import pandas as pd import matplotlib.pyplot as plt import seaborn as sns from sklearn.preprocessing import OneHotEncoder import tensorflow as tf from keras.models import Sequential from keras.layers import Dense

headers = pd.read_csv('./Data/handles-data.csv') tweets = pd.read_csv('./Data/tweets-data.csv') pd.options.display.float_format = '{:.2f}'.format

headers.shape

headers.head(6)

headers.info()

headers.describe()

headers[headers['tweets']==-1]

tweets.shape

tweets.info()

tweets.describe()

tweets.head()

union = pd.merge(headers, tweets, left_on='username', right_on='username' ) union.describe() union.to_csv("./Data/union.csv",sep=";", index=False)

tweets.shape

headers.shape

union.shape

print(union.columns.tolist())

union = pd.read_csv("./Data/union.csv", sep=";")

union[['nlikes','likes']].describe()

union[['nlikes','likes']].head(10)

union[['nlikes','likes']].corr()

tweets[tweets['username'] == 'latimes']['nlikes'].sum()

headers[headers['username'] == 'latimes']['likes'].sum()

tweets['nlikes'].head(10)

union['nlikes'].describe(percentiles=[0.5, 0.6, 0.7, 0.8, 0.85, 0.9])

union.boxplot(column =['nlikes'], figsize=(10,10))

n_likes = union.groupby(['username'])['nlikes'].count().reset_index() n_likes.sort_values('nlikes',ascending=False) union[union['username'] == 'latimes'][['join_date','date','tweet','followers','following']].sort_values('date', ascending=False) union[union['username'] == 'latimes'][['nlikes']]

union[['nlikes', 'nretweets']].corr()

figura = plt.figure() axes= figura.add_subplot() axes.set_ylim(0,400) tweets.boxplot(column =['nretweets'], figsize=(10,10))

nlikes_400 = union[union["nlikes"] < 400]["nlikes"].count() nlikes_400

nlikes_400_array = union["nlikes"][union["nlikes"] < 400] nlikes_400_df = pd.DataFrame(data=nlikes_400_array, columns=["nlikes"])

sns.histplot(data=nlikes_400_df, x="nlikes")

union['language'].unique() union.groupby(['language'])['language'].count().sort_values(ascending=False).to_frame()

union.groupby(['tweets','username'])['username'].count().sort_index(ascending=False).head(10)

union["followers"].max()

followers_median = union["followers"].median() followers_median

followers_array = union["followers"][union["followers"] < followers_median] followers_df = pd.DataFrame(data=followers_array, columns=["followers"])

sns.boxplot(x=followers_df["followers"])

# Si tiene urls union["have_urls"] = union["urls"] != "[]" # Si contiene videos union["have_video"] = union["video"] != 0 # Si tiene fotos union["have_photos"] = union["urls"] != "[]" # Si contiene hashtags union["have_hashtags"] = union["hashtags"] != "[]" # Cuando la Bio es superior a 90 caracteres union["large_bio"] = union["bio"].str.len() >= 90

union[['have_urls', 'have_video']].corr()

union[['have_hashtags', 'have_urls']].corr()

union["large_bio"].value_counts() union[union['bio'].str.len() >= 100].count()

tweets['date'][0]

from datetime import datetime print(datetime.strptime(tweets['date'][0], '%Y-%m-%d %H:%M:%S').isoweekday())

tweets['date'].isnull().sum()

def calcular_dia(fecha): return datetime.strptime(fecha, '%Y-%m-%d %H:%M:%S').isoweekday() tweets['dia_semana'] = tweets['date'].aggregate([calcular_dia]) tweets[["dia_semana"]]

tweets['dia_semana'].value_counts().sort_index()

import matplotlib.pyplot as plt import seaborn as sns figura = plt.figure(figsize=(15,8)) figura.suptitle("Día y Hora Tweets") axes = figura.add_subplot(2,2,1) axes2 = figura.add_subplot(2,2,2) axes3 = figura.add_subplot(2,2,3) axes4 = figura.add_subplot(2,2,4) axes.set_title("Hist. Hora") axes.set_xlabel("Hora") sns.histplot(x=tweets['hour'],ax=axes, kde=True) sns.histplot(x=tweets['day'],ax=axes2, kde=True) axes3.set_title("Hora / Nº Likes") axes3.set_xlabel("Hora") axes3.scatter(tweets['hour'].values, tweets['nlikes'].values, label="Nº Likes") axes3.scatter(tweets['hour'].values, tweets['nretweets'].values, color="red", label="Nº Retweets") axes3.legend() axes4.set_title("Día / Nº Likes") axes4.scatter(tweets['day'].values, tweets['nlikes'].values) axes4.scatter(tweets['day'].values, tweets['nretweets'].values, color="red")

union[union["have_hashtags"] == False ]["have_hashtags"].count()

union[union["have_urls"] == True ]["have_urls"].count()

union[union["have_photos"] == True ]["have_photos"].count()

union[union["have_video"] == True ]["have_video"].count()

print(tweets.columns.tolist())

datosInteresantes = union[['day','hour','followers','have_urls']] df_c = datosInteresantes.copy() df_c = df_c[df_c["followers"] < 8000000] target = union["nretweets"][union["followers"] < 8000000] print(df_c.count()) print(target.count())

df_c.isnull().sum()

df_c.dtypes

enc_ohe = OneHotEncoder() enc_ohe.fit(df_c[["have_urls"]])

data_encoder = enc_ohe.transform(df_c[['have_urls']]).toarray() print(data_encoder)

df_encoder = pd.DataFrame(data_encoder) df_encoder.head() df_encoder.isnull().sum()

df_encoder.shape

df_encoder.count()

df_c.shape

merge = pd.concat([df_c.reset_index(drop=True), df_encoder.reset_index(drop=True)],axis=1) merge.count() merge

df_c = pd.concat([df_c.reset_index(drop=True),df_encoder.reset_index(drop=True)],axis=1) df_c.drop(['have_urls'], axis=1, inplace=True)

df_c.count()

from sklearn.model_selection import train_test_split

df_c

X_train, X_test, y_train, y_test = train_test_split(df_c, target, test_size=0.3) print(X_train.shape) print(X_test.shape)

import numpy as np from sklearn.linear_model import LinearRegression

# Poongo un seed para empezar siempre desde el mismo número. np.random.seed(5)

model=LinearRegression()

model.fit(X_train, y_train)

predictions = model.predict(X_test) X_tested = X_test.copy() print(X_tested.shape) print(predictions.shape) X_tested['predictions'] = predictions X_tested['Target_test'] = y_test.values X_tested.head(25)

X_train

model=Sequential() model.add(Dense(16, activation='relu',input_dim=5)) model.add(Dense(32, activation='relu')) model.add(Dense(16, activation='relu')) model.add(Dense(8, activation='relu')) model.add(Dense(1, activation='linear')) model.compile(loss='mean_squared_error')

np.random.seed(5) tf.random.set_seed(5) hstory = model.fit(X_train, y_train,epochs=50)

hstory.history["loss"]

predictions = model.predict(X_test) X_tested = X_test.copy() print(X_tested.shape) print(predictions.shape) X_tested['predictions'] = predictions X_tested['Target_test'] = y_test.values X_tested.head(35)