import tensorflow as tf
from tensorflow import keras
from sklearn.model_selection import train_test_split
from random import random
import numpy as np

targets = []
inputs = []

CLASS_A = 0
CLASS_B = 1

def sort_by_probability(pair):
    return pair[0]

def create_model(input_count):
    model = keras.Sequential([
        keras.layers.Dense(3, input_dim=input_count, activation=tf.nn.relu),
        keras.layers.Dense(3, activation=tf.nn.relu, kernel_regularizer=keras.regularizers.l2(0.0000)),
        keras.layers.Dense(1, activation=tf.nn.sigmoid, kernel_regularizer=keras.regularizers.l2(0.00000))
    ])
    model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
    return model

def assess_accuracy(model, targets_validation, inputs_validation):
    predictions = [(float(v[0]), targets_validation[i], inputs_validation[i]) for i, v in enumerate(model.predict(np.asarray(inputs_validation, dtype=np.float32)))]
    predictions.sort(key=sort_by_probability)
    predictions.reverse()
    correct = 0
    for pair in predictions[:100]:
        if(pair[1] == CLASS_B):
            correct += 1
    print(str(correct) + "% correct")

for v in range(0, 100000):
    target = CLASS_A if random() < 0.8 else CLASS_B
    feature = [0.45 + 0.1 * random()] if random() < 0.2 and  target == CLASS_B else [random()]
    targets.append(target)
    inputs.append(feature)
inputs_training, inputs_validation, targets_training, targets_validation = train_test_split(inputs, targets, test_size=0.3, random_state=42)
model = create_model(1)
model.fit(np.array(inputs_training), np.array(targets_training), epochs=10, verbose=0, class_weight = {0:1, 1:4})
print("With one feature")
assess_accuracy(model, targets_validation, inputs_validation)

print("Now with three features, each equally meaningful as the first one")
inputs = []
targets = []
for v in range(0, 100000):
    target = CLASS_A if random() < 0.8 else CLASS_B
    featureA = 0.45 + 0.1 * random() if random() < 0.2 and target == CLASS_B else random()
    featureB = 0.45 + 0.1 * random() if random() < 0.2 and target == CLASS_B else random()
    featureC = 0.45 + 0.1 * random() if random() < 0.2 and target == CLASS_B else random()
    targets.append(target)
    inputs.append([featureA, featureB, featureC])
inputs_training, inputs_validation, targets_training, targets_validation = train_test_split(inputs, targets, test_size=0.3, random_state=42)
model = create_model(3)
model.fit(np.array(inputs_training), np.array(targets_training), epochs=10, verbose=0, class_weight = {0:1, 1:4})
assess_accuracy(model, targets_validation, inputs_validation)