petrosDemetrakopoulos

from matplotlib import pyplot as plt
import numpy as np
import pandas as pd
from scipy.stats import skew, kurtosis
from sklearn.decomposition import PCA
from sklearn.cluster import KMeans

def main():
    data = read_data()
    normalized_data = normalize(data)

petrosDemetrakopoulos

from sklearn.cluster import KMeans
# fit K-Means
kmeans = KMeans(n_clusters=2)
kmeans.fit(extracted_features)
predictions = kmeans.predict(extracted_features)

petrosDemetrakopoulos

def feature_extraction(signal_df, window_size=100, overlap=50):
    num_components = signal_df.shape[1]
    num_samples = signal_df.shape[0]

    # used to rename columns in the extracted features (mentioning the statistical moment)
    new_col_names_mean = dict()
    new_col_names_std = dict()
    new_col_names_skewness = dict()
    new_col_names_kurtosis = dict()

petrosDemetrakopoulos

def normalize(data):
    data_norm = data.copy()
    columns_for_normalization = ['x_acc','y_acc','z_acc',
                                 'x_gravity', 'y_gravity', 'z_gravity',
                                 'x_gyro', 'y_gyro', 'z_gyro']
 
    for column in columns_for_normalization:
        data_norm[column] = (data_norm[column] - data_norm[column].mean()) / data_norm[column].std()
    
    return data_norm

petrosDemetrakopoulos

import pandas as pd

def read_data():
    accelerometer_data = pd.read_csv('./data/Accelerometer.csv',header=0, names=['time','seconds_elapsed','z_acc','y_acc','x_acc'])
    gyro_data = pd.read_csv('./data/Gyroscope.csv',header=0, names=['time','seconds_elapsed','z_gyro','y_gyro','x_gyro'])
    gravity_data = pd.read_csv('./data/Gravity.csv',header=0, names=['time','seconds_elapsed','z_gravity','y_gravity','x_gravity'])

    # drop columns that are common in all 3 dataframes and keep only the ones from the accelerometer dataframe
    gyro_data = gyro_data.drop(['time','seconds_elapsed'],axis=1)
    gravity_data = gravity_data.drop(['time','seconds_elapsed'],axis=1)

petrosDemetrakopoulos

,batch,yeast,hop 1 name,hop 1 weight,hop 1 form,hop 1 phase,hop 1 time,hop 2 name,hop 2 weight,hop 2 form,hop 2 phase,hop 2 time,hop 3 name,hop 3 weight,hop 3 form,hop 3 phase,hop 3 time,hop 4 name,hop 4 weight,hop 4 form,hop 4 phase,hop 4 time,fermentable 1 weight,fermentable 1 name,fermentable 2 weight,fermentable 2 name,fermentable 3 weight,fermentable 3 name,fermentable 4 weight,fermentable 4 name,fermentable 5 weight,fermentable 5 name
0,18.309166931525436,White Labs - California Ale Yeast WLP001,Cluster,7.269675936122823,Pellet,First Wort,38.87988377656771,Zeus,17.583570003261514,Leaf/Whole,Dry Hop,13.42101097713062,Tettnanger,166.09849747403285,Pellet,First Wort,4.812608839959528,Northern Brewer,64.48508972798278,Pellet,Boil,4.7735462463325185,3.5972188703805634,American - Pale 2-Row,6.1346832837085135,United Kingdom - Munich,9.54155046053587,Flaked Oats,6.962939560440913,Flaked Oats,-0.0003344600787370782,0
1,66.00718958026789,Escarpment Labs - Vermont Ale,Southern Cross,59.02688552767579,Pellet,Boil,

petrosDemetrakopoulos

discrete_columns = [
    'hop 1 form',
    'hop 2 form',
    'hop 3 form',
    'hop 4 form',
    'hop 1 phase',
    'hop 2 phase',
    'hop 3 phase',
    'hop 4 phase',
    'hop 1 name',

petrosDemetrakopoulos

# pick a random index from validation dataset
random_index = np.random.choice(range(len(X_val)), size=1)
test_serie_X = X_val[random_index[0]]
test_serie_Y = y_val[random_index[0]]

first_frames = test_serie_X
original_frames = test_serie_Y
# predict the next 18 fames
new_prediction = model.predict(np.expand_dims(first_frames, axis=0))
new_prediction = np.squeeze(new_prediction, axis=0)

petrosDemetrakopoulos

def create_model():
    model = Sequential()
    model.add(ConvLSTM2D(filters=64, kernel_size=(7, 7),
                    input_shape=(18,344,315,1),
                    padding='same',activation=LeakyReLU(alpha=0.01), return_sequences=True))
    model.add(BatchNormalization())
    model.add(ConvLSTM2D(filters=64, kernel_size=(5, 5),
                    padding='same',activation=LeakyReLU(alpha=0.01), return_sequences=True))
    model.add(BatchNormalization())
    model.add(ConvLSTM2D(filters=64, kernel_size=(3, 3),

petrosDemetrakopoulos

def create_dataset_from_raw(directory_path, resize_to):
    resize_width = resize_to[0]
    resize_height = resize_to[1]
    batch_names = [directory_path + name for name in os.listdir(directory_path) if os.path.isdir(os.path.join(directory_path, name))]
    dataset = np.zeros(shape=(len(batch_names),36,resize_height,resize_width)) # (samples, filters, rows = height, cols = width)

    for batch_idx,batch in enumerate(batch_names):
        files = [x for x in os.listdir(batch) if x != '.DS_Store']
        files.sort()
        crn_batch = np.zeros(shape=(36, resize_height, resize_width))