In [1]:
import pandas as pd
import numpy as np
np.random.seed(2017) #important to set the seed before importing keras
from keras.models import Sequential
from keras.layers import Dense
from keras.utils import to_categorical
from keras.callbacks import EarlyStopping
from keras.optimizers import SGD
import matplotlib
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
import seaborn as sns
In [2]:
%matplotlib inline
In [3]:
sns.set()
sns.set_style('ticks')
In [4]:
#load the dataset
df = pd.read_csv("HR_comma_sep.csv")
In [5]:
#preview
df.head()
Out[5]:
In [6]:
df.sales.value_counts()
Out[6]:
In [7]:
df.rename(columns={'sales':'department'},inplace=True)
In [8]:
#let's convert the two categorical variables 'department' and 'salary' into dummy-variables for modelling
In [9]:
df = pd.get_dummies(df,columns=['department','salary'])
In [10]:
x,y = df.drop('left',axis=1).values, df.left.values
In [11]:
# let's do a training-test split for validation later on
x_train,x_test,y_train,y_test = train_test_split(x,y,test_size=0.3, random_state=2017)
In [12]:
# let's convert our output variable into categorical format for keras
num_classes = np.max(y_train)+1
y_train = to_categorical(y_train,num_classes)
y_test = to_categorical(y_test,num_classes)
In [13]:
x_train.shape
Out[13]:
In [14]:
n_cols = x_train.shape[1]
In [15]:
#set-up early-stopping monitor
early_stopping_monitor = EarlyStopping(patience=5)
Baseline Model¶
In [16]:
# configure our neural-net
np.random.seed(2017) #important to set the seed for reproducibility
model = Sequential()
model.add(Dense(50,activation='relu',input_shape=(n_cols,)))
model.add(Dense(2,activation='softmax'))
#compile
model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])
#fit
history = model.fit(x_train,y_train,
epochs=100,verbose=1,validation_split=0.2,
callbacks=[early_stopping_monitor],shuffle=False)
model.summary()
#plot training and validation los
plt.plot(history.history['loss'],'r',label='training')
plt.plot(history.history['val_loss'],'b',label='validation')
plt.xlabel('epochs')
plt.ylabel('loss')
plt.legend()
#plot training and validation accuracy
plt.figure(figsize=(8,7))
plt.plot(history.history['acc'],'r',label='training')
plt.plot(history.history['val_acc'],'b',label='validation')
plt.xlabel('epochs')
plt.ylabel('accuracy')
plt.legend()
# so we see that our baseline model has quite good accuracy just after 20 epochs
#let's evaluate our model to predict on our hold-out data
model.evaluate(x_test,y_test)
Out[16]:
In [17]:
# Let's increase number of nodes in the hidden layer and repeat
Model-2 : Increasing number of nodes in hidden layer¶
In [18]:
# configure our neural-net
np.random.seed(2017) #important to set the seed for reproducibility
model = Sequential()
model.add(Dense(100,activation='relu',input_shape=(n_cols,))) #changed number of hidden layers from 50 to 100
model.add(Dense(2,activation='softmax'))
#compile
model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])
#fit
history = model.fit(x_train,y_train,epochs=100,verbose=1,
validation_split=0.2,callbacks=[early_stopping_monitor],
shuffle=False)
model.summary()
#plot training and validation los
plt.plot(history.history['loss'],'r',label='training')
plt.plot(history.history['val_loss'],'b',label='validation')
plt.xlabel('epochs')
plt.ylabel('loss')
plt.legend()
#plot training and validation accuracy
plt.figure(figsize=(8,7))
plt.plot(history.history['acc'],'r',label='training')
plt.plot(history.history['val_acc'],'b',label='validation')
plt.xlabel('epochs')
plt.ylabel('accuracy')
plt.legend()
# so we see that our baseline model has quite good accuracy just after 20 epochs
#let's evaluate our model to predict on our hold-out data
print "Model Evaluation on test dataset [loss,accuracy] ",model.evaluate(x_test,y_test)
In [19]:
# we see that increasing the number of nodes in the hidden layer didn't help, let's move on to model-3
Model 3: Adding more hidden layers¶
In [20]:
# configure our neural-net
np.random.seed(2018) #important to set the seed for reproducibility
model = Sequential()
model.add(Dense(50,activation='relu',input_shape=(n_cols,)))
model.add(Dense(50,activation='relu'))
model.add(Dense(2,activation='softmax'))
#compile
model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])
#fit
history = model.fit(x_train,y_train,epochs=20,verbose=1,
validation_split=0.2,callbacks=[early_stopping_monitor],
shuffle=False)
model.summary()
#plot training and validation los
plt.plot(history.history['loss'],'r',label='training')
plt.plot(history.history['val_loss'],'b',label='validation')
plt.xlabel('epochs')
plt.ylabel('loss')
plt.legend()
#plot training and validation accuracy
plt.figure(figsize=(8,7))
plt.plot(history.history['acc'],'r',label='training')
plt.plot(history.history['val_acc'],'b',label='validation')
plt.xlabel('epochs')
plt.ylabel('accuracy')
plt.legend()
# so we see that our baseline model has quite good accuracy just after 20 epochs
#let's evaluate our model to predict on our hold-out data
print "Model Evaluation on test dataset [loss,accuracy] ",
model.evaluate(x_test,y_test)
Out[20]:
In [21]:
#increasing the number of layers certainly helped which leads us to our next model where make the model deeper
Model 4 : Making it more deep model¶
In [22]:
# configure our neural-net
np.random.seed(2019)
model = Sequential()
model.add(Dense(50,activation='relu',input_shape=(n_cols,)))
model.add(Dense(50,activation='relu'))
# model.add(Dense(50,activation='relu'))
model.add(Dense(2,activation='softmax'))
#compile
model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])
#fit
history = model.fit(x_train,y_train,epochs=50,
verbose=1,validation_split=0.2,
callbacks=[early_stopping_monitor],shuffle=False)
model.summary()
#plot training and validation los
plt.plot(history.history['loss'],'r',label='training')
plt.plot(history.history['val_loss'],'b',label='validation')
plt.xlabel('epochs')
plt.ylabel('loss')
plt.legend()
#plot training and validation accuracy
plt.figure(figsize=(8,7))
plt.plot(history.history['acc'],'r',label='training')
plt.plot(history.history['val_acc'],'b',label='validation')
plt.xlabel('epochs')
plt.ylabel('accuracy')
plt.legend()
# so we see that our baseline model has quite good accuracy just after 20 epochs
#let's evaluate our model to predict on our hold-out data
print "Model Evaluation on test dataset [loss,accuracy] ",
model.evaluate(x_test,y_test)
Out[22]:
