def main(path):
train,test,wordVocab = loaddata(path)
maxSentLen = 140
train_sent = pad_sequences(train["sent"],maxlen=maxSentLen,padding="post")
train_gen = pad_sequences(train["gen"],maxlen=5,padding="post")
train_disease = pad_sequences(train["disease"],maxlen=5,padding="post")
test_sent = pad_sequences(test["sent"],maxlen=maxSentLen,padding="post")
test_gen = pad_sequences(test["gen"],maxlen=5,padding="post")
test_disease = pad_sequences(test["disease"],maxlen=5,padding="post")
train_rel = np.array(train["rel"])
test_rel = np.array(test["rel"])
print(train_rel.shape)
print(test_rel.shape)
# load pre-embedding
print("loading embedding...")
embedding_size = 200
embeddingVocab_size = len(wordVocab)
# w2v_dir_path = "/media/network/watching_dog/embedding/bio_nlp_vec/PubMed-shuffle-win-30.bin"
w2v_dir_path = "/media/kazane/watching_dog/embedding/bio_nlp_vec/PubMed-shuffle-win-30.bin"
word2vec = KeyedVectors.load_word2vec_format(w2v_dir_path, binary=True, unicode_errors='ignore')
print("build embedding weights...")
embedding_weights = np.zeros((embeddingVocab_size + 1, embedding_size))
unknow_words = []
know_words = []
for word, index in wordVocab.items():
try:
embedding_weights[index, :] = word2vec[word.lower()]
know_words.append(word)
except KeyError as E:
# print(E)
unknow_words.append(word)
embedding_weights[index, :] = np.random.uniform(-0.025, 0.025, embedding_size)
print("unknow_per: ", len(unknow_words) / embeddingVocab_size, " unkownwords: ", len(unknow_words), " vocab_size: ",
embeddingVocab_size)
model = myModel(sent_lenth=maxSentLen,word_embedding=embedding_weights)
model.attn(embedding=embedding_weights)
model.train(inputs=[train_sent,train_gen,train_disease,],
label=[train_rel,],
save_path="./outputs/checkpoint",
validation_split=0.1,
batch_size=128,
epochs=5,
)
y_ = model.predict([test_sent,test_gen,test_disease])
# np.save(file=path+"/"+"outputs.npy",arr=y_)
is_gate = False
if is_gate is False:
y_ = get_Result(y_,)
get_F1Value(y_=y_,y=test_rel,path=path)
else:
get_ROC(y_,test_rel,path=path,gate=100)
def testmodel(filepath):
"""
according to the image feature, predict the img label
"""
print("test model....")
images, tags = loaddata(filepath)
feature_info = extractfeature(images, tags)
x = [info[0] for info in feature_info]
y = [info[1] for info in feature_info]
clf = joblib.load('train_model.m')
samples_proba = clf.predict_proba(x) # predict the test images probability
top5_index = np.argsort(-samples_proba, axis=1)[:, :5].tolist()
res = []
for (i, tag) in enumerate(y):
res.append(tag in top5_index[i])
return res
def testmodel(path):
def getFeatVec(features, clf):
featVec = np.zeros((1, 1000))
res = clf.predict(features)
for i in res:
featVec[0][i] += 1
return featVec
def result(predict_y, test_y, num):
print("###############################")
right = 0
res = []
for i, tag in enumerate(test_y):
if tag in predict_y[i][21 - num:21]:
right += 1
res.append(True)
else:
res.append(False)
print("%d/%d = %f" % (right, len(test_y), right * 1.0 / len(test_y)))
print("###############################")
return res
print('load model')
svm = joblib.load("./05svm.model")
clf = joblib.load("./05vocab.pkl")
data, tags = loaddata(path)
features, tags = extractfeature(data, tags)
print('predict')
test_x = np.float32([]).reshape(0, 1000)
for feature in features:
featVec = getFeatVec(feature, clf)
test_x = np.append(test_x, featVec, axis=0)
p = svm.predict_proba(test_x)
p = p.argsort()
res = result(p, tags, 5)
return res
def testmodel(path):
print("[test] start")
data, tags = loaddata(path)
feats, types = extractfeature(data, tags)
svm = joblib.load("svm.pkl")
pca = joblib.load("pca.pkl")
normalizer = joblib.load("normalizer.pkl")
feats = pca.transform(feats)
feats = normalizer.transform(feats)
probability = svm.predict_proba(feats)
args = probability.argsort(axis=1)
re = []
for i in range(args.shape[0]):
if int(types[i]) in args[i][-5:]:
re.append("True")
else:
re.append("False")
print("[test] end")
return re
import pandas as pd
import numpy as np
from load import loaddata
from scipy import stats
from sklearn.preprocessing import Imputer
from sklearn.model_selection import KFold
from sklearn.svm import SVR
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import r2_score
from sklearn.metrics import mean_squared_error
data = loaddata('train.csv', 0.05)
X = data.loc[:, 'Feature_1':'Ret_120'].values
y = data.loc[:, 'Ret_121':'Ret_PlusTwo'].values
X_feature = data.loc[:, 'Feature_1':'Feature_25'].values
X_mins = data.loc[:, 'Ret_2':'Ret_120'].values
X_ret = data.loc[:, 'Ret_MinusTwo':'Ret_MinusOne'].values
daily_weights = data['Weight_Daily'].values
intraday_weights = data['Weight_Intraday'].values
#print(X.shape, y.shape)
imp = Imputer(missing_values='NaN', strategy="mean", axis=0)
X_imp = imp.fit_transform(X)
X_featureimp = imp.fit_transform(X_feature)
X_minsimp = imp.fit_transform(X_mins)
X_retimp = imp.fit_transform(X_ret)
import pandas as pd
import numpy as np
import os
import matplotlib.pyplot as plt
import random
import tensorflow as tf
from sklearn.preprocessing import Imputer
from sklearn.model_selection import KFold
from sklearn.decomposition import PCA
from load import loaddata
#-------------------------------------------------------------------------------
# read data
#-------------------------------------------------------------------------------
data = loaddata('train.csv', 1.0)
features = data.loc[:, 'Feature_1':'Feature_25'].values
rets = data.loc[:, 'Ret_MinusTwo':'Ret_PlusTwo'].values
daily_weights = data['Weight_Daily'].values
intraday_weights = data['Weight_Intraday'].values
test_data = loaddata('test_2.csv', 3.0)
test_features = test_data.loc[:, 'Feature_1':'Feature_25'].values
test_rets = test_data.loc[:, 'Ret_MinusTwo':'Ret_120'].values
print("test rets shape ", test_rets.shape)
#-------------------------------------------------------------------------------
# preprocess
#-------------------------------------------------------------------------------
imp_axis0 = Imputer(missing_values='NaN', strategy="mean", axis=0)
imp_axis1 = Imputer(missing_values='NaN', strategy="mean", axis=1)
def main():
# only read raw data if so required (cleaned files do not exist yet)
if READ_RAW_DATA:
# train set
dataset = '../../data/training_set_VU_DM.csv'
# test set (turn off relevance score in this case!)
#dataset = '../../data/test_set_VU_DM.csv'
# take the first 1000 lines of the dataset only - use this for testing
# to make the code less slow! Comment it out for finalizing
# dataset = '../data/testfile.csv'
# loading in the right file
data = load.loaddata(dataset)
# # create competitor features
data = features.create_competitor_features(data)
# # create other features
data = features.other_features(data)
# # add relevance grades
data = features.relevance_score(data)
# remove outliers
data = eda.remove_outliers(data)
# # handling missing values
data = eda.missing_values(data)
if PLOT:
# take a sample of the data to make plotting feasible
sample_data = data.sample(n=500000)
# plot distributions
eda.plot_distributions(sample_data)
# plot correlations between sets of variables
eda.plot_correlations(sample_data)
# plot impact of price of competitor on booking
eda.plot_competitor_price_impact(sample_data)
# get correlations of the features
correlations.show_correlations(sample_data)
# divide data into train and test set (and save these)
train_data, test_data = process.split_train_test(data)
# downsample data to create class balance (and save)
downsampled_train_data = process.downsample(train_data)
# upsample data to create class balance (and save it)
upsampled_train_data = process.upsample(train_data)
# data is already loaded - only need to load it from file
# test for the best set of hyperparameters
if HYPERPARAM:
# get the appropriate training set
if SAMPLING_METHOD == "downsample":
traindataset = '../data/downsampled_crossvalidation_set.csv'
elif SAMPLING_METHOD == "upsample":
traindataset = "../data/upsampled_crossvalidation_set.csv"
elif SAMPLING_METHOD == "none":
traindataset = "../data/full_crossvalidation_set.csv"
# loading in the data
train_data = load.loaddata(traindataset)
# remove columns not in test dataset
keep_cols = [col for col in train_data.columns if col not in ['booking_bool', 'click_bool']]
# sample a smaller subset to make this all feasible
train_data = train_data[keep_cols].sample(n=4000)
print(train_data.columns)
# Train lambdamart for different hyperparam values and evaluate on validation set
trees = [5, 10, 50, 100, 150, 300, 400]
lrs = [0.15, 0.10, 0.8, 0.05, 0.01]
indices = []
for i in range(np.array(train_data.shape[0])):
items = [0, 1]
indices.append(items)
indices = np.array(indices)
# K-fold cross validation for different parameter combinations
for tree in trees:
for lr in lrs:
# indices = np.array(train_data.shape[0])
kf = KFold(n_splits = 5)
ndcgs = []
for train_index, test_index in kf.split(indices):
train_index = train_index.tolist()
test_index = test_index.tolist()
# Split up data
X_train, X_validation = train_data.iloc[train_index], train_data.iloc[test_index]
# Run lambdamart on training data and evaluate on validation data
ndcg = models.lambdamart(X_train, X_validation, tree, lr, SAMPLING_METHOD)
print(ndcg)
ndcgs.append(ndcg)
average_ndcg = np.mean(ndcgs)
# Save NDCG
file = '../results/hyperparams/crossvalidation_' + SAMPLING_METHOD + '.txt'
with open(file, 'a') as f:
line = 'trees: ' + str(tree) + ', lr: ' + str(lr) + ', average_ndcg: ' + str(average_ndcg) + '\n'
print(line)
f.write(line)
f.close()
# run the full model
if LAMBDAMART:
# test data is always the same
testdataset = '../data/testing_set_only.csv'
# get the appropriate training set
if SAMPLING_METHOD == "downsample":
traindataset = '../data/downsampled_training_set_only.csv'
elif SAMPLING_METHOD == "upsample":
traindataset = "../data/upsampled_training_set_only.csv"
elif SAMPLING_METHOD == "none":
traindataset = "../data/full_training_set_only.csv"
# loading in the data
train_data = load.loaddata(traindataset)
# loading in final test set
test_data = load.loaddata(testdataset)
# hyperparameters
trees = 2
lrs = 0.10
# train lambdamart and evaluate on test set
ndcg = models.lambdamart(train_data, test_data, trees, lrs, SAMPLING_METHOD)
print(ndcg)
from load import loaddata
from feature import extractfeature
from classify import trainmodel
from predict import testmodel
from datetime import datetime
start = datetime.now()
images, tags = loaddata("../data/images/train")
feature_info = extractfeature(images, tags)
trainmodel(feature_info)
result = testmodel("../data/images/test")
print("The accuracy is {}.".format(sum(result) / len(result)))
print("The whole proess cost {}.".format(datetime.now() - start))
from load import loaddata
from feature import extractfeature
from classify import trainmodel
from predict import testmodel
data, tags = loaddata("/home/acytoo/train_dataset/")
features = extractfeature(data, tags)
trainmodel(features)
res = testmodel("/home/acytoo/test_dataset/")
print "precise: ", res.count("True")*1.0/len(res)
from load import loaddata
from feature import extractfeature
from classify import trainmodel
from predict import testmodel
if __name__ == '__main__':
data,tags = loaddata('./train')
features = extractfeature(data,tags)
trainmodel(features)
res = testmodel('./test')
dtest = xgb.DMatrix(X_test, label=y_test)
# hyperparameters
param = {'max_depth': 5, 'eta': 0.01, 'objective': 'rank:ndcg'}
num_round = 20
# specify model
model = xgb.train(param, dtrain, num_round)
# predict
preds = model.predict(dtest)
return preds, y_test
if __name__ == '__main__':
traindataset = '../data/downsampled_training_set.csv'
train_data = load.loaddata(traindataset)
train_data = train_data.drop('date_time', axis=1)
train_data = features.relevance_score(train_data)
testdataset = '../data/test_subset.csv'
test_data = load.loaddata(testdataset)
test_data = test_data.drop('date_time', axis=1)
test_data = features.relevance_score(test_data)
y_pred, y_test = lambda_mart(train_data, test_data)
print(len(y_test))
print(len(y_pred))
#print('Accuracy: {0:.4f}'.format(accuracy_score(y_test, y_pred)))
from gensim import corpora, models, similarities
from load import loaddata
from nltk.corpus import brown
from nltk.corpus import stopwords
import jieba
import logging
import nltk
import load
#Query
target = '房貸要如何申請?最近的利率優惠如何?'
#1. get jieba
jieba_target = list(jieba.cut(target,cut_all=False))
jieba_target_stopword=[]
for i in range(0,len(jieba_target)):
if jieba_target[i] not in loaddata().chinese_stopwords:
jieba_target_stopword.append(jieba_target[i])
#2. get vec
vec_target = dictionary.doc2bow(jieba_target_stopword)
#3. get lda score
score_target = lsi[vec_target]
#sims是LSI的結果
index = similarities.MatrixSimilarity(lsi[bank_questions_corpus])
sims = index[score_target]
###建立權重分數
#建立權重陣列
tier1=[]
tier2=[]
