.SH COPYRIGHT
Copyright \(co 2024 jvech
.PP
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
|
