This post shows how to fix the random seed to get reproducible results with Keras.<\/p>\n\n\n\n
For an illustration, a simple DNN model is implemented in the Jupyter Notebook using a function for fixing the random seed. The function seed_everything()<\/strong> is borrowed from https:\/\/dacon.io\/codeshare\/2363.<\/p>\n\n\n\n Every time we run this code as a whole, the result is always 96.562645 and is not changed.<\/p>\n\n\n\n However, when we reuse and rerun some parts of the above code in the same file, we can encounter some unexpected results as follows.<\/p>\n\n\n\n# Import library and define functions\n \n# In[1]:\n_________________________________________________________________\nfrom keras.models import Sequential\nfrom keras.layers import Dense\n \nimport tensorflow as tf\nimport numpy as np\nimport random\nimport os\n \ndef seed_everything(seed: int = 42):\n random.seed(seed)\n np.random.seed(seed)\n os.environ[\"PYTHONHASHSEED\"] = str(seed)\n tf.random.set_seed(seed)\n_________________________________________________________________\n \n# Simple DNN model with a fixed random seed \n# In[2]:\n_________________________________________________________________\nx = np.array([[10, 20, 30], [20, 30, 40], [30, 40, 50], [40, 50, 60]])\ny = np.array([40, 50, 60, 70])\n \nseed_everything(1) # fix the random seed\n \nmodel = Sequential()\nmodel.add(Dense(50, activation='relu', input_dim=3))\nmodel.add(Dense(1))\nmodel.compile(optimizer='adam', loss='mse')\nmodel.fit(x, y, epochs=1000, verbose=0)\n \nx_input = np.array([60, 70, 80])\nx_input = x_input.reshape((1,3))\npred = model.predict(x_input, verbose=0)\nprint(pred)\n_________________________________________________________________\n \n[[96.562645]]<\/code><\/pre>\n\n\n\nRobustness checks<\/h3>\n\n\n\n
# rerun model fit\n# In[3]:\n_________________________________________________________________\nmodel.fit(x, y, epochs=1000, verbose=0)\nx_input = np.array([60, 70, 80])\nx_input = x_input.reshape((1,3))\npred = model.predict(x_input, verbose=0)\nprint(pred)\n_________________________________________________________________\n[[90.000145]]\n \n# rerun model.fit with a fixed random seed\n# In[4]:\n_________________________________________________________________\nseed_everything(1)\n \nmodel.fit(x, y, epochs=1000, verbose=0)\nx_input = np.array([60, 70, 80])\nx_input = x_input.reshape((1,3))\npred = model.predict(x_input, verbose=0)\nprint(pred)\n_________________________________________________________________\n[[90.00001]]\n \n \n# rerun model.compile & model.fit with a fixed random seed\n# In[5]:\n_________________________________________________________________\nseed_everything(1)\n \nmodel.compile(optimizer='adam', loss='mse')\nmodel.fit(x, y, epochs=1000, verbose=0)\nx_input = np.array([60, 70, 80])\nx_input = x_input.reshape((1,3))\npred = model.predict(x_input, verbose=0)\nprint(pred)\n_________________________________________________________________\n[[89.99999]]\n \n# call seed_everything(1) model = Sequential()\n# In[6]:\n_________________________________________________________________\nseed_everything(1)\n \nmodel = Sequential()\nmodel.add(Dense(50, activation='relu', input_dim=3))\nmodel.add(Dense(1))\nmodel.compile(optimizer='adam', loss='mse')\nmodel.fit(x, y, epochs=1000, verbose=0)\nx_input = np.array([60, 70, 80])\nx_input = x_input.reshape((1,3))\npred = model.predict(x_input, verbose=0)\nprint(pred)\n_________________________________________________________________\n[[96.562645]]<\/code><\/pre>\n\n\n\n