Merge pull request #197 from deepskies/feature/small_tweaks_mostly_co…

…smetic

Feature/small tweaks mostly cosmetic

.github/workflows/test.yaml

@@ -44,7 +44,3 @@ jobs:
       env:
         PYTHONPATH: ${{ env.PYTHONPATH }}
         ENV_FILE: ${{ runner.workspace }}/.env
-
-    - name: Upload coverage to Codecov
-      run: bash <(curl -s https://codecov.io/bash)
-

CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## [0.1.5] - 2024-11-01
+### Fixed
+- verbosity in data.py and analyze.py
+- can load rs DE model in analyze.py
+- torch.Tensor error addressed in data.py
+
 ## [0.1.4] - 2024-10-28
 ### Added
 - test_data

README.md

@@ -1,7 +1,7 @@
 # DeepUQ
 DeepUQ is a package for injecting and measuring different types of uncertainty in ML models.
 
-[![PyPi](https://img.shields.io/badge/PyPi-0.1.4-blue)](https://pypi.org/project/deepuq/) 
+[![PyPi](https://img.shields.io/badge/PyPi-0.1.5-blue)](https://pypi.org/project/deepuq/) 
 [![License](https://img.shields.io/badge/License-MIT-lightgrey)](https://opensource.org/licenses/MIT)
 [![Downloads](https://static.pepy.tech/personalized-badge/deepuq?period=month&units=international_system&left_color=black&right_color=brightgreen&left_text=Total%20Downloads)](https://pepy.tech/project/deepuq)
 

deepuq/analyze/analyze.py

@@ -39,12 +39,13 @@ def load_checkpoint(
         path="DeepUQResources/checkpoints/",
         BETA=0.5,
         nmodel=1,
-        COEFF=0.5,
-        loss="SDER",
+        COEFF=0.01,
+        loss="DER",
         load_rs_chk=False,
         rs=42,
         load_nh_chk=False,
         nh=64,
+        verbose=False
     ):
         """Loads a PyTorch model checkpoint from a .pt file, constructing the
         file name based on model type (DE or DER) and other configuration
@@ -60,8 +61,8 @@ def load_checkpoint(
             path (str): Directory path to the model checkpoints
             (default: "models/").
             BETA (float): Beta value for DE models (default: 0.5).
-            nmodel (int): Number of models in the ensemble (default: 1).
-            COEFF (float): Coefficient for the DER model (default: 0.5).
+            nmodel (int): Number of the model to select in the DE (default: 1).
+            COEFF (float): Coefficient for the DER model (default: 0.01).
             loss (str): Type of loss used for DER models (e.g., 'SDER', 'DER').
             load_rs_chk (bool): Flag to indicate if random seed checkpoint is
             used (default: False).
@@ -70,6 +71,8 @@ def load_checkpoint(
             configuration is used (default: False).
             nh (int): Number of hidden units if load_nh_chk is True
             (default: 64).
+            verbose (bool): Whether to print out the checkpoint
+            (default: False).
 
         Returns:
             dict: The loaded checkpoint containing model weights and
@@ -82,53 +85,19 @@ def load_checkpoint(
                 + f"{model_name}_{inject_type}_{data_dim}"
                 + f"_noise_{noise}_loss_{loss}_COEFF_{COEFF}_epoch_{epoch}"
             )
-            if load_rs_chk:
-                file_name += f"_rs_{rs}"
-            if load_nh_chk:
-                file_name += f"_n_hidden_{nh}"
-            file_name += ".pt"
         elif model_name[0:2] == "DE":
             file_name = (
                 str(path)
                 + f"{model_name}_{inject_type}_{data_dim}"
-                f"_noise_{noise}_beta_{BETA}_nmodel_{nmodel}_epoch_{epoch}.pt"
+                f"_noise_{noise}_beta_{BETA}_nmodel_{nmodel}_epoch_{epoch}"
             )
-        print("loading this chk", file_name)
+        if load_rs_chk:
+            file_name += f"_rs_{rs}"
+        if load_nh_chk:
+            file_name += f"_n_hidden_{nh}"
+        file_name += ".pt"
+        if verbose:
+            print("loading this chk", file_name)
         checkpoint = torch.load(file_name, map_location=device,
                                 weights_only=False)
         return checkpoint
-
-    def ep_al_checkpoint_DE(self, checkpoint):
-        """Extracts mean and variance validation metrics from a loaded DE model
-        checkpoint.
-
-        Parameters:
-            checkpoint (dict): The loaded DE model checkpoint.
-
-        Returns:
-            tuple: A tuple containing mean validation and variance validation
-            metrics.
-        """
-        # Extract additional information
-        # loaded_epoch = checkpoint.get("epoch", None)
-        mean_validation = checkpoint.get("valid_mean", None).detach().numpy()
-        var_validation = checkpoint.get("valid_var", None).detach().numpy()
-        return mean_validation, var_validation
-
-    def ep_al_checkpoint_DER(self, checkpoint):
-        """Extracts mean and variance validation metrics from a loaded DER
-        model checkpoint.
-
-        Parameters:
-            checkpoint (dict): The loaded DER model checkpoint.
-
-        Returns:
-            tuple: A tuple containing mean validation and variance validation
-            metrics.
-        """
-        mean_u_ep = checkpoint.get("mean_u_ep_validation", None)
-        mean_u_al = checkpoint.get("mean_u_al_validation", None)
-        std_u_ep = checkpoint.get("std_u_ep_validation", None)
-        std_u_al = checkpoint.get("std_u_al_validation", None)
-
-        return mean_u_ep, mean_u_al, std_u_ep, std_u_al

deepuq/data/data.py

@@ -78,7 +78,8 @@ def load_data_h5(self, data_name, path="../data/"):
         loaded_data = {}
         with h5py.File(file_name, "r") as file:
             for key in file.keys():
-                loaded_data[key] = torch.Tensor(file[key][...])
+                loaded_data[key] = torch.tensor(file[key][...],
+                                                dtype=torch.float32)
         return loaded_data
 
 
@@ -422,6 +423,7 @@ def simulate_data_2d(
         image_size=32,
         inject_type="output",
         rs_simulate_2D=40,
+        verbose=False
     ):
         """Simulates 2D image data based on provided parameters and noise
         levels.
@@ -447,6 +449,7 @@ def simulate_data_2d(
                 - "input": Noise is added to the image pixels directly.
                 Default is "output".
             rs (int): Random seed for reproducibility. Default is 40.
+            verbose (bool): Display printouts? Default False.
 
         Returns:
             tuple:
@@ -506,10 +509,10 @@ def simulate_data_2d(
             # total_brightness_prop_noisy.append(np.sum(noisy_image))
         self.input = image_array
         self.output = total_brightness
-        print(
-            f"2D data generated, \
-                with noise injected type: {inject_type}."
-        )
+        if verbose:
+            print(
+                f"2D data generated, with noise injected type: {inject_type}."
+            )
         return image_array, total_brightness
 
     def simulate_data(
@@ -565,7 +568,8 @@ def simulate_data(
         # convert to numpy array (if tensor):
         thetas = np.atleast_2d(thetas)
         n_sim = thetas.shape[0]
-        print("number of sims", n_sim)
+        if verbose:
+            print("number of sims", n_sim)
         # Check if the input has the correct shape
         if thetas.shape[1] != 2:
             raise ValueError(
@@ -587,12 +591,14 @@ def simulate_data(
         # Generate random noise (epsilon) based
         # on a normal distribution with mean 0 and standard deviation sigma
         if vary_sigma:
-            print("YES WERE VARYING SIGMA")
+            if verbose:
+                print("YES WERE VARYING SIGMA")
             new_sig = self.get_sigma_m(sigma, m)
             ε = rs.normal(loc=0, scale=new_sig, size=(len(x), n_sim))
             scale = new_sig
         else:
-            print("NO WERE NOT VARYING SIGMA")
+            if verbose:
+                print("NO WERE NOT VARYING SIGMA")
             ε = rs.normal(loc=0, scale=sigma, size=(len(x), n_sim))
             scale = sigma
         if verbose:
@@ -621,18 +627,17 @@ def simulate_data(
                 x_noisy[:, i] = x + ε[:, i]
 
         if inject_type == "output":
-            # self.input = x
-            self.input = torch.Tensor(np.tile(x, thetas.shape[0]).T)
-            self.output = torch.Tensor(y_noisy.T)
-            # self.output_err = ε[:, i].T
+            self.input = torch.tensor(np.tile(x, thetas.shape[0]).T,
+                                      dtype=torch.float32)
+            self.output = torch.tensor(y_noisy.T, dtype=torch.float32)
         elif inject_type == "input":
-            self.input = torch.Tensor(x_noisy.T)
-            self.output = torch.Tensor(y.T)
+            self.input = torch.tensor(x_noisy.T, dtype=torch.float32)
+            self.output = torch.tensor(y.T, dtype=torch.float32)
             # self.output_err = ε[:, i].T
-        print(
-            f"0D data generated, \
-                with noise injected type: {inject_type}."
-        )
+        if verbose:
+            print(
+                f"0D data generated, with noise injected type: {inject_type}."
+            )
         return
 
     def sample_params_from_prior(