def create_bike(x, y, bias=None, kernels=None, epochs=5): # Start my custom stuff d = Dense(x, 1) d2 = Dense(d, 1) d3 = Dense(d2, 1) # Assign Kernel matrix to weights if kernels is not None: print 'k' d.W = kernels[0] #d2.W = kernels[1] d.b = bias[0] #d2.b = bias[1] relu = ReLu() sess = Session() sess.add_node(d) sess.add_node(ReLu()) ''' sess.add_node(d2) sess.add_node(ReLu()) sess.add_node(d3) sess.add_node(relu) ''' sess.add_node(subtr(d3, y)) sess.add_node(MeanSqrError(d3)) print '\n' return sess
def run_bike(x, y, kernels=None, epochs=5): # Start my custom stuff d = Dense(x, 3) d2 = Dense(d, 1) # Assign Kernel matrix to weights if kernels: d.W = kernels[0] weg = d.W relu = ReLu() sq = Square() sess = Session() sess.add_node(d) sess.add_node(d2) sess.add_node(relu) print '\n' losses = [] for i in range(epochs): ''' x1 = d2.forward(x1) print 'shape:%s\n'%str(x1.shape),x1 x1 =relu.forward(x1) print 'shape:%s\n'%str(x1.shape),x1 ''' #d.W = kernels[i] print "Weight matr " #print str(d.W) err = sess.step(x, y) #print "err na\n", err loss = np.mean(np.square(err)) print 'epoch ', i, "LOSS:", loss losses.append(loss) print hp.plot(d.W.flatten()) return losses