diff --git a/Dockerfile b/Dockerfile
index 40237d0e..f24e6428 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -73,7 +73,6 @@ RUN apt-get install -y libfreetype6-dev && \
     # Pinned to match GPU version. Update version together.
     pip install lightgbm==3.2.1 && \
     pip install pydot && \
-    pip install keras && \
     pip install keras-tuner && \
     pip install flake8 && \
     # Pinned because it breaks theano test with the latest version (b/178107003).
@@ -248,10 +247,6 @@ RUN pip install tensorpack && \
 RUN pip install --upgrade cython && \
     pip install --upgrade cysignals && \
     pip install pyfasttext && \
-    # ktext has an explicit dependency on Keras 2.2.4 which is not
-    # compatible with TensorFlow 2.0 (support was added in Keras 2.3.0).
-    # Add the package back once it is fixed upstream.
-    # pip install ktext && \
     pip install fasttext && \
     apt-get install -y libhunspell-dev && pip install hunspell && \
     pip install annoy && \
@@ -450,7 +445,7 @@ ENV PYTHONPATH=$PYTHONPATH:/opt/facets/facets_overview/python/
 ENV MKL_THREADING_LAYER=GNU
 
 # Temporary fixes and patches
-    # Temporary patch for Dask getting downgraded, which breaks Keras
+# Temporary patch for Dask getting downgraded, which breaks Keras
 RUN pip install --upgrade dask && \
     # Stop jupyter nbconvert trying to rewrite its folder hierarchy
     mkdir -p /root/.jupyter && touch /root/.jupyter/jupyter_nbconvert_config.py && touch /root/.jupyter/migrated && \
diff --git a/tests/test_keras.py b/tests/test_keras.py
deleted file mode 100644
index de57de27..00000000
--- a/tests/test_keras.py
+++ /dev/null
@@ -1,82 +0,0 @@
-import unittest
-
-import keras
-import numpy as np
-import pandas as pd
-
-from keras.models import Sequential
-from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPooling2D, LSTM
-from keras.optimizers import RMSprop, SGD
-from keras.utils.np_utils import to_categorical
-
-from common import gpu_test
-
-class TestKeras(unittest.TestCase):
-    def test_train(self):
-        train = pd.read_csv("/input/tests/data/train.csv")
-
-        x_train = train.iloc[:,1:].values.astype('float32')
-        y_train = to_categorical(train.iloc[:,0].astype('int32'))
-
-        model = Sequential()
-        model.add(Dense(units=10, input_dim=784, activation='softmax'))
-
-        model.compile(
-            loss='categorical_crossentropy',
-            optimizer=RMSprop(lr=0.001),
-            metrics=['accuracy'])
-
-        model.fit(x_train, y_train, epochs=1, batch_size=32)
-
-    # Uses convnet which depends on libcudnn when running on GPU
-    def test_conv2d(self):
-        # Generate dummy data
-        x_train = np.random.random((100, 100, 100, 3))
-        y_train = keras.utils.to_categorical(np.random.randint(10, size=(100, 1)), num_classes=10)
-        x_test = np.random.random((20, 100, 100, 3))
-        y_test = keras.utils.to_categorical(np.random.randint(10, size=(20, 1)), num_classes=10)
-
-        model = Sequential()
-        # input: 100x100 images with 3 channels -> (100, 100, 3) tensors.
-        # this applies 32 convolution filters of size 3x3 each.
-        model.add(Conv2D(32, (3, 3), activation='relu', input_shape=(100, 100, 3)))
-        model.add(Conv2D(32, (3, 3), activation='relu'))
-        model.add(MaxPooling2D(pool_size=(2, 2)))
-        model.add(Dropout(0.25))
-
-        model.add(Conv2D(64, (3, 3), activation='relu'))
-        model.add(Conv2D(64, (3, 3), activation='relu'))
-        model.add(MaxPooling2D(pool_size=(2, 2)))
-        model.add(Dropout(0.25))
-
-        model.add(Flatten())
-        model.add(Dense(256, activation='relu'))
-        model.add(Dropout(0.5))
-        model.add(Dense(10, activation='softmax'))
-
-        sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
-
-        # This throws if libcudnn is not properly installed with on a GPU
-        model.compile(loss='categorical_crossentropy', optimizer=sgd)
-        model.fit(x_train, y_train, batch_size=32, epochs=1)
-        
-        model.evaluate(x_test, y_test, batch_size=32)
-
-    def test_lstm(self):
-        x_train = np.random.random((100, 100, 100))
-        y_train = keras.utils.to_categorical(np.random.randint(10, size=(100, 1)), num_classes=10)
-        x_test = np.random.random((20, 100, 100))
-        y_test = keras.utils.to_categorical(np.random.randint(10, size=(20, 1)), num_classes=10)
-
-        sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
-
-        model = Sequential()
-        model.add(LSTM(32, return_sequences=True, input_shape=(100, 100)))
-        model.add(Flatten())
-        model.add(Dense(10, activation='softmax'))
-
-
-        model.compile(loss='categorical_crossentropy', optimizer=sgd)
-        model.fit(x_train, y_train, batch_size=32, epochs=1)
-        model.evaluate(x_test, y_test, batch_size=32)
-