# %%
import lzma
import sys
from io import StringIO
from sklearn.feature_extraction.text import TfidfVectorizer
import pandas as pd
import numpy

pathX = "./train/train.tsv.xz"
# pathX = "./train/in.tsv"
# pathY = "./train/meta.tsv.xz"
nrows = 100000

# %%
# data = lzma.open(pathX, mode='rt', encoding='utf-8').read()
# stringIO = StringIO(data)
# df = pd.read_csv(stringIO, sep="\t", header=None)
df = pd.read_csv(pathX, sep='\t', nrows=nrows, header=None)
# df = df.drop(df.columns, axis=1)
# topics = pd.read_csv(pathY, sep='\t', nrows=nrows, header=None)

# %%
print(len(df.index))

# print(len(topics.index))


# %%
def mergeTexts(a, b, c):
    return str(a) + " " + str(b) + " " + str(c)

# %%
def getMean(a, b):
    return ((a + b)/2)

# %%
df["year"] = df.apply(lambda x: getMean(x[0], x[1]), axis = 1)
df["text"] = df.apply(lambda x: x[4], axis = 1)

# %%
df = df.drop(columns = [0,1,2,3,4], axis=1)
df.sample(5)

# %%
topics = df.pop('year')
df.sample(5)

# %%
topics.sample(5)

# %%
vectorizer = TfidfVectorizer(lowercase=True, stop_words=['polish'])
X = vectorizer.fit_transform(df.to_numpy().ravel())
vectorizer.get_feature_names_out()


# %%
# vectorizer.transform("Ala ma kotka".lower().split())

# %%
# df = df.reset_index()

# %%
tfidfVector = vectorizer.transform(df["text"])

    

# %%
# from sklearn.model_selection import train_test_split
# from sklearn.naive_bayes import GaussianNB
# 
# gnb = GaussianNB()
# gnb.fit(tfidfVector.todense(), topics)

# %%
from sklearn.linear_model import LinearRegression

reg = LinearRegression().fit(tfidfVector, topics)


# %%
testXPath = "./dev-0/in.tsv"
testYPath = "./dev-0/expected.tsv"

testX = pd.read_csv(testXPath, sep='\t', nrows=19998, header=None)

testY = pd.read_csv(testYPath, sep='\t', nrows=19998, header=None)


# %%
testX.sample(5)

# %%
testY.sample()


# %%
testXtfidfVector = vectorizer.transform(testX[0])


# %%
reg.score(testXtfidfVector, testY[0])


# %%
testXPath = "./dev-1/in.tsv"
testYPath = "./dev-1/out.tsv"

testX = pd.read_csv(testXPath, sep='\t', nrows=nrows, header=None)

# testY = pd.read_csv(testYPath, sep='\t', nrows=nrows, header=None)
testXtfidfVector = vectorizer.transform(testX[0])


# %%
pred = reg.predict(testXtfidfVector)
print(pred)

import csv
with open(testYPath, 'w', newline='') as f_output:
    tsv_output = csv.writer(f_output, delimiter='\n')
    tsv_output.writerow(pred)