astoeckl

import pandas as pd
import numpy as np

df = pd.read_csv("news_en.csv", sep=',',index_col = "id")
print(df.shape)
df.head()

astoeckl

import gensim

def read_corpus(fname, tokens_only=False):
    for i, line in enumerate(fname):
        if tokens_only:
            yield gensim.utils.simple_preprocess(line)
        else:
            yield gensim.models.doc2vec.TaggedDocument(gensim.utils.simple_preprocess(line), [i])
            
train_corpus = list(read_corpus(df_train["Body"]))

astoeckl

allvecs=[]
for i in range(len(train_corpus)):
    allvecs.append(model.docvecs[i])
    
df_train["vec"] = allvecs

astoeckl

import quandl
quandl.ApiConfig.api_key = "YOURKEY"
apple = quandl.get("EOD/AAPL", start_date="2016-02-22", end_date="2018-07-23")
print(apple.shape)
apple.head()

astoeckl

apple["Daily Change"] = apple["Adj_Close"].diff(periods=1)
apple.head()

astoeckl

result = pd.merge(df_train, apple, on='Date')
print(result.shape)

astoeckl

import numpy as np

feat, y = result["vec"].values, result["Daily Change"].values
frames = [pd.DataFrame(result["vec"].values[i]).transpose() for i in range(result.shape[0]) ]
X = pd.concat(frames)

astoeckl

from keras.models import Sequential
from keras.layers import Input, Dense, Dropout
from keras.models import Model

regmodel = Sequential()
regmodel.add(Dense(150, activation='relu', input_shape=(300,)))
regmodel.add(Dropout(0.5))
regmodel.add(Dense(75, activation='relu'))
regmodel.add(Dropout(0.5))
regmodel.add(Dense(1, activation='linear'))

astoeckl

regmodel.compile(loss='mse', optimizer='adam')
history = regmodel.fit(X, y, epochs=500, verbose=2)

astoeckl

from scipy.stats.stats import pearsonr

ytrain = np.squeeze(regmodel.predict(X))
pearsonr(ytrain, y)