Added additional input checks

lightgbmlss/distributions/Beta.py

@@ -35,13 +35,20 @@ def __init__(self,
                  response_fn: str = "exp",
                  loss_fn: str = "nll"
                  ):
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if loss_fn not in ["nll", "crps"]:
+            raise ValueError("Invalid loss function. Please choose from 'nll' or 'crps'.")
+
         # Specify Response Functions
-        if response_fn == "exp":
-            response_fn = exp_fn
-        elif response_fn == "softplus":
-            response_fn = softplus_fn
+        response_functions = {"exp": exp_fn, "softplus": softplus_fn}
+        if response_fn in response_functions:
+            response_fn = response_functions[response_fn]
         else:
-            raise ValueError("Invalid response function. Please choose from 'exp' or 'softplus'.")
+            raise ValueError(
+                "Invalid response function. Please choose from 'exp' or 'softplus'.")
 
         # Set the parameters specific to the distribution
         distribution = Beta_Torch

lightgbmlss/distributions/Cauchy.py

@@ -35,13 +35,20 @@ def __init__(self,
                  response_fn: str = "exp",
                  loss_fn: str = "nll"
                  ):
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if loss_fn not in ["nll", "crps"]:
+            raise ValueError("Invalid loss function. Please choose from 'nll' or 'crps'.")
+
         # Specify Response Functions
-        if response_fn == "exp":
-            response_fn = exp_fn
-        elif response_fn == "softplus":
-            response_fn = softplus_fn
+        response_functions = {"exp": exp_fn, "softplus": softplus_fn}
+        if response_fn in response_functions:
+            response_fn = response_functions[response_fn]
         else:
-            raise ValueError("Invalid response function. Please choose from 'exp' or 'softplus'.")
+            raise ValueError(
+                "Invalid response function. Please choose from 'exp' or 'softplus'.")
 
         # Set the parameters specific to the distribution
         distribution = Cauchy_Torch

lightgbmlss/distributions/Expectile.py

@@ -31,6 +31,17 @@ def __init__(self,
                  expectiles: List = [0.1, 0.5, 0.9],
                  penalize_crossing: bool = False,
                  ):
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if not isinstance(expectiles, list):
+            raise ValueError("Expectiles must be a list.")
+        if not all([0 < expectile < 1 for expectile in expectiles]):
+            raise ValueError("Expectiles must be between 0 and 1.")
+        if not isinstance(penalize_crossing, bool):
+            raise ValueError("penalize_crossing must be a boolean. Please choose from True or False.")
+
         # Set the parameters specific to the distribution
         distribution = Expectile_Torch
         torch.distributions.Distribution.set_default_validate_args(False)

lightgbmlss/distributions/Gamma.py

@@ -35,13 +35,20 @@ def __init__(self,
                  response_fn: str = "exp",
                  loss_fn: str = "nll"
                  ):
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if loss_fn not in ["nll", "crps"]:
+            raise ValueError("Invalid loss function. Please choose from 'nll' or 'crps'.")
+
         # Specify Response Functions
-        if response_fn == "exp":
-            response_fn = exp_fn
-        elif response_fn == "softplus":
-            response_fn = softplus_fn
+        response_functions = {"exp": exp_fn, "softplus": softplus_fn}
+        if response_fn in response_functions:
+            response_fn = response_functions[response_fn]
         else:
-            raise ValueError("Invalid response function. Please choose from 'exp' or 'softplus'.")
+            raise ValueError(
+                "Invalid response function. Please choose from 'exp' or 'softplus'.")
 
         # Set the parameters specific to the distribution
         distribution = Gamma_Torch

lightgbmlss/distributions/Gaussian.py

@@ -35,13 +35,20 @@ def __init__(self,
                  response_fn: str = "exp",
                  loss_fn: str = "nll"
                  ):
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if loss_fn not in ["nll", "crps"]:
+            raise ValueError("Invalid loss function. Please choose from 'nll' or 'crps'.")
+
         # Specify Response Functions
-        if response_fn == "exp":
-            response_fn = exp_fn
-        elif response_fn == "softplus":
-            response_fn = softplus_fn
+        response_functions = {"exp": exp_fn, "softplus": softplus_fn}
+        if response_fn in response_functions:
+            response_fn = response_functions[response_fn]
         else:
-            raise ValueError("Invalid response function. Please choose from 'exp' or 'softplus'.")
+            raise ValueError(
+                "Invalid response function. Please choose from 'exp' or 'softplus'.")
 
         # Set the parameters specific to the distribution
         distribution = Gaussian_Torch

lightgbmlss/distributions/Gumbel.py

@@ -35,13 +35,20 @@ def __init__(self,
                  response_fn: str = "exp",
                  loss_fn: str = "nll"
                  ):
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if loss_fn not in ["nll", "crps"]:
+            raise ValueError("Invalid loss function. Please choose from 'nll' or 'crps'.")
+
         # Specify Response Functions
-        if response_fn == "exp":
-            response_fn = exp_fn
-        elif response_fn == "softplus":
-            response_fn = softplus_fn
+        response_functions = {"exp": exp_fn, "softplus": softplus_fn}
+        if response_fn in response_functions:
+            response_fn = response_functions[response_fn]
         else:
-            raise ValueError("Invalid response function. Please choose from 'exp' or 'softplus'.")
+            raise ValueError(
+                "Invalid response function. Please choose from 'exp' or 'softplus'.")
 
         # Set the parameters specific to the distribution
         distribution = Gumbel_Torch

lightgbmlss/distributions/Laplace.py

@@ -35,13 +35,20 @@ def __init__(self,
                  response_fn: str = "exp",
                  loss_fn: str = "nll"
                  ):
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if loss_fn not in ["nll", "crps"]:
+            raise ValueError("Invalid loss function. Please choose from 'nll' or 'crps'.")
+
         # Specify Response Functions
-        if response_fn == "exp":
-            response_fn = exp_fn
-        elif response_fn == "softplus":
-            response_fn = softplus_fn
+        response_functions = {"exp": exp_fn, "softplus": softplus_fn}
+        if response_fn in response_functions:
+            response_fn = response_functions[response_fn]
         else:
-            raise ValueError("Invalid response function. Please choose from 'exp' or 'softplus'.")
+            raise ValueError(
+                "Invalid response function. Please choose from 'exp' or 'softplus'.")
 
         # Set the parameters specific to the distribution
         distribution = Laplace_Torch

lightgbmlss/distributions/LogNormal.py

@@ -35,13 +35,20 @@ def __init__(self,
                  response_fn: str = "exp",
                  loss_fn: str = "nll"
                  ):
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if loss_fn not in ["nll", "crps"]:
+            raise ValueError("Invalid loss function. Please choose from 'nll' or 'crps'.")
+
         # Specify Response Functions
-        if response_fn == "exp":
-            response_fn = exp_fn
-        elif response_fn == "softplus":
-            response_fn = softplus_fn
+        response_functions = {"exp": exp_fn, "softplus": softplus_fn}
+        if response_fn in response_functions:
+            response_fn = response_functions[response_fn]
         else:
-            raise ValueError("Invalid response function. Please choose from 'exp' or 'softplus'.")
+            raise ValueError(
+                "Invalid response function. Please choose from 'exp' or 'softplus'.")
 
         # Set the parameters specific to the distribution
         distribution = LogNormal_Torch

lightgbmlss/distributions/NegativeBinomial.py

@@ -31,31 +31,36 @@ class NegativeBinomial(DistributionClass):
         Response function for transforming the distributional parameters to the correct support. Options are
         "sigmoid" (sigmoid).
     loss_fn: str
-        Loss function. Options are "nll" (negative log-likelihood) or "crps" (continuous ranked probability score).
-        Note that if "crps" is used, the Hessian is set to 1, as the current CRPS version is not twice differentiable.
-        Hence, using the CRPS disregards any variation in the curvature of the loss function.
+        Loss function. Options are "nll" (negative log-likelihood).
     """
     def __init__(self,
                  stabilization: str = "None",
                  response_fn_total_count: str = "relu",
                  response_fn_probs: str = "sigmoid",
                  loss_fn: str = "nll"
                  ):
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if loss_fn not in ["nll"]:
+            raise ValueError("Invalid loss function. Please select 'nll'.")
+
         #  Specify Response Functions for total_count
-        if response_fn_total_count == "exp":
-            response_fn_total_count = exp_fn
-        elif response_fn_total_count == "softplus":
-            response_fn_total_count = softplus_fn
-        elif response_fn_total_count == "relu":
-            response_fn_total_count = relu_fn
+        response_functions_total_count = {"exp": exp_fn, "softplus": softplus_fn, "relu": relu_fn}
+        if response_fn_total_count in response_functions_total_count:
+            response_fn_total_count = response_functions_total_count[response_fn_total_count]
         else:
-            raise ValueError("Invalid response function for total_count. Please choose from 'exp', 'softplus' or relu.")
+            raise ValueError(
+                "Invalid response function for total_count. Please choose from 'exp', 'softplus' or 'relu'.")
 
         #  Specify Response Functions for probs
-        if response_fn_probs == "sigmoid":
-            response_fn_probs = sigmoid_fn
+        response_functions_probs = {"sigmoid": sigmoid_fn}
+        if response_fn_probs in response_functions_probs:
+            response_fn_probs = response_functions_probs[response_fn_probs]
         else:
-            raise ValueError("Invalid response function for probs. Please select 'sigmoid'.")
+            raise ValueError(
+                "Invalid response function for probs. Please select 'sigmoid'.")
 
         # Set the parameters specific to the distribution
         distribution = NegativeBinomial_Torch

lightgbmlss/distributions/Poisson.py

@@ -24,24 +24,27 @@ class Poisson(DistributionClass):
         Response function for transforming the distributional parameters to the correct support. Options are
         "exp" (exponential), "softplus" (softplus) or "relu" (rectified linear unit).
     loss_fn: str
-        Loss function. Options are "nll" (negative log-likelihood) or "crps" (continuous ranked probability score).
-        Note that if "crps" is used, the Hessian is set to 1, as the current CRPS version is not twice differentiable.
-        Hence, using the CRPS disregards any variation in the curvature of the loss function.
+        Loss function. Options are "nll" (negative log-likelihood).
     """
     def __init__(self,
                  stabilization: str = "None",
                  response_fn: str = "relu",
                  loss_fn: str = "nll"
                  ):
-        #  # Specify Response Functions
-        if response_fn == "exp":
-            response_fn = exp_fn
-        elif response_fn == "softplus":
-            response_fn = softplus_fn
-        elif response_fn == "relu":
-            response_fn = relu_fn
+
+        # Input Checks
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Please choose from 'None', 'MAD' or 'L2'.")
+        if loss_fn not in ["nll"]:
+            raise ValueError("Invalid loss function. Please select 'nll'.")
+
+        # Specify Response Functions
+        response_functions = {"exp": exp_fn, "softplus": softplus_fn, "relu": relu_fn}
+        if response_fn in response_functions:
+            response_fn = response_functions[response_fn]
         else:
-            raise ValueError("Invalid response function for total_count. Please choose from 'exp', 'softplus' or relu.")
+            raise ValueError(
+                "Invalid response function for total_count. Please choose from 'exp', 'softplus' or 'relu'.")
 
         # Set the parameters specific to the distribution
         distribution = Poisson_Torch

lightgbmlss/distributions/SplineFlow.py

@@ -59,45 +59,22 @@ def __init__(self,
                  loss_fn: str = "nll"
                  ):
 
-        # Check if stabilization method is valid.
-        if not isinstance(stabilization, str):
-            raise ValueError("stabilization must be a string.")
-        if stabilization not in ["None", "MAD", "L2"]:
-            raise ValueError("Invalid stabilization method. Options are 'None', 'MAD' or 'L2'.")
-
-        # Check if loss function is valid.
-        if not isinstance(loss_fn, str):
-            raise ValueError("loss_fn must be a string.")
-        if loss_fn not in ["nll", "crps"]:
-            raise ValueError("Invalid loss_fn. Options are 'nll' or 'crps'.")
-
-        # Number of parameters
-        if not isinstance(order, str):
-            raise ValueError("order must be a string.")
-        if order == "quadratic":
-            n_params = 2*count_bins + (count_bins-1)
-        elif order == "linear":
-            n_params = 3*count_bins + (count_bins-1)
-        else:
-            raise ValueError("Invalid order specification. Options are 'linear' or 'quadratic'.")
-
         # Specify Target Transform
         if not isinstance(target_support, str):
             raise ValueError("target_support must be a string.")
-        if target_support == "real":
-            target_transform = identity_transform
-            discrete = False
-        elif target_support == "positive":
-            target_transform = SoftplusTransform()
-            discrete = False
-        elif target_support == "positive_integer":
-            target_transform = SoftplusTransform()
-            discrete = True
-        elif target_support == "unit_interval":
-            target_transform = SigmoidTransform()
-            discrete = False
+
+        transforms = {
+            "real": (identity_transform, False),
+            "positive": (SoftplusTransform(), False),
+            "positive_integer": (SoftplusTransform(), True),
+            "unit_interval": (SigmoidTransform(), False)
+        }
+
+        if target_support in transforms:
+            target_transform, discrete = transforms[target_support]
         else:
-            raise ValueError("Invalid target_support. Options are 'real', 'positive', 'positive_integer' or 'unit_interval'.")
+            raise ValueError(
+                "Invalid target_support. Options are 'real', 'positive', 'positive_integer', or 'unit_interval'.")
 
         # Check if count_bins is valid
         if not isinstance(count_bins, int):
@@ -109,6 +86,32 @@ def __init__(self,
         if not isinstance(bound, float):
             raise ValueError("bound must be a float.")
 
+        # Number of parameters
+        if not isinstance(order, str):
+            raise ValueError("order must be a string.")
+
+        order_params = {
+            "quadratic": 2 * count_bins + (count_bins - 1),
+            "linear": 3 * count_bins + (count_bins - 1)
+        }
+
+        if order in order_params:
+            n_params = order_params[order]
+        else:
+            raise ValueError("Invalid order specification. Options are 'linear' or 'quadratic'.")
+
+        # Check if stabilization method is valid.
+        if not isinstance(stabilization, str):
+            raise ValueError("stabilization must be a string.")
+        if stabilization not in ["None", "MAD", "L2"]:
+            raise ValueError("Invalid stabilization method. Options are 'None', 'MAD' or 'L2'.")
+
+        # Check if loss function is valid.
+        if not isinstance(loss_fn, str):
+            raise ValueError("loss_fn must be a string.")
+        if loss_fn not in ["nll", "crps"]:
+            raise ValueError("Invalid loss_fn. Options are 'nll' or 'crps'.")
+
         # Specify parameter dictionary
         param_dict = {f"param_{i + 1}": identity_fn for i in range(n_params)}
         torch.distributions.Distribution.set_default_validate_args(False)