Written in the Stars: Predict Your Future With Tensorflow and MongoDB Charts

1
from transformers import AutoConfig
2
from transformers import TFAutoModelForSequenceClassification
3
from transformers import AutoTokenizer
4
import numpy
5
import numpy as np
6
from scipy.special import softmax
7
import tensorflow as tf

1
# distilbert model we are using
2
distilbert = "distilbert-base-uncased-finetuned-sst-2-english"
3
4
5
tokenizer = AutoTokenizer.from_pretrained(distilbert)
6
config = AutoConfig.from_pretrained(distilbert)
7
model = TFAutoModelForSequenceClassification.from_pretrained(distilbert)
8
9
10
def sentiment_finder(horoscope):
11
 input = tokenizer(horoscope, padding=True, truncation=True, max_length=512, return_tensors='tf')
12
 output = model(input)
13
 scores = output.logits[0].numpy()
14
 scores = softmax(scores)
15
 ranking = np.argsort(scores)
16
 ranking = ranking[::-1]
17
 return config.id2label[ranking[0]]
18
19
20
# test and see if works before we try on our csv file
21
horoscope = "Things might get a bit confusing for you today, Capricorn. Don't feel like you need to make sense of it all. In fact, this task may be impossible. Just be yourself. Let your creative nature shine through. Other people are quite malleable, and you should feel free to take the lead in just about any situation. Make sure, however, that you consider other people's needs."
22
sentiment = sentiment_finder(horoscope)
23
print(f"Horoscope is {sentiment}")

1
# function to apply sentiment against each horoscope
2
def apply_sentiment(horoscope):
3
 sentiment = sentiment_finder(horoscope)
4
 return 1 if sentiment == "POSITIVE" else 0

1
# import pandas
2
!pip install pandas
3
import pandas as pd

1
df = pd.read_csv("anaiya-six-months-horoscopes.csv")

1
df["sentiment"] = df["horoscope"].apply(apply_sentiment)

1
df.to_csv("anaiya-six-months-horoscopes-sentiment.csv")
2
print("saved to new file called anaiya-six-months-horoscopes-sentiment.csv")

1
import tensorflow_hub as hub

1
df = pd.read_csv("anaiya-six-months-horoscopes-sentiment.csv")
2
df = df[["horoscope", "sentiment"]]

1
df = df.sample(frac=1, random_state=42)

1
X = df["horoscope"]
2
y = df["sentiment"]

1
from imblearn.under_sampling import RandomUnderSampler
2
from sklearn.model_selection import train_test_split
3
4
5
# now do for training and testing
6
X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y, shuffle=True, test_size=0.15, random_state=42)
7
8
9
# create our RandomUnderSampler object
10
rus = RandomUnderSampler(random_state=42, sampling_strategy='majority')
11
12
13
# apply our RUS technique
14
X_resampled, y_resampled = rus.fit_resample(X_train.to_frame(), y_train)
15
16
17
# now convert it back to our dataframe
18
balanced_trained = pd.DataFrame({'horoscope': X_resampled['horoscope'], 'sentiment': y_resampled})

1
sentiment_amount_training = balanced_trained['sentiment'].value_counts()
2
print(sentiment_amount_training)

1
train, val = train_test_split(balanced_trained, test_size=0.2, stratify=balanced_trained['sentiment'], random_state=42)

1
print("Training set:", len(train))
2
print("Validation set:", len(val))
3
print("Test set:", len(X_test))

1
# combine back our X_test and y_test to a df
2
test = pd.DataFrame({'horoscope': X_test, 'sentiment': y_test})

1
# code changed to meet my specific needs
2
def df_to_dataset(dataframe, shuffle=True, batch_size=32):
3
 df = dataframe.copy()
4
 labels = df.pop('sentiment')
5
 df = df['horoscope']
6
 ds = tf.data.Dataset.from_tensor_slices((df, labels))
7
 if shuffle:
8
   ds = ds.shuffle(buffer_size=len(dataframe))
9
 ds = ds.batch(batch_size)
10
 ds = ds.prefetch(tf.data.AUTOTUNE)
11
 return ds

1
train_data = df_to_dataset(train)
2
val_data = df_to_dataset(val)
3
test_data = df_to_dataset(test)

1
list(train_data)[0]

1
# using the embedding layer from tensorflow hub
2
embedding = "https://tfhub.dev/google/nnlm-en-dim50/2"
3
hub_layer = hub.KerasLayer(embedding, dtype=tf.string, trainable=True)

1
# model
2
model = tf.keras.Sequential() # since layer by layer so sequential. most basic form
3
model.add(hub_layer)
4
model.add(tf.keras.layers.Dense(128, activation='relu')) # first neural network layer
5
model.add(tf.keras.layers.Dropout(0.4))
6
model.add(tf.keras.layers.Dense(64, activation='relu')) # second layer
7
model.add(tf.keras.layers.Dropout(0.3)) # another dropout layer
8
model.add(tf.keras.layers.Dense(32, activation='relu')) # third layer
9
model.add(tf.keras.layers.Dropout(0.2)) # another dropout layer
10
# sigmoid is used for binary, so great for sentiment analysis
11
model.add(tf.keras.layers.Dense(1, activation='sigmoid')) # output layer

1
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.0001),
2
             loss=tf.keras.losses.BinaryCrossentropy(),
3
             metrics=['accuracy'])

1
history = model.fit(train_data, epochs=5, validation_data=val_data)

1
loss, accuracy = model.evaluate(test_data)
2
print(f'Loss: {loss}')
3
print(f'Accuracy: {accuracy}')

1
def predict_sentiment(horoscope):
2
   # convert to so tensorflow can understand
3
   encoded_input = tf.constant([horoscope])
4
5
6
   prediction = model.predict(encoded_input)
7
8
9
   # prediction from probability
10
   probability = tf.squeeze(prediction).numpy()
11
   print(f"model probability: {probability}")
12
13
14
   # set it so that we can see if it's negative or positive
15
   sentiment = 1 if probability > 0.5 else 0
16
17
18
   return sentiment
19
20
21
# daily horoscope
22
positive_horoscope = "You're incredibly productive, with good business sense, Libra."
23
negative_horoscope = "This isn't the most cheerful time, Leo, because important issues are rearing their heads again and forcing you to address them."
24
pos_sentiment = predict_sentiment(positive_horoscope)
25
neg_sentiment = predict_sentiment(negative_horoscope)
26
27
28
print(f"This should be positive: {pos_sentiment}")
29
print(f"This should be negative: {neg_sentiment}")

1
file = 'new-week-horoscopes2.csv'
2
df = pd.read_csv(file)
3
4
5
df['sentiment'] = df['horoscope'].apply(predict_sentiment)
6
7
8
for index, row in df.iterrows():
9
 zodiac = row['zodiac']
10
 horoscope = row['horoscope']
11
 sentiment = row['sentiment']
12
 print(f"{zodiac} horoscope is {sentiment}")

1
pip install pymongo

1
# first connect to MongoDB Atlas
2
import getpass
3
from pymongo import MongoClient
4
5
# set up your MongoDB connection
6
connection_string = getpass.getpass(prompt= "Enter connection string WITH USER + PASS here")
7
client = MongoClient(connection_string)
8
9
# we are creating a new collection in the same database as before
10
database = client['horoscopes']
11
collection = database['new_week_horoscope']
12
13
for index, row in df.iterrows():
14
 zodiac = row['zodiac']
15
 horoscope = row['horoscope']
16
 sentiment = row['sentiment']
17
18
 dict = {
19
     "zodiac": zodiac,
20
     "horoscope": horoscope,
21
     "sentiment": sentiment
22
 }
23
24
 collection.insert_one(dict)
25
print("saved in! go check")