john-adeojo

import seaborn as sns 
import matplotlib 
import matplotlib.pyplot as plt

x = Train.SalePrice
sns.set_style('whitegrid')
sns.distplot(x)
plt.show()

Train['SalePrice_log'] = np.log(Train.SalePrice)

john-adeojo

# Lets explore the correlations in our data set 
plt.figure(figsize=(20,20))
sns.heatmap(Train.corr())

john-adeojo

# Model 1: Ranbdom Forest Rgressor
import sklearn
from sklearn.model_selection import GridSearchCV
from sklearn.ensemble import RandomForestRegressor

# Split training data into features (x_train) and labels (Y_train)
x_train = Train_encoded.drop(columns=['Id','SalePrice','SalePrice_log'])
Y_train = Train_encoded.SalePrice_log

john-adeojo

# Build sequential neural network

model = keras.models.Sequential()  # Keras sequential model is composed of a single stack of layers connected sequentially
model.add(keras.layers.Input(shape=(784)))  # Input layer is a 1D array (no parameters, first layer so specify input shape, not including batch size just size of instances)
model.add(keras.layers.Dense(300, activation="relu"))  # Dense hidden layer with 300 neurons and ReLU activation fucntion ( each dense layer manages it's own weight matrix and bias vector)
model.add(keras.layers.Dense(100, activation="relu"))  # Second Dense hidden layer of 100 neurons
model.add(keras.layers.Dense(10, activation="softmax"))  # Final layer is a dense output layer of 10 neurons with softmax activation function (there are 10 exxclusive classes)

# Displays all the model's layers 
model.summary()

john-adeojo

# Compile the model 1: sequential NN

model.compile(loss="sparse_categorical_crossentropy",  # We have sparse labels: each instance there is a target class from 0 to 9 and classes are exclusive
             optimizer="sgd",  # model trained useing stochastic gradient descent 
             metrics=["accuracy"])  

john-adeojo

# Compile the model 1: sequential NN

model.compile(loss="sparse_categorical_crossentropy",  # We have sparse labels: each instance there is a target class from 0 to 9 and classes are exclusive
             optimizer="sgd",  # model trained useing stochastic gradient descent 
             metrics=["accuracy"])  

john-adeojo

# Fit  model1 : sequential NN
history = model.fit(X_train, y_train, epochs=30,  # Set the number of epochs else Keras defaults to 1 epoch
                   validation_data=(X_valid, y_valid))  # Passing the validation sets enables Keras to measure the loss and accuracy metrics on the validation set

john-adeojo

# Plot the training history for the network
def plot_performance (history, save):
    pd.DataFrame(history.history).plot(figsize=(8, 5))
    plt.grid(True)
    plt.gca().set_ylim(0, 1)  # Sets the verticle axis range from 0 to 1
    plt.show()
    plt.savefig(save)
    
plot_performance (history, 'model1')

john-adeojo

# Build a neural network with deep and wide architecture

input_ = keras.layers.Input(shape=(784))  # Input layer 
hidden1 = keras.layers.Dense(300, activation="relu")(input_)  # Create first hidden layer and feed input layer to it
hidden2 = keras.layers.Dense(100, activation="relu")(hidden1)  # Create 2nd hidden layer and feed output of first hidden layer 
concat = keras.layers.Concatenate()([input_, hidden2])  # Concatenate output from 2nd hidden layer and input layer
output = keras.layers.Dense(10, activation="softmax")(concat)  # Feed concatenation of first and second input layer to output layer creating wide and deep network
model2 = keras.Model(inputs=[input_], outputs=[output])

john-adeojo

# THIS FUNCTION TAKES A CSV FILE FROM A GITHUB URL AND READS IT INTO A PANDAS DATA FRAME
import pandas as pd

def read_file(url):

    """
    Takes GitHub url as an argument,
    pulls CSV file located @ github URL.

    """