Machine Learning & Deep Learning practice code.3

• get the dataset

boston = pd.read_csv('boston_house.csv')
boston

• check the first rows

boston.head()

• check the statistic

boston.describe()

• see the values of selected columns

boston['MEDV']

• check the column's distribution using histplot()

sns.histplot(data=boston, x='MEDV')

• visualize the graph using plot(scatter)
  kind='scatter', x='RM', y='MEDV'

boston.plot(kind='scatter', x='RM', y='MEDV')

• print four columns' values

boston[['MEDV', 'RM', 'AGE', 'CHAS']]

• identify relationships between columns
  plt.figure(figsize=(10,10))

sns.pairplot(boston[['MEDV', 'RM', 'AGE', 'CHAS']])
plt.show()

• see the rows correlation

boston_corr = boston[['MEDV', 'RM', 'AGE', 'CHAs']].corr()
boston_corr

• check the correlations between columns using seaborn's heatmap
  annot=True, cmap='Blues'
  plt.figure(figsize=(10,10)), sns.set(font_scale=0.6)

plt.figure(figsize=(10,10))
sns.set(font_scale=0.6)
sns.heatmap(boston_corr, annot=True, cmap='Blues')
plt.show()

• finding the most relatable column to MEDV using corr(). except last value, absolute value, ascending sort

boston_corr['MEDV'][:-1].abs().sort_values(ascending=False)

• separate data X and answers y

X = boston.drop('MEDV', axis=1)
X

y = bosotn['MEDV']
y

• change the form of DataFrame, Series to numpy array

X = X.values
y = y.values

print(X[:2])
print(y[:2])

• scaling with MinMaxScaler

from sklearn.preprocessing import MinMaxScaler

• 1. define the function mmx

2. label encoding X with fit_transform function and re-save it to X
3. check the result

mmx = MinMaxScaler()
X = mmx.fit_transform(X)
print(X)

X.shape

• modeling Decision Tree

dt = DecisionTreeRegressor()
dt.fit(X, y)
dt.score(X, y)

• modeling RandomForest

from sklearn.ensemble import RandomForestRegressor

rf = RandomForestRegressor()
rf.fit(X, y)
rf.score(X, y)

• print sample data

print(X[149])
print(y[149])

• insert the sample data into the model and predict using rf.predict

pred = rf.predict([X[149]])
print(pred)

• get libraries for deep learning

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense

• make a Sequential model --> and save to model variable
  input layer: (13, )
  hidden layer: 6 unit, activation='relu'
  output layer: 1 unit

model = Sequential()
model.add(Dense(6, activation='relu', input_shape=(13,)))
model.add(Dense(1))

• model compile
  loss='mse'
  optimizer='adam'
  metrics=['mse','mae']

model.compile(loss='mse', optimizer='adam', metrics=['mse', 'mae'])

• model learning: fit
  X, y, epochs=10, batch_size=8
  save the learned result: history

history = model.fit(X, y, epochs=10, batch_size=8)

-improve the accuracy. increase epochs -> 50

history = model.fit(X, y, epochs=50, batch_size=8)

• evaluate the performance of the deep learning

plt.plot(history.history['loss'], 'r')
plt.plot(history.history['mse'], 'b')
plt.title('Loss and Accuracy')
plt.xlabel("Epochs")
plt.ylabel("Loss")
plt.legent(["Loss", "MSE"])
plt.show()