| Parameters: |
-
eqns
(list)
–
Equations to solve in form of list of strings. function and derivatives represented as "u", "ut", "utt",
etc. for first equation. "x", "xt", etc. for second equation. "y", "yt", etc. for third equation.
For including function, i.e. cos(u), use tf.cos(u), or for ln(t), np.log(t). Write equation as would be written in code.
Must be eqn = 0. Rearrange equation to equal 0.
-
orders
(list)
–
list of orders of equations (highest derivative used). Can be 1-3. ex. [1, 3, 2], corresponding to
a highest derivative of "ut", "xttt", "ytt".
-
init_data
(list)
–
list of lists of inital data for each deriviatve. Previously descirbed orders would have
[ [u(t0) ], [ x(t0), xt(t0), xtt(t0) ], [ y(t0), yt(t0) ]], with t0 being inital t in t_bdry.
-
t_bdry
(list, default:
[0, 1]
)
–
list of two elements, the interval of t to be solved on.
-
N_pde
(int, default:
100
)
–
(Number points for differnetial equation).
Number of randomly sampled collocation points along t which PINN uses in training.
-
epochs
(int, default:
1000
)
–
Number of epochs PINN gets trained for.
-
net_layers
(int, default:
4
)
–
Number of internal layers of PINN
-
net_units
(int, default:
40
)
–
Number of units in each internal layer
-
constraint
(string, default:
'soft'
)
–
Determines hard constrainting inital conditions or network learning inital conditions.
Only "soft" implemented currently.
-
model
(PINN, default:
None
)
–
User may pass in user constructed network, however no guarentee of correct training.
|