NNsPOD #221
NNsPOD #221
Conversation
MMRROOO
force-pushed
the
NNsPOD
branch
3 times, most recently
from
be97435
to
37cf70a
Compare
MMRROOO
force-pushed
the
NNsPOD
branch
2 times, most recently
from
820cdd7
to
dea1126
Compare
ezyrb/ann.py
Outdated
| @@ -122,11 +122,10 @@ def fit(self, points, values): | |||
| """ | |||
|
|
|||
| self._build_model(points, values) | |||
| self.optimizer = torch.optim.Adam(self.model.parameters()) | |||
| self.optimizer = torch.optim.Adam(self.model.parameters(), lr = 0.01) | |||
There was a problem hiding this comment.
Can we add the possibility to pass the optimizer (Adam by default) and the learning_rate (0.001 by default)?
ezyrb/ann.py
Outdated
| @@ -143,7 +142,7 @@ def fit(self, points, values): | |||
| elif isinstance(criteria, float): # stop criteria is float | |||
| if loss.item() < criteria: | |||
| flag = False | |||
|
|
|||
| print(loss.item()) | |||
There was a problem hiding this comment.
Can we have something more customizable? Like decide the print frequency (frequency_print should be an optional argument in the constructor) like in this snippet
if epoch % frequency_print == 0 or epoch == 1:
print('[epoch {:05d}] {:.6e}'.format(self.trained_epoch, loss.item()))
ezyrb/ann.py
Outdated
| def predict_tensor(self, new_point): | ||
|
|
||
| return self.model(new_point) |
There was a problem hiding this comment.
Such a method is just a type specification of the predict method. Please avoid that, and instead check the type inside predict and use conversion only if the input is NumPy
ezyrb/ann.py
Outdated
|
|
||
|
|
ezyrb/ann.py
Outdated
|
|
||
|
|
ezyrb/nnspod.py
Outdated
|
|
||
| def reshape2dto1d(self, x, y): | ||
| """ | ||
| reshapes graphable data into data that can go through the Neural Net |
There was a problem hiding this comment.
What do you mean with "graphable" data?
ezyrb/nnspod.py
Outdated
| def reshape2dto1d(self, x, y): | ||
| """ | ||
| reshapes graphable data into data that can go through the Neural Net | ||
| :param array x: x value of data |
ezyrb/nnspod.py
Outdated
| return snapshots.reshape(int(np.sqrt(len(snapshots))), int(np.sqrt(len(snapshots)))) | ||
|
|
||
|
|
||
| def train_InterpNet1d(self,ref_data, interp_layers, interp_function, interp_stop_training, interp_loss, retrain = False): |
ezyrb/nnspod.py
Outdated
|
|
||
|
|
||
|
|
||
| def train_InterpNet2d(self,ref_data, interp_layers, interp_function, interp_stop_training, interp_loss, retrain = False): |
There was a problem hiding this comment.
Please merge function for 1D e 2D. Keep always one general function, instead of specifying several methods for different dimensions
|
In the NNsPOD class the |
MMRROOO
force-pushed
the
NNsPOD
branch
2 times, most recently
from
0f054a3
to
1171901
Compare
| @@ -99,7 +99,7 @@ def _build_model(self, points, values): | |||
| layers_torch.append(nn.Linear(layers[-2], layers[-1])) | |||
| self.model = nn.Sequential(*layers_torch) | |||
|
|
|||
| def fit(self, points, values): | |||
| def fit(self, points, values, optimizer = torch.optim.Adam, learning_rate = 0.001, frequency_print = 0): | |||
There was a problem hiding this comment.
optimizer and learning_rate should be constructor arguments
|
Merged in #232 |
implementation of NNsPOD for ezyrb