update keras

examples/mnist-keras/client/entrypoint → examples/mnist-keras/client/entrypoint.py

@@ -1,4 +1,3 @@
-#!./.mnist-keras/bin/python
 import json
 import os
 

examples/mnist-keras/client/fedn.yaml

@@ -1,10 +1,12 @@
 python_env: python_env.yaml
 entry_points:
   build:
-    command: python entrypoint init_seed
+    command: python entrypoint.py init_seed
+  startup:
+    command: python get_data.py
   train:
-    command: python entrypoint train $ENTRYPOINT_OPTS
+    command: python entrypoint.py train $ENTRYPOINT_OPTS
   validate:
-    command: python entrypoint validate $ENTRYPOINT_OPTS
+    command: python entrypoint.py validate $ENTRYPOINT_OPTS
   infer:
-    command: python entrypoint infer $ENTRYPOINT_OPTS
+    command: python entrypoint.py infer $ENTRYPOINT_OPTS

examples/mnist-keras/bin/split_data → examples/mnist-keras/client/get_data.py

@@ -1,9 +1,8 @@
-#!./.mnist-keras/bin/python
 import os
 from math import floor
 
-import fire
 import numpy as np
+import tensorflow as tf
 
 
 def splitset(dataset, parts):
@@ -38,5 +37,17 @@ def split(dataset='data/mnist.npz', outdir='data', n_splits=2):
                  y_test=data['y_test'][i])
 
 
+def get_data(out_dir='data'):
+    # Make dir if necessary
+    if not os.path.exists(out_dir):
+        os.mkdir(out_dir)
+
+    # Download data
+    (x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
+    np.savez(f'{out_dir}/mnist.npz', x_train=x_train,
+             y_train=y_train, x_test=x_test, y_test=y_test)
+
+
 if __name__ == '__main__':
-    fire.Fire(split)
+    get_data()
+    split()