#!/bin/python3
import math
import os
import random
import re
import sys


#
# Complete the 'predictMissingPrices' function below.
#
# The function is expected to return an INTEGER_ARRAY.
# The function accepts following parameters:
#  1. STRING startDate
#  2. STRING endDate
#  3. STRING_ARRAY knownTimestamps
#  4. INTEGER_ARRAY pricesAtKnownTimeStamps
#  5. STRING_ARRAY unknownTimestamps
#

# y = mx + b
def makeLinearPrediction(value_map, i, X):
    # need two points
    X1 = None
    X2 = None
    if i == 0:
        j = i + 1
        while X1 == None or X2 == None:
            if X[j] in value_map:
                if X1 == None:
                    X1 = (j, value_map[X[j]])
                else:
                    X2 = (j, value_map[X[j]])
            j += 1
    elif i == len(X) - 1:
        j = i - 1
        while X1 == None or X2 == None:
            if X[j] in value_map:
                if X1 == None:
                    X1 = (j, value_map[X[j]])
                else:
                    X2 = (j, value_map[X[j]])
            j -= 1
    else:
        distance = 1
        while X1 == None or X2 == None:
            j = i - distance
            if 0 < j and X[j] in value_map:
                if X1 == None:
                    X1 = (j, value_map[X[j]])
                else:
                    X2 = (j, value_map[X[j]])
            j = i + distance
            if j < len(X) and X[j] in value_map:
                if X1 == None:
                    X1 = (j, value_map[X[j]])
                else:
                    X2 = (j, value_map[X[j]])

            distance += 1

    m = (X2[1]-X1[1]) / float((X2[0]-X1[0]))
    b = X1[1] - (m * X1[0])

    x = i
    return round(m * x + b)

def timestampToFeatures(ts):
    ts_split = ts.split(' ')
    date_split = ts_split[0].split('-')
    year = int(date_split[0])
    month = int(date_split[1])
    day = int(date_split[2])
    hour = int(ts_split[1].split(':')[0])

    return (year, month, day, hour)


def predictMissingPrices(startDate, endDate, knownTimestamps, pricesAtKnownTimeStamps, unknownTimestamps):
    value_map = {}
    X_train = [timestampToFeatures(ts) for ts in knownTimestamps]
    for i in range(len(pricesAtKnownTimeStamps)):
        value_map[X_train[i]] = pricesAtKnownTimeStamps[i]

    X_test = [timestampToFeatures(ts) for ts in unknownTimestamps]
    X = sorted(X_train + X_test)

    predictions = []
    for i in range(len(X)):
        Xi = X[i]
        if Xi not in value_map:
            predictions.append(makeLinearPrediction(value_map, i, X))

    return predictions

"""
from sklearn.svm import SVR

def predictMissingPrices(startDate, endDate, knownTimestamps, pricesAtKnownTimeStamps, unknownTimestamps):
    X_train = [timestampToFeatures(ts) for ts in knownTimestamps]
    y_train = pricesAtKnownTimeStamps
    X_test = [timestampToFeatures(ts) for ts in unknownTimestamps]

    model = SVR(kernel='rbf', C=10, gamma=0.1, epsilon=.1)

    model.fit(X_train, y_train)

    y_test = model.predict(X_test)
    return y_test
"""


if __name__ == '__main__':
    # IO handling and parsing