Regression test for checking

import seaborn as sns
import tensorflow as tf
from tensorflow import keras
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import tensorflow_docs as tfdocs
import tensorflow_docs.plots
import tensorflow_docs.modeling
from scipy.stats import gaussian_kde

################################################################################

def build_model(n):
# Define the prediction model. The NN takes 5 input features (magnitudes), and outputs
# the redshift.
  model = keras.Sequential([
    keras.layers.Dense(5, activation='relu', input_shape=(n,)), # 5 inputs (mags)
    keras.layers.Dense(4, activation='relu'), # do I need this layer?
    keras.layers.Dense(1) # 1 output (z)
    ])
  model.compile(loss='mae',
                optimizer = tf.keras.optimizers.RMSprop(0.001),
                metrics = ['mae', 'mse'])
  return model
################################################################################
# These are convenience functions for plotting the results of the model's predictions.
def plot_mae():
    plotter = tfdocs.plots.HistoryPlotter(smoothing_std=2)
    plotter.plot({'Basic': history}, metric = "mae")
    plt.title('Mean Absolute Error evolution for %s'%datasetname)
    plt.ylabel(r'$\Delta z$')
    plt.show()

def plot_mse():
    plotter = tfdocs.plots.HistoryPlotter(smoothing_std=2)
    plotter.plot({'Basic': history}, metric = "mse")
    plt.title('Mean Standard Error evolution for %s'%datasetname)
    plt.ylabel(r'$\Delta z$')
    plt.show()

def plot_z():
    x, y = valid_dataset['z'], valid_dataset['Predicted z']
    xy = np.vstack([x,y])
    z = gaussian_kde(xy)(xy)
    # Sort the points by density, so that the densest points are plotted last
    # idx = z.argsort()
    # x, y, z = x[idx], y[idx], z[idx]
    plt.scatter(valid_dataset['z'], valid_dataset['Predicted z'],
                s = 50,
                alpha = 0.5,
                marker = '.',
                # edgecolor = '',
                c = z)
    plt.colorbar()
    plt.title('Redshift predictions for %s\ncompared to spectroscopic redshift'%datasetname)
    plt.xlabel(r'$z_{spec}$')
    plt.ylabel('Predicted z')
    plt.show()

def plot_deltaz():
    valid_dataset['Delta z'] = valid_dataset['Predicted z'] - valid_dataset['z']
    plt.scatter(valid_dataset['z'], valid_dataset['Delta z'],
                s = 2, alpha = 0.5, marker = '.')
    plt.title('Deviation of redshift predictions for %s\nfrom spectroscopic redshift'%datasetname)
    plt.xlabel(r'$z_{spec}$')
    plt.ylabel(r'$\Delta z$')
    # plt.xlim([0, 1.5])
    plt.show()

def plot_z_boxplot(outliers = False):
    columns = [valid_dataset['Predicted z'], valid_dataset['z']]
    df = pd.DataFrame(data = valid_dataset,
                columns = columns)
    sns.boxplot(data = pd.melt(df),
        x = None, y = None, # This screws it up
        linewidth = 0.5,
        flierprops = dict(markerfacecolor = '0.1', markersize = 0.2),
        showfliers = outliers
        )
    # valid_dataset.boxplot(column = ['Predicted z', 'z'])
    plt.title('Distribution of statistical parameters\nfor %s'%datasetname)
    plt.tight_layout()
    plt.show()

def plot_delta_z_hist():
    stats = valid_dataset['Delta z'].describe()
    valid_dataset['Delta z'] = valid_dataset['Predicted z'] - valid_dataset['z']
    valid_dataset['Delta z'].hist(label = ' mean = %.3f\n std dev = %.3f'%
                                  (stats[1], stats[2]),
                                  bins = 100)
    plt.title(r'Distribution of $\Delta z$ for %s'%datasetname)
    plt.legend()
    plt.xlabel(r'$\Delta z$')
    plt.ylabel('Count')
    plt.show()
################################################################################
dataset = pd.read_csv('sdss12.csv')
datasetname = 'SDSS DR12 QSOs'
mags = ['umag', 'gmag', 'rmag', 'imag', 'zmag']
columns = mags + ['z']

num_features = len(columns)

dataset = dataset[columns]
train_dataset = dataset.sample(frac = 0.8, random_state = 1)
valid_dataset = dataset.drop(train_dataset.index)
train_labels = train_dataset.iloc[:, len(mags):].values # all but the last column
valid_labels = valid_dataset.iloc[:, len(mags):].values # all but the last column

print('Training set: \n',train_dataset)
print('Validation set: \n', valid_dataset)

model = build_model(num_features)
model.summary()

N = 50
early_stop = keras.callbacks.EarlyStopping(monitor='val_loss', patience=500)
history = model.fit(train_dataset.iloc[:,:num_features], train_labels, epochs = N,
                    validation_split = 0.2, verbose = 0,
                    callbacks = [
                        early_stop,
                        tfdocs.modeling.EpochDots()
                        ])
model.save("model2.h5")

# Model testing
valid_predictions = model.predict(valid_dataset.iloc[:,:num_features])
valid_dataset['Predicted z'] = valid_predictions
print("\nPredicted\n")
print(valid_dataset)

# Visualise the model's training progress using the stats in the history object
hist = pd.DataFrame(history.history)
hist['epoch'] = history.epoch
print(hist.tail)

print('Predicted z stats: \n', valid_dataset['Predicted z'].describe())
print('Spectroscopic z stats: \n', valid_dataset['z'].describe())

plot_mse()
plot_mae()
plot_z()
plot_deltaz()
plot_delta_z_hist()
plot_z_boxplot(True)
plot_z_boxplot(False)