def run_network(sklearn,
NN,
lmbd,
eta,
act_h,
act_o,
hn,
epochs,
tol,
batch_size,
n_categories,
color_iter=0,
length=1):
costs = None
if sklearn:
if NN: # NN Regress using sklearn - Somewhat reasonable, needs tuning.
mlpr = MLPRegressor(hidden_layer_sizes=hn,
activation=act_h,
solver="adam",
alpha=lmbd,
learning_rate_init=eta)
mlpr.fit(XTrain, yTrain)
yTrue, yPred = yTest, mlpr.predict(XTest)
ypred = mlpr.predict(X)
R2 = mlpr.score(XTest, yTrue.ravel())
MSE = 1.0 #Not implemented
else: # Linear regression using Sklearn - Initial results prefer this
mlpr = LinearRegression()
mlpr.fit(XTrain, yTrain)
yTrue, yPred = yTest, mlpr.predict(XTest)
R2 = mlpr.score(XTest, yTrue.ravel())
MSE = 1.0
ypred = mlpr.predict(X)
else:
if NN: # My regression NN - Outputs 50 :)
mlpr = NeuralNetwork(XTrain,
yTrain.reshape(-1, 1),
XTest,
yTest.reshape(-1, 1),
eta=eta,
lmbd=lmbd,
n_hidden_neurons=hn,
act_h=act_h,
act_o=act_o,
epochs=epochs,
tol=tol,
batch_size=batch_size,
nn_type="regression",
cost="mse",
n_categories=n_categories,
length=length)
costs, scores = mlpr.train()
# Find full data prediction
#mlpr.plot_costs(color_iter) # Plot cost
mlpr.plot_scores(color_iter) # Plot scores
MSE = scores[-1, 1, 0]
R2 = scores[-1, 1, 1]
# full prediction
ypred = mlpr.predict_a_o(X)
else: # Run my regression
raise NotImplementedError("Insert lasso n stuff")
return MSE, R2, ypred, costs, mlpr
