fix breaking dnn input formatting tests

neuro_py/ensemble/decoding/pipeline.py

@@ -2,6 +2,7 @@
 import os
 import random
 
+from re import X
 from typing import List, Tuple, Dict, Optional,  Any
 
 import sklearn.preprocessing
@@ -202,8 +203,11 @@ def zscore_trial_segs(
         Normalized train data, normalized rest features, and normalization parameters.
     """
     is_2D = train[0].ndim == 1
-    concat_train = train if is_2D else np.concatenate(train)
-    train_mean = normparams['X_train_mean'] if normparams is not None else bn.nanmean(concat_train, axis=0)
+    concat_train = train if is_2D else np.concatenate(train).astype(float)
+    train_mean = (
+        normparams['X_train_mean'] if normparams is not None
+        else bn.nanmean(concat_train, axis=0)
+    )
     train_std = normparams['X_train_std'] if normparams is not None else bn.nanstd(concat_train, axis=0)
 
     train_notnan_cols = train_std != 0
@@ -213,15 +217,17 @@ def zscore_trial_segs(
         # if train is not jagged, it gets converted completely to object
         # np.ndarray. Hence, cannot exclusively use normed_train.loc
         if isinstance(normed_train, pd.DataFrame):
-            normed_train = normed_train.loc
-        normed_train[:, train_nan_cols] = 0
+            normed_train.loc[:, train_nan_cols] = 0
+        else:
+            normed_train[:, train_nan_cols] = 0
     else:
         normed_train = np.empty_like(train)
         for i, nsvstseg in enumerate(train):
             zscored = np.divide(nsvstseg-train_mean, train_std, where=train_notnan_cols)
             if isinstance(zscored, pd.DataFrame):
-                zscored = zscored.loc
-            zscored[:, train_nan_cols] = 0
+                zscored.loc[:, train_nan_cols] = 0
+            else:
+                zscored[:, train_nan_cols] = 0
             normed_train[i] = zscored
 
     normed_rest_feats = []
@@ -230,16 +236,18 @@ def zscore_trial_segs(
             if is_2D:
                 normed_feats = np.divide(feats-train_mean, train_std, where=train_notnan_cols)
                 if isinstance(normed_feats, pd.DataFrame):
-                    normed_feats = normed_feats.loc
-                normed_feats[:, train_nan_cols] = 0
+                    normed_feats.loc[:, train_nan_cols] = 0
+                else:
+                    normed_feats[:, train_nan_cols] = 0
                 normed_rest_feats.append(normed_feats)
             else:
                 normed_feats = np.empty_like(feats)
                 for i, trialSegROI in enumerate(feats):
                     zscored = np.divide(feats[i]-train_mean, train_std, where=train_notnan_cols)
                     if isinstance(zscored, pd.DataFrame):
-                        zscored = zscored.loc
-                    zscored[:, train_nan_cols] = 0
+                        zscored.loc[:, train_nan_cols] = 0
+                    else:
+                        zscored[:, train_nan_cols] = 0
                     normed_feats[i] = zscored
                 normed_rest_feats.append(normed_feats)
 
@@ -351,6 +359,13 @@ def minibatchify(
     """
     g_seed = torch.Generator()
     g_seed.manual_seed(seed)
+    if Xtrain.ndim == 2:  # handle object arrays
+        Xtrain = Xtrain.astype(np.float32)
+        Xval = Xval.astype(np.float32)
+        Xtest = Xtest.astype(np.float32)
+        ytrain = ytrain.astype(np.float32)
+        yval = yval.astype(np.float32)
+        ytest = ytest.astype(np.float32)
     train = torch.utils.data.TensorDataset(
         torch.from_numpy(Xtrain).type(torch.float32),
         torch.from_numpy(ytrain).type(torch.float32))