diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..ce07276
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,2 @@
+**/__pycache__/*
+example/*.h5
\ No newline at end of file
diff --git a/Dockerfile b/Dockerfile
new file mode 100644
index 0000000..794128a
--- /dev/null
+++ b/Dockerfile
@@ -0,0 +1,5 @@
+FROM tensorflow/tensorflow:2.3.0rc2-gpu-jupyter
+
+RUN python -m pip install --upgrade pip
+COPY requirements.txt .
+RUN pip install -r requirements.txt
diff --git a/parametric_tSNE/core.py b/parametric_tSNE/core.py
index ff39bd0..7fb0400 100644
--- a/parametric_tSNE/core.py
+++ b/parametric_tSNE/core.py
@@ -18,8 +18,8 @@
 import numpy as np
 
 import tensorflow as tf
-from tensorflow.contrib.keras import models
-from tensorflow.contrib.keras import layers
+from tensorflow.keras import models
+from tensorflow.keras import layers
 
 from .utils import calc_betas_loop
 from .utils import get_squared_cross_diff_np
@@ -147,8 +147,8 @@ def _get_normed_sym_tf(X_, batch_size):
         symmetric probabilities, making the assumption that P(i|j) = P(j|i)
         Diagonals are all 0s."""
     toset = tf.constant(0, shape=[batch_size], dtype=X_.dtype)
-    X_ = tf.matrix_set_diag(X_, toset)
-    norm_facs = tf.reduce_sum(X_, axis=0, keep_dims=True)
+    X_ = tf.linalg.set_diag(X_, toset)
+    norm_facs = tf.reduce_sum(X_, axis=0, keepdims=True)
     X_ = X_ / norm_facs
     X_ = 0.5*(X_ + tf.transpose(X_))
     
@@ -216,12 +216,12 @@ def kl_loss(y_true, y_pred, alpha=1.0, batch_size=None, num_perplexities=None, _
         #yrange = tf.range(zz*batch_size, (zz+1)*batch_size)
         #cur_beta_P = tf.slice(P_, [zz*batch_size, [-1, batch_size])
         #cur_beta_P = P_
-        kl_matr = tf.multiply(cur_beta_P, tf.log(cur_beta_P + _tf_eps) - tf.log(Q_ + _tf_eps), name='kl_matr')
+        kl_matr = tf.math.multiply(cur_beta_P, tf.math.log(cur_beta_P + _tf_eps) - tf.math.log(Q_ + _tf_eps), name='kl_matr')
         toset = tf.constant(0, shape=[batch_size], dtype=kl_matr.dtype)
-        kl_matr_keep = tf.matrix_set_diag(kl_matr, toset)
+        kl_matr_keep = tf.linalg.set_diag(kl_matr, toset)
         kl_total_cost_cur_beta = tf.reduce_sum(kl_matr_keep)
         kls_per_beta.append(kl_total_cost_cur_beta)
-    kl_total_cost = tf.add_n(kls_per_beta)
+    kl_total_cost = tf.math.add_n(kls_per_beta)
     #kl_total_cost = kl_total_cost_cur_beta
     
     return kl_total_cost
@@ -272,7 +272,7 @@ def __init__(self, num_inputs, num_outputs, perplexities,
         self.do_pretrain = do_pretrain
         self._loss_func = None
         
-        tf.set_random_seed(seed)
+        tf.random.set_seed(seed)
         np.random.seed(seed)
         
         # If no layers provided, use the same architecture as van der maaten 2009 paper
diff --git a/requirements.txt b/requirements.txt
index aa60ac2..130403b 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,10 +1,6 @@
-numpy
-# Tensorflow API is rapidly evolving (last time I checked)
-# This is the version I designed against
-tensorflow~=1.4.0
 # For saving/loading trained model
 h5py
 # Optional, for example script
-matplotlib~=2.0.0
-seaborn~=0.8.0
-scikit-learn~=0.19.1
+matplotlib>=2.0.0
+seaborn>=0.8.0
+scikit-learn>=0.19.1