

In [ ]:

    
%matplotlib inline



In [ ]:

    
from ailib.ml.datasets.regression_toy_problems import gen_1d_polynomial_samples

Basic 1D non-linear regression with Keras

TODO: see https://stackoverflow.com/questions/44998910/keras-model-to-fit-polynomial

Install Keras

https://keras.io/#installation

Install dependencies

Install TensorFlow backend: https://www.tensorflow.org/install/

pip install tensorflow

Insall h5py (required if you plan on saving Keras models to disk): http://docs.h5py.org/en/latest/build.html#wheels

pip install h5py

Install pydot (used by visualization utilities to plot model graphs): https://github.com/pydot/pydot#installation

pip install pydot

Install Keras

pip install keras

Import packages and check versions



In [ ]:

    
import tensorflow as tf
tf.__version__



In [ ]:

    
import keras
keras.__version__



In [ ]:

    
import h5py
h5py.__version__



In [ ]:

    
import pydot
pydot.__version__

Make the dataset



In [ ]:

    
df_train = gen_1d_polynomial_samples(n_samples=100, noise_std=0.05)

x_train = df_train.x.values
y_train = df_train.y.values

plt.plot(x_train, y_train, ".k");



In [ ]:

    
df_test = gen_1d_polynomial_samples(n_samples=100, noise_std=None)

x_test = df_test.x.values
y_test = df_test.y.values

plt.plot(x_test, y_test, ".k");

Make the regressor



In [ ]:

    
model = keras.models.Sequential()

#model.add(keras.layers.Dense(units=1000, activation='relu', input_dim=1))
#model.add(keras.layers.Dense(units=1))
#model.add(keras.layers.Dense(units=1000, activation='relu'))
#model.add(keras.layers.Dense(units=1))

model.add(keras.layers.Dense(units=5, activation='relu', input_dim=1))
model.add(keras.layers.Dense(units=1))
model.add(keras.layers.Dense(units=5, activation='relu'))
model.add(keras.layers.Dense(units=1))
model.add(keras.layers.Dense(units=5, activation='relu'))
model.add(keras.layers.Dense(units=1))

model.compile(loss='mse',
              optimizer='adam')

model.summary()



In [ ]:

    
hist = model.fit(x_train, y_train, batch_size=100, epochs=3000, verbose=None)



In [ ]:

    
plt.plot(hist.history['loss']);



In [ ]:

    
model.evaluate(x_test, y_test)



In [ ]:

    
y_predicted = model.predict(x_test)



In [ ]:

    
plt.plot(x_test, y_test, ".r")
plt.plot(x_test, y_predicted, ".k");