diff --git a/deeptables/_version.py b/deeptables/_version.py
index 13a85f7..44b1806 100644
--- a/deeptables/_version.py
+++ b/deeptables/_version.py
@@ -1 +1 @@
-__version__ = '0.2.5'
+__version__ = '0.2.6'
diff --git a/deeptables/models/deeptable.py b/deeptables/models/deeptable.py
index 8a3ec1d..66dea62 100644
--- a/deeptables/models/deeptable.py
+++ b/deeptables/models/deeptable.py
@@ -5,7 +5,6 @@
 import pickle
 import time
 
-import dask
 import numpy as np
 import pandas as pd
 from joblib import Parallel, delayed
@@ -415,6 +414,7 @@ def fit_cross_validation(self, X, y, X_eval=None, X_test=None, num_folds=5, stra
             oof_proba = np.full((X_shape[0], 1), np.nan)
 
         if is_dask_installed and DaskToolBox.exist_dask_object(X, y):
+            import dask
             X = DaskToolBox.reset_index(DaskToolBox.to_dask_frame_or_series(X))
             y = DaskToolBox.to_dask_type(y)
             if DaskToolBox.is_dask_dataframe_or_series(y):
diff --git a/deeptables/utils/dataset_generator.py b/deeptables/utils/dataset_generator.py
index 1629a17..4702bd6 100644
--- a/deeptables/utils/dataset_generator.py
+++ b/deeptables/utils/dataset_generator.py
@@ -6,14 +6,12 @@
 from distutils.version import LooseVersion
 from functools import partial
 
-import dask
-import dask.dataframe as dd
 import numpy as np
 import tensorflow as tf
 from tensorflow.keras.utils import to_categorical as tf_to_categorical
 
 from deeptables.utils import consts, dt_logging
-
+from hypernets.tabular import get_tool_box, is_dask_installed
 logger = dt_logging.get_logger(__name__)
 
 TFDG_DASK_CHUNK = 100
@@ -105,6 +103,7 @@ def __call__(self, X, y=None, *, batch_size, shuffle, drop_remainder):
         return ds
 
     def _to_ds20(self, X, y=None, *, batch_size, shuffle, drop_remainder):
+        import dask
         ds_types = {}
         ds_shapes = {}
         meta = self._get_meta(X)
@@ -118,6 +117,7 @@ def _to_ds20(self, X, y=None, *, batch_size, shuffle, drop_remainder):
                 ds_types[k] = 'int32'
 
         if y is not None:
+            import dask.dataframe as dd
             if isinstance(y, dd.Series):
                 y = y.to_dask_array(lengths=True)
             if self.task == consts.TASK_MULTICLASS:
@@ -149,6 +149,7 @@ def to_spec(name, dtype, idx):
         sig = {k: to_spec(k, dtype, idx) for k, (dtype, idx) in meta.items()}
 
         if y is not None:
+            import dask.dataframe as dd
             if isinstance(y, dd.Series):
                 y = y.to_dask_array(lengths=True)
             if self.task == consts.TASK_MULTICLASS:
@@ -167,6 +168,7 @@ def to_spec(name, dtype, idx):
 
     @staticmethod
     def _generate(meta, X, y, *, batch_size, shuffle, drop_remainder):
+        import dask
         total_size = dask.compute(X.shape)[0][0]
         chunk_size = min(total_size, batch_size * TFDG_DASK_CHUNK)
         fn = partial(_TFDGForDask._compute_chunk, X, y, chunk_size)
@@ -205,6 +207,7 @@ def _generate(meta, X, y, *, batch_size, shuffle, drop_remainder):
 
     @staticmethod
     def _to_categorical(y, *, num_classes):
+        import dask
         if len(y.shape) == 1:
             y = y.reshape(dask.compute(y.shape[0])[0], 1)
         fn = partial(tf_to_categorical, num_classes=num_classes, dtype='float32')
@@ -213,6 +216,7 @@ def _to_categorical(y, *, num_classes):
 
     @staticmethod
     def _compute_chunk(X, y, chunk_size, i):
+        import dask
         try:
             Xc = X[i:i + chunk_size]
             yc = y[i:i + chunk_size] if y is not None else None
@@ -236,7 +240,12 @@ def _range(start, stop, step, shuffle):
 def to_dataset(config, task, num_classes, X, y=None, *,
                batch_size, shuffle, drop_remainder,
                categorical_columns, continuous_columns, var_len_categorical_columns):
-    cls = _TFDGForDask if isinstance(X, dd.DataFrame) else _TFDGForPandas
+
+    if is_dask_installed:
+        import dask.dataframe as dd
+        cls = _TFDGForDask if isinstance(X, dd.DataFrame) else _TFDGForPandas
+    else:
+        cls = _TFDGForPandas
     logger.info(f'create dataset generator with {cls.__name__}, '
                 f'batch_size={batch_size}, shuffle={shuffle}, drop_remainder={drop_remainder}')
 
diff --git a/requirements.txt b/requirements.txt
index e8f0e21..7c9cb4f 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -6,7 +6,6 @@ numpy>=1.16.5
 scikit-learn>=0.22.1
 lightgbm>=2.2.0
 category_encoders>=2.1.0
-hypernets>=0.2.5.1
+hypernets>=0.3.0
 h5py>=2.10.0
 eli5
-dask
\ No newline at end of file