Cacher is the code snippet organizer for pro developers

import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np

from sklearn.cross_validation import train_test_split
from sklearn.linear_model import LogisticRegressionCV

from keras.models import Sequential
from keras.layers.core import Dense, Activation
from keras.utils import np_utils

import ssl
ssl._create_default_https_context = ssl._create_unverified_context

# https://github.com/fastforwardlabs/keras-hello-world/blob/master/kerashelloworld.ipynb

def one_hot_encode_object_array(arr):
    uniques, ids = np.unique(arr, return_inverse=True)
    return np_utils.to_categorical(ids, len(uniques))

def iris_sklearn():
    iris = sns.load_dataset("iris")
    
    X = iris.values[:, :4]
    y = iris.values[:, 4]

    train_X, test_X, train_y, test_y = train_test_split(X, y, train_size=0.5, random_state=0)
    print(train_X.shape, test_X.shape, train_y.shape, test_y.shape)

    lr = LogisticRegressionCV()
    lr.fit(train_X, train_y)

    print("Accuracy = {:.2f}".format(lr.score(test_X, test_y)))

def iris_keras():
    iris = sns.load_dataset("iris")
    
    X = iris.values[:, :4]
    y = iris.values[:, 4]

    train_X, test_X, train_y, test_y = train_test_split(X, y, train_size=0.5, random_state=0)
    print(train_X.shape, test_X.shape, train_y.shape, test_y.shape)

    train_y_ohe = one_hot_encode_object_array(train_y)
    test_y_ohe = one_hot_encode_object_array(test_y)

    model = Sequential()
    model.add(Dense(16, input_shape=(4, )))
    model.add(Activation('sigmoid'))

    model.add(Dense(3))
    model.add(Activation('softmax'))

    model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
    model.fit(train_X, train_y_ohe, nb_epoch=100, batch_size=1, verbose=0)
    loss, accuracy = model.evaluate(test_X, test_y_ohe, verbose=0)
    print("Accuracy = {:.2f}".format(accuracy))

if __name__ == "__main__":
#    iris_sklearn()
    iris_keras()