Clean up scripts

clberube · Aug 5, 2022 · 29faf79 · 29faf79
1 parent 3990b47
commit 29faf79
Show file tree

Hide file tree

Showing 13 changed files with 19 additions and 41 deletions.
diff --git a/0-create_data_LHS.py b/0-create_data_LHS.py
@@ -4,7 +4,7 @@
 # @Author: charles
 # @Date:   2021-01-24 20:01:11
 # @Last modified by:   charles
-# @Last modified time: 2021-01-24 20:01:57
+# @Last modified time: 2022-08-05 15:08:83
 
 
 import torch

diff --git a/1-plot_fits_learning_curves.py b/1-plot_fits_learning_curves.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-08 14:09:74
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-08 14:09:99
+# @Last modified time: 2022-08-05 15:08:27
 
 
 import os

diff --git a/2-dimension_reduction.py b/2-dimension_reduction.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-08 14:09:74
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-08 14:09:99
+# @Last modified time: 2022-08-05 15:08:49
 
 
 import os

diff --git a/3-sensitivity_analysis.py b/3-sensitivity_analysis.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-08 14:09:74
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-08 14:09:99
+# @Last modified time: 2022-08-05 15:08:85
 
 
 import numpy as np

diff --git a/4a-parameter_estimation.py b/4a-parameter_estimation.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-08 14:09:74
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-08 14:09:99
+# @Last modified time: 2022-08-05 15:08:85
 
 
 import os

diff --git a/4b-cond_parameter_estimation.py b/4b-cond_parameter_estimation.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-08 14:09:74
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-08 14:09:99
+# @Last modified time: 2022-08-05 15:08:68
 
 
 import string

diff --git a/5a-generative_para.py b/5a-generative_para.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-08 14:09:74
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-08 14:09:99
+# @Last modified time: 2022-08-05 15:08:30
 
 
 import os

diff --git a/5b-cond_generative_para.py b/5b-cond_generative_para.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-08 14:09:74
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-08 14:09:99
+# @Last modified time: 2022-08-05 15:08:85
 
 
 import numpy as np

diff --git a/autoencoders/__init__.py b/autoencoders/__init__.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-09 15:09:95
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-14 13:09:49
+# @Last modified time: 2022-08-05 15:08:12
 
 
 from .lstm_ae import LSTM_AE

diff --git a/autoencoders/reg_VAE.py b/autoencoders/reg_VAE.py
@@ -1,15 +1,14 @@
 # @Author: charles
 # @Date:   2021-05-15 15:05:15
+# @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-05-15 15:05:38
+# @Last modified time: 2022-08-05 15:08:10
 
 
 import math
 
 import torch
 import torch.nn as nn
-import torch.nn.functional as F
-import torch.nn.utils.prune as prune
 
 from utilities import softclip
 
@@ -39,11 +38,11 @@ def __init__(self, input_dim, num_hidden=3, hidden_dim=128, latent_dim=2,
         self.log_sigma = torch.zeros([])
 
     def reconstruction_loss(self, x_hat, x):
-        """ Computes the NLL of the data given the latent variable.
+        """ Computes the NLL of x given z.
         """
         self.log_sigma = ((x - x_hat) ** 2).mean([0, 1], keepdim=True).sqrt().log()
-        # Learning the variance can become unstable in some cases. Softly limiting log_sigma to a minimum of -6
-        # ensures stable training.
+        # Softly limiting log_sigma to a minimum of -6
+        # ensures stable training. (not actually necessary here)
         log_sigma = softclip(self.log_sigma, -6)
         rec = self.gaussian_nll(x_hat, log_sigma, x).sum()
         return rec

diff --git a/plotlib.py b/plotlib.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-24 09:09:18
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-24 09:09:02
+# @Last modified time: 2022-08-05 15:08:87
 
 
 import numpy as np

diff --git a/ppip_model.py b/ppip_model.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-08 13:09:44
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-08 13:09:13
+# @Last modified time: 2022-08-05 15:08:16
 
 
 import numpy as np

diff --git a/utilities.py b/utilities.py
@@ -2,7 +2,7 @@
 # @Date:   2021-09-08 14:09:79
 # @Email:  charles.berube@polymtl.ca
 # @Last modified by:   charles
-# @Last modified time: 2021-09-08 14:09:28
+# @Last modified time: 2022-08-05 15:08:61
 
 
 import os
@@ -21,11 +21,8 @@
     "ignore", message="Initializing zero-element tensors is a no-op")
 
 
-def softclip(tensor, min):
-    """ Clips the tensor values at the minimum value min in a softway.
-    Taken from Handful of Trials """
-    result_tensor = min + F.softplus(tensor - min)
-    return result_tensor
+def softclip(x, min):
+    return min + F.softplus(x - min)
 
 
 def min_max_scale(x):
@@ -169,9 +166,6 @@ def train(model, train_loader, verbose, lr, n_epoch, device=None, beta=None,
     if device is None:
         device = 'cuda' if torch.cuda.is_available() else 'cpu'
 
-    # if beta is None:
-    #     beta = torch.ones(n_epoch)
-
     train_losses = ['log_sigma', 'NLL', 'KLD', 'AUX', 'train']
     valid_losses = ['valid']
     grads = ['input_grad_total']
@@ -182,10 +176,6 @@ def train(model, train_loader, verbose, lr, n_epoch, device=None, beta=None,
 
     optimizer = torch.optim.Adam(model.parameters(), lr=lr)
 
-    # reg_loss = torch.nn.MSELoss(reduction='sum')
-    # reg_loss = torch.nn.L1Loss(reduction='mean')
-    # mlp_loss = torch.nn.BCEWithLogitsLoss()
-
     start_time = timer()
     model.to(device)
     for e in range(n_epoch):
@@ -205,10 +195,7 @@ def train(model, train_loader, verbose, lr, n_epoch, device=None, beta=None,
             Xp, mu, logvar, p = model(X, c)
             if y.shape[-1] == 0:
                 AUX = torch.tensor(0)
-            # elif y.shape[-1] == 1:
-            #     AUX = mlp_loss(p, y)
             elif y.shape[-1] > 1:
-                # AUX = reg_loss(p, y)
                 AUX = model.reconstruction_loss(p, y)
 
             NLL, KLD = model.vae_loss(Xp, X, mu, logvar)
@@ -221,8 +208,6 @@ def train(model, train_loader, verbose, lr, n_epoch, device=None, beta=None,
             running_loss['AUX'] += AUX.item()*X.size(0)
             running_loss['train'] += total_loss.item()
 
-            # if (e + 1 == n_epoch) and (c.shape[-1] > 0):
-            # if (e == 0) and (c.shape[-1] > 0):
             if c.shape[-1] > 0:
                 y_grad_total = torch.autograd.grad(
                     NLL, c, retain_graph=True)[0]
@@ -256,16 +241,10 @@ def train(model, train_loader, verbose, lr, n_epoch, device=None, beta=None,
                 Xp, mu, logvar, p = model(X, c)
                 if y.shape[-1] == 0:
                     AUX = torch.tensor(0)
-                # elif y.shape[-1] == 1:
-                #     AUX = mlp_loss(p, y)
                 elif y.shape[-1] > 1:
-                    # AUX = reg_loss(p, y)
                     AUX = model.reconstruction_loss(p, y)
 
                 NLL, KLD = model.vae_loss(Xp, X, mu, logvar)
-                # total_loss = NLL + beta[e]*KLD + AUX
-                # NLL_scaled = X.shape[-1]*NLL/(2*logvar.exp())
-                # total_loss = X.shape[-1]*sigma.log() + NLL_scaled + KLD + AUX
                 total_loss = NLL + KLD + AUX
                 running_loss['valid'] += total_loss.item()