solve.py

import pandas as pd

from preprocess import load_data_set, get_encoder, filter_string
from train import forward
import numpy as np


def load_weights():
    W = pd.read_csv("results/best_weights.csv", header=None)
    W = W.as_matrix()
    return W


def load_bias():
    b = pd.read_csv("results/bias.csv", header=None)
    b = b.as_matrix()
    return b


def encoder():
    return get_encoder(load_data_set()[0])


np.set_printoptions(threshold=np.nan)


def get_encoded_data(string, encoder):
    string = filter_string(string)
    data = encoder.transform([string]).toarray()
    return data


def get_sentiment(word):
    x = get_encoded_data(word, enc)
    sentiment = forward(x, W, bias)
    print(sentiment)
    return sentiment[0][0]


def resolve_sentiment(sentiment):
    result = "Neutral"
    if sentiment < 0.3:
        result = "Negative"

    if sentiment > 0.7:
        result = "Positive",
    return result

from tools import serialize, deserialize


def start_solver():
    global W, bias, enc
    W = load_weights()
    bias = load_bias()
    enc = deserialize("encoder")
    # enc = encoder()
    while True:
        print("Paste comment")
        string = input()
        result = get_sentiment(string)

        print(resolve_sentiment(result))