get_dot_h.py

from keras.models import Sequential
from keras.layers import Dense
import numpy as np
from keras.models import load_model
from utility import get_dataset_path, get_dataframe_split
from everywhereml.code_generators.tensorflow import convert_model

pathDSTrain = get_dataset_path("TrainSHAPSNNImpactfulFeatures")
pathDSTest = get_dataset_path("TestSHAPSNNImpactfulFeatures")

'''pathDSTrain = get_dataset_path("TrainBinarySHAPSNNImpactfulFeatures")
pathDSTest = get_dataset_path("TestBinarySHAPSNNImpactfulFeatures")'''

x_train , y_train = get_dataframe_split(pathDSTrain)
x_test , y_test = get_dataframe_split(pathDSTest)

#one hot encode labels to adapt with model converter multiclass
x_test = np.asarray(x_test)
y_test = np.asarray(y_test)
y_test = np.zeros((len(y_test), 3), dtype=int)
for i in range(len(y_test)):
    y_test[i, y_test[i]] = 1

#one hot encode labels to adapt with model converter binary
'''x_test = np.asarray(x_test)
y_test = np.asarray(y_test)
y_test = np.eye(2)[y_test].astype(int)'''

#avoid the custom layer problem snn
'''model = Sequential()
model.add(Dense(41, input_dim=x_train.shape[1], activation='relu'))
model.add(Dense(20, activation='relu'))
model.add(Dense(units=3, activation="softmax")) 
model.compile(loss="sparse_categorical_crossentropy", optimizer='adam', metrics=['accuracy'])
model.fit(x_train, y_train, epochs=5, batch_size=32, verbose=1)
model.save("TrainedModels\\SHAPSNNImportantFeatures.h5")'''

#convert a snn trained model in .h module
snn = load_model("TrainedModels\\SHAPSNNImportantFeatures.h5")
c_header = convert_model(snn, x_test, y_test, model_name='snn_model')
print(c_header)