-
Notifications
You must be signed in to change notification settings - Fork 3
/
Settings.txt
115 lines (88 loc) · 4.82 KB
/
Settings.txt
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
# This file controls the program settings. The file consists of blank, comment,
# and setting lines. Each comment line begins with '#'. Each setting line uses
# the following format:
# <Setting Name> [<data type>] : <value>
# <value>'s data type must match <data type>. You can place any number of spaces
# between the colon and <value> (the spacing after the colon doesn't matter, so
# long as the setting is all on one line). Nothing can appear after <value> on a
# setting line. You CAN NOT change the order of the setting lines.
#
# The reader will ignore all blank and comment lines. Thus, you can add them as
# you see fit, as they will not break the reader.
################################################################################
# Save, Load Settings
# The "Floating Point Precision" setting does not need to match that of the
# saved state. Thus, for example, you can load a save that used singles into a
# network that uses doubles.
#
# The Load/Save File Name, and the Data File Name can not contain spaces. The
# Data File Name should include a file extension.
#
# The code will NOT load the Optimizer if in Extraction mode. In this case, the
# Reader ignores the load optimizer setting.
Load Sol Network State [bool] : False
Load PDE Network State [bool] : False
Load Optimizer State [bool] : False
Load File Name [str] : Burgers_Sine_N50_P5000_Rat_Adam
Save State [Bool] : True
Save File Name [str] : Burgers_Sine_N50_P5000_Rat_Adam
################################################################################
# Mode
# The "Discovery, or Extraction mode" setting controls what the code
# does. The allowed values for this setting are "Discovery" or
# "Extraction". Actually, the reader will interpret any word that begins with
# "d" or "D" as "Discovery" and any word that begins with "e" or "E" as
# "Extraction".
#
# In Discovery mode, the code trains both the Solution and PDE networks. The
# loss consists of data and collocation losses. In this mode, the reader will
# ignore the Extraction settings (so you can too!)
#
# In Extraction mode, the code trains neither network. Instead, it attempts to
# extract a PDE from the PDE network (which means you should load a trained
# PDE network). In this mode, you must set the "Extracted PDE term degree" and
# "Number of Extraction Points" settings. The former controls the maximum degree
# of the terms that we try to extract (the term du_dx has degree one, u*du_dx
# has degree 2, u^2 (du_dx)^3 d^2u_dx^2 has order 6, etc.). The latter controls
# the number of points (in the PDE domain) at which we attempt to fit the PDE.
Discovery, or Extraction mode [str] : Discovery
# Which dataset do we want to use?
DataSet Name [str] : Burgers_Sine_N50_P5000
# Discovery mode specific settings.
Number of Training Collocation Points [int] : 5000
Number of Testing Collocation Points [int] : 1000
# Extraction mode specific settings.
Extracted PDE maximum term degree [int] : 3
Number of Extraction Points [int] : 20000
Train on CPU or GPU [GPU, CPU] : CPU
################################################################################
# PDE settings.
# These settings control the nature of the PDE we try to uncover. Our goal is
# to find a PDE of the following form:
# D_t^m U = N(U, D_x U, D_x^2 U, ... , D_x^n U)
# Here, m = Time Derivative Order, and n = Spatial Derivative Order.
PDE - Time Derivative Order [int] : 1
PDE - Spatial Derivative Order [int] : 2
################################################################################
# Network Settings
# If you Load the Optimizer state, the Optimizer type must match that of the
# saved optimizer. If you want to switch Optimizers, don't load the Optimizer.
#
# If you plan to load the state of either network (set above), then the network
# Architecture settings below must match those of the saved network. (same
# number of layers and neurons per layer).
#
# Each Network's Activation function must be in {Rational, Tanh, Sine}
Sol Network - Number of Hidden Layers [int] : 5
Sol Network - Neurons per Hidden Layer [int] : 50
Sol Network - Activation Function [str] : Rat
PDE Network - Normalize Inputs [bool] : False
PDE Network - Number of Hidden Layers [int] : 2
PDE Network - Neurons per Hidden Layer [int] : 100
PDE Network - Activation Function [str] : Rat
Optimizer [Adam, LBFGS, SGD] : Adam
################################################################################
# Learning hyper-parameters
Number of Epochs [int] : 1000
Learning Rate [float] : .001
Epochs between testing [int] : 10