print 'batch_size_train', batch_size_train
print 'batch_size_test', batch_size_test
print 'lambda_r:', lambda_r
print 'vali_test', vali_test
print 'epoch', epoch
print
'''*************************main****************************'''
'''*************************main****************************'''
for i in range(1):
# dataset list
dataset_list = ['', '_Women', '_Men', '_CLothes', '_Shoes', '_Jewelry']
# load data
[train_data, train_data_aux, validate_data, test_data, P,
Q] = readdata(dataset_list[dataset])
# load feature
[E, F] = get_feature(dataset)
'''for i in range(len(E)):
E[i] = E[i] / math.sqrt(mat(E[i]) * mat(E[i]).T)
for i in range(len(F)):
F[i] = F[i] / math.sqrt(mat(F[i]) * mat(F[i]).T)'''
E = E[:, 0:K1]
F = F[:, 0:K2]
Fc = get_feature_cnn(dataset_list[dataset])
Kc = len(Fc[0])
# select validate or test dataset
if vali_test == 0:
Test = validate_data
#This assumes that the input data is of length 4
#my packages
import readdata
#other packages
import numpy as np
import tensorflow as tf
#Get Data
trainx, trainy = readdata.readdata(999)
testx, testy = readdata.readdata(100)
numtrain = len(trainx)
numtest = len(testx)
#hyperparameters
learning_rate = 0.5
epochs = 30
#placeholders
X = tf.placeholder("float32")
Y = tf.placeholder("float32")
#model weights
#tf.variables are trainable by default
W = tf.Variable(np.array(np.random.rand(1, 4), dtype='float32'), name="weight")
#pred = tf.reduce_sum(tf.multiply(X, W))
pred = tf.matmul(W, X)
import numpy as np
# importing the database
import readdata
df = readdata.readdata('database.sqlite')
print(df.head())
#Shuffle the rows of df so we get a distributed sample when we display top few rows
df = df.reindex(np.random.permutation(df.index))
# Defining the X anf y parameters
X = df.iloc[:,1:] # 'other than overall rating, i.e. column 1'
y = df.iloc[:,0] # Need to predict 'Overall rating'
# Identifying the unique values for preferred_foot colomns
print("Unique values in preferred_foot feature are: ", end=" ")
print(df.preferred_foot.unique())
print("Lets apply one Hot encoding to preferred_foot feature")
# Applying One hot encoding on only catagorical columns, i.e. preferred_foot
from sklearn.preprocessing import LabelEncoder
label_encoder = LabelEncoder()
X.preferred_foot = label_encoder.fit_transform(X.preferred_foot)
# Splitting the train and test data
from sklearn.model_selection import train_test_split
X_train,X_test,y_train,y_test = train_test_split(X,y,test_size=0.10,random_state=0)
# Find the most significant features using backward elemination
print('-'*50)
import backwardEle
#my packages
import readdata
#other packages
import numpy as np
import tensorflow as tf
import pickle
#######change lbl to a list of vectors, not integers###########
#Get Data
trainimg,trainlbl=readdata.readdata('train')
testimg,testlbl=readdata.readdata('test')
xshape,yshape,dummy = np.shape(trainimg[0])
numtrain = len(trainimg)
numtest = len(testimg)
#hyperparameters
learning_rate = 0.0001#3e-3 #tf documentation says default = 0.001
beta1=0.94 #0.9
beta2=0.999 #0.999
epsilon=1e-8 #1e-8
dropout_rate = 0.80