Merge pull request #32 from ziatdinovmax/mpdkl

Minor bug fixes for vidkl and vigp

gpax/models/vi_mtdkl.py

@@ -217,7 +217,7 @@ def get_mvn_posterior(self,
         z_test = self.nn_module.apply(
             nn_params, jax.random.PRNGKey(0),
             X_new if self.shared_input else X_new[:, :-1])
-        if self.shared_input:
+        if not self.shared_input:
             z_train = jnp.column_stack((z_train, X_train[:, -1]))
             z_test = jnp.column_stack((z_test, X_new[:, -1]))
         # compute kernel matrices for train and test data

gpax/models/vidkl.py

@@ -8,7 +8,7 @@
 """
 
 from functools import partial
-from typing import Callable, Dict, Optional, Tuple
+from typing import Callable, Dict, Optional, Tuple, Union
 
 import jax
 import jax.numpy as jnp
@@ -30,7 +30,7 @@ class viDKL(ExactGP):
 
     Args:
         input_dim:
-            Number of input dimensions
+            Input features dimensions (e.g. 64*64 for a stack of flattened 64-by-64 images)
         z_dim:
             Latent space dimensionality (defaults to 2)
         kernel:
@@ -66,7 +66,7 @@ class viDKL(ExactGP):
         >>> y_mean, y_var = dkl.predict(key2, X_new)
     """
 
-    def __init__(self, input_dim: int, z_dim: int = 2, kernel: str = 'RBF',
+    def __init__(self, input_dim: Union[int, Tuple[int]], z_dim: int = 2, kernel: str = 'RBF',
                  kernel_prior: Optional[Callable[[], Dict[str, jnp.ndarray]]] = None,
                  nn: Optional[Callable[[jnp.ndarray], jnp.ndarray]] = None,
                  latent_prior: Optional[Callable[[jnp.ndarray], Dict[str, jnp.ndarray]]] = None,
@@ -229,6 +229,7 @@ def sample_from_posterior(self, rng_key: jnp.ndarray,
 
     def predict_in_batches(self, rng_key: jnp.ndarray,
                            X_new: jnp.ndarray,  batch_size: int = 100,
+                           params: Optional[Dict[str, jnp.ndarray]] = None,
                            noiseless: bool = False,
                            **kwargs
                            ) -> Tuple[jnp.ndarray, jnp.ndarray]:
@@ -237,10 +238,11 @@ def predict_in_batches(self, rng_key: jnp.ndarray,
         by spitting the input array into chunks ("batches") and running
         self.predict on each of them one-by-one to avoid a memory overflow
         """
-        predict_fn = lambda xi: self.predict(rng_key, xi, noiseless=noiseless, **kwargs)
+        predict_fn = lambda xi: self.predict(
+            rng_key, xi, params, noiseless=noiseless, **kwargs)
         cat_dim = 1 if self.X_train.ndim == len(self.data_dim) + 2 else 0
         mean, var = self._predict_in_batches(
-            rng_key, X_new, batch_size, cat_dim, predict_fn=predict_fn)
+            rng_key, X_new, batch_size, cat_dim, params, predict_fn=predict_fn)
         mean = jnp.concatenate(mean, cat_dim)
         var = jnp.concatenate(var, cat_dim)
         return mean, var
@@ -319,7 +321,7 @@ def single_fit_predict(key):
             self.fit(key, X, y, num_steps, step_size,
                      print_summary, progress_bar, **kwargs)
             mean, var = self.predict_in_batches(
-                key, X_new, batch_size, noiseless, **kwargs)
+                key, X_new, batch_size, None, noiseless, **kwargs)
             return mean, var
 
         if n_models > 1 and ensemble_method not in ["vectorized", "parallel"]:

gpax/models/vigp.py

@@ -138,13 +138,13 @@ def predict_in_batches(self, rng_key: jnp.ndarray,
         """
         predict_fn = lambda xi:  self.predict(
                 rng_key, xi, samples, noiseless, **kwargs)
-        y_pred, y_sampled = self._predict_in_batches(
+        y_pred, y_var = self._predict_in_batches(
             rng_key, X_new, batch_size, 0, samples,
             predict_fn=predict_fn, noiseless=noiseless,
             device=device, **kwargs)
         y_pred = jnp.concatenate(y_pred, 0)
-        y_sampled = jnp.concatenate(y_sampled, -1)
-        return y_pred, y_sampled
+        y_var = jnp.concatenate(y_var, 0)
+        return y_pred, y_var
 
     def predict(self, rng_key: jnp.ndarray, X_new: jnp.ndarray,
                 samples: Optional[Dict[str, jnp.ndarray]] = None,