Merge pull request #480 from ModECI/keras_interface

Keras interface

.gitignore

@@ -278,6 +278,7 @@ Thumbs.db #thumbnail cache on Windows
 /examples/NeuroML/arm64
 /examples/MDF/RNN/iaf.net
 /examples/MDF/RNN/iaf.net2
+/examples/TensorFlow/Keras/keras_to_MDF
 /examples/NeuroML/PyNN/All*
 /examples/NeuroML/PyNN/*dat
 /examples/NeuroML/PyNN/*1.png
@@ -288,6 +289,10 @@ Thumbs.db #thumbnail cache on Windows
 /examples/NeuroML/PyNN/*.mdf
 /examples/NeuroML/PyNN/*.spikes
 *_code.gen.c
+examples/TensorFlow/Keras/Keras_to_MDF_IRIS/keras_to_MDF
+examples/TensorFlow/Keras/keras_to_MDF
+/examples/TensorFlow/Keras/Keras_MDF/Keras_to_MDF_IRIS/keras_to_MDF
+/examples/TensorFlow/Keras/Keras_MDF/keras_to_MDF
 /examples/NeuroML/PyNN/HH.yaml
 /examples/NeuroML/PyNN/InputWeights.yaml
 /examples/NeuroML/PyNN/Net1.yaml
@@ -298,3 +303,7 @@ Thumbs.db #thumbnail cache on Windows
 /examples/NeuroML/PyNN/SimSimpleNet.yaml
 /examples/NeuroML/PyNN/SimpleNet.yaml
 /examples/NeuroML/PyNN/SimInputWeights.yaml
+/examples/TensorFlow/Keras/IRIS/keras_to_MDF
+/examples/TensorFlow/Keras/MNIST/keras_to_MDF
+/examples/TensorFlow/Keras/MNIST/keras_to_MDF.1
+/examples/TensorFlow/Keras/IRIS/keras_to_MDF.1

examples/TensorFlow/Keras/IRIS/README.md

@@ -0,0 +1,34 @@
+# Keras to MDF example: IRIS dataset
+
+**For more details on Keras and the current state of the Keras->MDF mapping see the [MNIST example](../MNIST).**
+
+This model uses the [IRIS dataset](https://en.wikipedia.org/wiki/Iris_flower_data_set) in the [trained Keras model](keras_model.py) and the MDF equivalent.
+
+### Summarize Model
+
+Below is the summary image of the trained Keras model. We can clearly see the output shape and number of weights in each layer:
+
+![summary](summary.png)
+
+
+### Keras Model
+
+Visualization of the model from Keras:
+
+<p align="center"><img src="model_on_iris_plot.png"/></p>
+<br>
+
+### MDF Model
+
+Graphviz is used to generate visualization for the MDF graph. Below is the visualization of the MDF graph after converting the keras model to MDF.
+
+![keras_to_MDF](keras_to_MDF.1.png)
+
+More detailed graphical representation of the MDF:
+
+<p align="center"><img src="keras_to_MDF.png" width="400"/></p>
+
+##### Netron
+Below is the visualization of this model using netron
+
+![keras-model-to-netron](layers_netron.png)

examples/TensorFlow/Keras/IRIS/keras_model.py

@@ -0,0 +1,91 @@
+import sys
+import tensorflow as tf
+
+# tf.__version__
+
+from keras.layers import Dense
+from keras.utils.vis_utils import plot_model
+from keras.models import Sequential
+
+# from keras_visualizer import visualizer
+from keras import layers
+
+import matplotlib.pyplot as plt
+import numpy as np
+
+import matplotlib.pyplot as plt
+import numpy as np
+
+from sklearn.datasets import load_iris
+from sklearn.model_selection import train_test_split
+from sklearn.preprocessing import StandardScaler
+
+# Controlling randomness, to facilitate testing/reproducibility
+# based on https://keras.io/examples/keras_recipes/reproducibility_recipes/
+from keras.utils import set_random_seed
+
+set_random_seed(1234)
+tf.config.experimental.enable_op_determinism()
+
+print("Loading data")
+iris = load_iris()
+
+X, y = iris.data, iris.target
+
+X_train, X_test, y_train, y_test = train_test_split(
+    X, y, test_size=0.2, random_state=42
+)
+
+# standardize the data
+scaler = StandardScaler()
+X_train = scaler.fit_transform(X_train)
+X_test = scaler.fit_transform(X_test)
+
+
+# Build Model
+print("Building the Model")
+model = tf.keras.models.Sequential()
+model.add(
+    tf.keras.layers.Dense(32, activation=tf.nn.relu, input_dim=4, name="first_layer")
+)  # hidden layers
+model.add(tf.keras.layers.Dense(64, activation=tf.nn.relu, name="second_layer"))
+model.add(tf.keras.layers.Dense(64, activation=tf.nn.relu, name="third_layer"))
+model.add(
+    tf.keras.layers.Dense(3, activation=tf.nn.softmax, name="fourth_layer")
+)  # output layer
+
+# Compile Model
+print("Compiling the model")
+model.compile(
+    optimizer="adam", loss="sparse_categorical_crossentropy", metrics=["accuracy"]
+)
+
+# train the model
+print("Training the Model")
+model.fit(X_train, y_train, epochs=3)
+
+
+# check if the model actually generalize the data.
+print("Accuracy of our model is:")
+val_loss, val_acc = model.evaluate(X_test, y_test)
+print(val_loss, val_acc)
+
+
+# Print summary
+model.summary()
+
+# plot the model
+print("Plotting the model")
+plot_model(
+    model, to_file="model_on_iris_plot.png", show_shapes=True, show_layer_names=True
+)
+
+
+# Saving model in h5 for Ploting Nuetron visual
+model.save("keras_model_on_iris.h5")
+
+
+# predict example for index 0
+print("Predict value at index 0:")
+predictions = model.predict([X_test])
+print("The predicted number at index 0 is", np.argmax(predictions[0]))