Skip to content 
import pandas as pd
from sklearn.preprocessing import MinMaxScaler, RobustScaler
import requests
from datetime import datetime
import tensorflow as tf
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import LabelBinarizer,OneHotEncoder,StandardScaler
from sklearn.compose import ColumnTransformer
from sklearn.model_selection import train_test_split
from sklearn.model_selection import TimeSeriesSplit
from sklearn.model_selection import GridSearchCV
from sklearn import svm
from xgboost import XGBRegressor
from sklearn.ensemble import RandomForestRegressor
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.metrics import MeanSquaredError
from tensorflow.keras import regularizers
import numpy as np

from sklearn.metrics import mean_squared_error

Preprocess Data

train = pd.read_csv('/content/drive/MyDrive/courses/HKUST/MSBD5001/project/data/group project/disney_shanghai.csv', parse_dates=['Time'])
test = pd.read_csv('/content/drive/MyDrive/courses/HKUST/MSBD5001/project/data/group project/disney_shanghai_test.csv', 
                   names=['Time', 'Facility ID', 'Name', 'Wait time', 'Ride yype', 
                          'Temperature', 'Max temperature', 'Min temperature', 'Humidity', 
                          'Pressure', 'Wind degree', 'Wind speed', 'Cloud', 'Weather', 
                          'Weather description'],
                   parse_dates=['Time'])

print(train.info())

print(test.info())

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 32024 entries, 0 to 32023
Data columns (total 17 columns):
 #   Column               Non-Null Count  Dtype                                
---  ------               --------------  -----                                
 0   Name                 32024 non-null  object                               
 1   Ride type            32024 non-null  object                               
 2   Time                 32024 non-null  datetime64[ns, pytz.FixedOffset(480)]
 3   Fastpass-avaliable   32024 non-null  bool                                 
 4   Status               32024 non-null  object                               
 5   Wait time            14621 non-null  float64                              
 6   Weather              32024 non-null  object                               
 7   Weather description  32024 non-null  object                               
 8   Temperature          32024 non-null  float64                              
 9   Max temperature      32024 non-null  float64                              
 10  Min temperature      32024 non-null  float64                              
 11  Pressure             32024 non-null  int64                                
 12  Humidity             32024 non-null  int64                                
 13  Wind degree          32024 non-null  float64                              
 14  Wind speed           32024 non-null  float64                              
 15  Cloud                32024 non-null  int64                                
 16  Visibility           32024 non-null  int64                                
dtypes: bool(1), datetime64[ns, pytz.FixedOffset(480)](1), float64(6), int64(4), object(5)
memory usage: 3.9+ MB
None
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 10449 entries, 0 to 10448
Data columns (total 15 columns):
 #   Column               Non-Null Count  Dtype              
---  ------               --------------  -----              
 0   Time                 10449 non-null  datetime64[ns, UTC]
 1   Facility ID          10449 non-null  object             
 2   Name                 10449 non-null  object             
 3   Wait time            4016 non-null   float64            
 4   Ride yype            10449 non-null  object             
 5   Temperature          10449 non-null  int64              
 6   Max temperature      10449 non-null  float64            
 7   Min temperature      10449 non-null  float64            
 8   Humidity             10449 non-null  int64              
 9   Pressure             10449 non-null  int64              
 10  Wind degree          10449 non-null  int64              
 11  Wind speed           10449 non-null  float64            
 12  Cloud                10449 non-null  int64              
 13  Weather              10449 non-null  object             
 14  Weather description  10449 non-null  object             
dtypes: datetime64[ns, UTC](1), float64(4), int64(5), object(5)
memory usage: 1.2+ MB
None

show data

train = train[['Name', 'Time', 'Wait time', 'Weather', 'Temperature', 'Max temperature', 'Min temperature', 'Pressure', 'Humidity', 'Wind degree', 'Wind speed', 'Cloud']]
train.head(2)
Name Time Wait time Weather Temperature Max temperature Min temperature Pressure Humidity Wind degree Wind speed Cloud
0 Camp Discovery 2020-08-31 13:38:47.638019+08:00 0.0 Clouds 33.28 34.0 34.0 1009 66 110.0 5.0 75
1 Challenge Trails at Camp Discovery 2020-08-31 13:38:47.638035+08:00 10.0 Clouds 33.28 34.0 34.0 1009 66 110.0 5.0 75

show test

test = test[['Name', 'Time', 'Wait time', 'Weather', 'Temperature', 'Max temperature', 'Min temperature', 'Pressure', 'Humidity', 'Wind degree', 'Wind speed', 'Cloud']]
test.head(2)
Name Time Wait time Weather Temperature Max temperature Min temperature Pressure Humidity Wind degree Wind speed Cloud
0 Challenge Trails at Camp Discovery 2020-11-14 10:07:34+00:00 30.0 Clouds 18 18.0 18.0 1026 68 70 3.0 20
1 Vista Trail at Camp Discovery 2020-11-14 10:07:34+00:00 0.0 Clouds 18 18.0 18.0 1026 68 70 3.0 20

Change name into values

unique_names = train[['Name']].drop_duplicates()
unique_names['id'] = range(len(unique_names))
train = train.merge(unique_names, on='Name')
unique_names
Name id
0 Camp Discovery 0
1 Challenge Trails at Camp Discovery 1
2 Vista Trail at Camp Discovery 2
3 Soaring Over the Horizon 3
4 “Once Upon a Time” Adventure 4
5 Alice in Wonderland Maze 5
6 Frozen: A Sing-Along Celebration 6
7 Hunny Pot Spin 7
8 Peter Pan’s Flight 8
9 Seven Dwarfs Mine Train 9
10 The Many Adventures of Winnie the Pooh 10
11 Voyage to the Crystal Grotto 11
12 Dumbo the Flying Elephant 12
13 Fantasia Carousel 13
14 TRON Lightcycle Power Run – Presented by Chevr... 14
15 Stitch Encounter 15
16 Jet Packs 16
17 Buzz Lightyear Planet Rescue 17
18 Siren's Revenge 18
19 Shipwreck Shore 19
20 Pirates of the Caribbean Battle for the Sunken... 20
21 Explorer Canoes 21
22 Eye of the Storm: Captain Jack’s Stunt Spectac... 22
23 Roaring Rapids 23
24 Ignite the Dream - A Nighttime Spectacular of ... 24
25 Mickey’s Storybook Express 25
26 Golden Fairytale Fanfare 26
27 TRON Realm, Chevrolet Digital Challenge 27
28 Rex’s Racer 28
29 Slinky Dog Spin 29
30 Woody’s Roundup 30
31 Marvel Universe 31
32 Club Destin-E 32
33 Color Wall 33
34 Avengers Training Initiative 34
35 Wave Hello to Your Favorite Mickey Avenue Char... 35
36 Mickey Avenue Kiss Goodnight 36
37 Adventurous Friends Exploration 37
38 Catch a Glimpse of Jack Sparrow 38
39 Hundred Acre Wood Character Sighting 39
40 Princess Balcony Greetings 40
41 Avengers Assemble at the E-Stage 41
42 Woody’s Rescue Patrol 42

Dropna

train = train.dropna()
test = test.dropna()

Add busy label

def is_busy(time: int):
    if time < 30:
        return [1, 0, 0]
    elif time >= 30 and time <= 70:
        return [0, 1, 0]
    else:
        return [0, 0, 1]

Add weekend and public holiday

def hour_modify(x: datetime):
    Early_Morning = [4,5,6,7]
    Morning = [8,9,10,11]
    Afternoon = [12,13,14,15]
    Evening = [16,17,18,19]
    Night = [20,21,22,23]
    Late_Night = [0,1,2,3]


    if x.hour in Early_Morning:
        return 'Early_Morning'
    elif x.hour in Morning:
        return 'Morning'
    elif x.hour in Afternoon:
        return 'Afternoon'
    elif x.hour in Evening:
        return 'Evening'
    elif x.hour in Night:
        return 'Night'
    else:
        return 'Late_Night'

def add_holiday_and_weekend(df: pd.DataFrame, date_field_str='date') -> pd.DataFrame:
    """
    Add holiday and weekend to the dataset
    """
    new_df = df.copy()
    new_df['IsWeekend'] = new_df[date_field_str].apply(lambda x:0 if x.weekday() in [0,1,2,3,4] else 1)
    new_df['IsHoliday']=new_df[date_field_str].apply(lambda x:1 if (x.date().strftime('%Y-%m-%d') in [
           '2020-01-01', '2020-01-24', '2020-01-24', '2020-01-25', '2020-01-26', '2020-01-27', '2020-01-28',
           '2020-01-29', '2020-01-30', '2020-01-31', '2020-02-01', '2020-02-02', '2020-04-04', '2020-05-01',
           '2020-05-02', '2020-05-03', '2020-05-04', '2020-05-05', '2020-06-25', '2020-06-26', '2020-06-27',
           '2020-10-01', '2020-10-02', '2020-10-03', '2020-10-04', '2020-10-05', '2020-10-06', '2020-10-07',
           '2020-10-08', '2020-10-31'])
           or(x.weekday() in[6]) else 0)
    return new_df

convert time zone

train['Time'] = train['Time'].dt.tz_localize(None)

print(train['Time'])

0       2020-08-31 13:38:47.638019
1       2020-08-31 14:09:47.630487
2       2020-08-31 14:26:10.487907
3       2020-08-31 14:29:48.273505
4       2020-08-31 15:09:11.331162
                   ...            
23808   2020-10-24 15:23:47.123270
23809   2020-10-24 16:25:44.947592
23810   2020-10-24 17:23:16.431926
23816   2020-10-25 10:26:28.625480
23817   2020-10-25 11:24:10.125045
Name: Time, Length: 14621, dtype: datetime64[ns]

test['Time'] = test['Time'].dt.tz_convert("Asia/Shanghai").dt.tz_localize(None)
print(test['Time'])

0       2020-11-14 18:07:34
1       2020-11-14 18:07:34
2       2020-11-14 18:07:34
3       2020-11-14 18:07:34
4       2020-11-14 18:07:34
                ...        
10340   2020-12-01 19:07:47
10348   2020-12-01 19:07:47
10349   2020-12-01 19:07:47
10350   2020-12-01 19:07:47
10351   2020-12-01 19:07:47
Name: Time, Length: 4016, dtype: datetime64[ns]

Add public holidays

train = add_holiday_and_weekend(train, 'Time')
test = add_holiday_and_weekend(test, 'Time')

train['Hour modify'] = train['Time'].apply(hour_modify)
test['Hour modify'] = test['Time'].apply(hour_modify)
train['Is busy'] = train['Wait time'].apply(is_busy)

train.head(3)
Name Time Wait time Weather Temperature Max temperature Min temperature Pressure Humidity Wind degree Wind speed Cloud id IsWeekend IsHoliday Hour modify Is busy
0 Camp Discovery 2020-08-31 13:38:47.638019 0.0 Clouds 33.28 34.00 34.00 1009 66 110.0 5.0 75 0 0 0 Afternoon [1, 0, 0]
1 Camp Discovery 2020-08-31 14:09:47.630487 0.0 Clouds 33.80 35.56 35.56 1008 70 110.0 5.0 75 0 0 0 Afternoon [1, 0, 0]
2 Camp Discovery 2020-08-31 14:26:10.487907 0.0 Clouds 34.11 35.56 35.56 1008 66 110.0 6.0 75 0 0 0 Afternoon [1, 0, 0]

Plot Data

plt.figure(figsize=(6,4))
sns.boxplot('Wait time',data=train,orient='h',palette="Set3",linewidth=2.5)
plt.show()

/usr/local/lib/python3.6/dist-packages/seaborn/_decorators.py:43: FutureWarning: Pass the following variable as a keyword arg: x. From version 0.12, the only valid positional argument will be `data`, and passing other arguments without an explicit keyword will result in an error or misinterpretation.
  FutureWarning

png

train[['Wait time', 'Temperature', 'Humidity', 'Wind degree', 'Wind speed', 'Cloud']].describe()
Wait time Temperature Humidity Wind degree Wind speed Cloud
count 14621.000000 14621.000000 14621.000000 14621.000000 14621.000000 14621.000000
mean 23.616374 24.423189 64.933862 127.520347 5.352968 45.041174
std 26.388862 4.075267 14.270916 119.544200 1.962247 31.870902
min 0.000000 14.800000 33.000000 0.000000 0.450000 0.000000
25% 5.000000 21.570000 56.000000 30.000000 4.000000 20.000000
50% 15.000000 23.930000 61.000000 70.000000 5.000000 40.000000
75% 30.000000 26.900000 74.000000 180.000000 7.000000 75.000000
max 195.000000 35.050000 100.000000 360.000000 11.000000 100.000000

Train

train.columns

Index(['Name', 'Time', 'Wait time', 'Weather', 'Temperature',
       'Max temperature', 'Min temperature', 'Pressure', 'Humidity',
       'Wind degree', 'Wind speed', 'Cloud', 'id', 'IsWeekend', 'IsHoliday',
       'Hour modify', 'Is busy'],
      dtype='object')

def plot_history(history):
  hist = pd.DataFrame(history.history)
  hist['epoch'] = history.epoch

  plt.figure()
  plt.xlabel('Epoch')
  plt.ylabel('Mean Abs Error [MPG]')
  plt.plot(hist['epoch'], hist['mae'],
           label='Train Error')
  plt.plot(hist['epoch'], hist['val_mae'],
           label = 'Val Error')
  plt.legend()


  plt.figure()
  plt.xlabel('Epoch')
  plt.ylabel('Mean Square Error [$MPG^2$]')
  plt.plot(hist['epoch'], hist['mse'],
           label='Train Error')
  plt.plot(hist['epoch'], hist['val_mse'],
           label = 'Val Error')
  plt.legend()
  plt.show()
  plt.savefig('MSE.png')



def generate_training_data(data: pd.DataFrame, prediction_label, cat_vars=['id', 'IsWeekend','IsHoliday','Hour modify', 'Weather'], 
                           num_vars=['Temperature', 'Pressure', 'Humidity', 'Cloud', 'Wind degree'], 
                           should_reshape=True, should_split = True):
    x = train.copy()
    y = x[prediction_label].to_list()
    y = np.array(y)

    numeric_transformer=Pipeline(steps=[
    ('scaler', RobustScaler())])
    categorical_transformer=Pipeline(steps=[
    ('oneHot',OneHotEncoder(sparse=False))])

    preprocessor=ColumnTransformer(transformers=[
    ('num',numeric_transformer,num_vars),
    ('cat',categorical_transformer,cat_vars)])

    data_transformed=preprocessor.fit_transform(x)
    if should_split:
        if should_reshape:
            y = y.reshape(-1, 1)
            scaler = MinMaxScaler()
            scaled_y = scaler.fit_transform(y)
            return train_test_split(data_transformed,scaled_y ,test_size=0.02,random_state=42), scaler
        else:
            return train_test_split(data_transformed,y ,test_size=0.02,random_state=42)
    else:
        return data_transformed, y

data, scaler=generate_training_data(train, 'Wait time')
X_train,X_test,y_train,y_test = data
print(y_train.shape)
print(X_train.shape)

(14328, 1)
(14328, 44)

Training by using DNN

early_stop = keras.callbacks.EarlyStopping(monitor='val_loss', patience=20)
model = keras.Sequential(
    [
        layers.Dense(64, activation="relu",  
            input_shape=(X_train.shape[1],),),
        layers.Dense(512, activation="relu",  
                     input_shape=(X_train.shape[1],),),
        layers.Dropout(0.5),
        layers.Dense(256, activation="relu",  
                     input_shape=(X_train.shape[1],),),
        layers.Dropout(0.5),
        layers.Dense(128, activation="relu",  
                     input_shape=(X_train.shape[1],),),
        layers.Dense(y_train.shape[1] ),
    ]
)

opt = keras.optimizers.Adam(learning_rate=0.001)

model.compile(loss='mean_squared_error',
              optimizer=opt, metrics=['mae', 'mse'])

history = model.fit(X_train, y_train, epochs=1000, callbacks=[early_stop], validation_split=0.2)
plot_history(history)

Epoch 1/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0127 - mae: 0.0778 - mse: 0.0127 - val_loss: 0.0115 - val_mae: 0.0670 - val_mse: 0.0115
Epoch 2/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0069 - mae: 0.0563 - mse: 0.0069 - val_loss: 0.0065 - val_mae: 0.0515 - val_mse: 0.0065
Epoch 3/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0058 - mae: 0.0510 - mse: 0.0058 - val_loss: 0.0060 - val_mae: 0.0485 - val_mse: 0.0060
Epoch 4/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0052 - mae: 0.0480 - mse: 0.0052 - val_loss: 0.0061 - val_mae: 0.0472 - val_mse: 0.0061
Epoch 5/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0052 - mae: 0.0470 - mse: 0.0052 - val_loss: 0.0055 - val_mae: 0.0464 - val_mse: 0.0055
Epoch 6/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0046 - mae: 0.0449 - mse: 0.0046 - val_loss: 0.0051 - val_mae: 0.0482 - val_mse: 0.0051
Epoch 7/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0046 - mae: 0.0439 - mse: 0.0046 - val_loss: 0.0045 - val_mae: 0.0434 - val_mse: 0.0045
Epoch 8/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0044 - mae: 0.0434 - mse: 0.0044 - val_loss: 0.0053 - val_mae: 0.0485 - val_mse: 0.0053
Epoch 9/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0042 - mae: 0.0423 - mse: 0.0042 - val_loss: 0.0054 - val_mae: 0.0442 - val_mse: 0.0054
Epoch 10/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0039 - mae: 0.0410 - mse: 0.0039 - val_loss: 0.0049 - val_mae: 0.0448 - val_mse: 0.0049
Epoch 11/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0040 - mae: 0.0410 - mse: 0.0040 - val_loss: 0.0041 - val_mae: 0.0422 - val_mse: 0.0041
Epoch 12/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0039 - mae: 0.0409 - mse: 0.0039 - val_loss: 0.0043 - val_mae: 0.0416 - val_mse: 0.0043
Epoch 13/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0037 - mae: 0.0399 - mse: 0.0037 - val_loss: 0.0041 - val_mae: 0.0415 - val_mse: 0.0041
Epoch 14/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0038 - mae: 0.0403 - mse: 0.0038 - val_loss: 0.0040 - val_mae: 0.0432 - val_mse: 0.0040
Epoch 15/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0034 - mae: 0.0390 - mse: 0.0034 - val_loss: 0.0039 - val_mae: 0.0398 - val_mse: 0.0039
Epoch 16/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0034 - mae: 0.0383 - mse: 0.0034 - val_loss: 0.0038 - val_mae: 0.0405 - val_mse: 0.0038
Epoch 17/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0034 - mae: 0.0382 - mse: 0.0034 - val_loss: 0.0038 - val_mae: 0.0391 - val_mse: 0.0038
Epoch 18/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0034 - mae: 0.0381 - mse: 0.0034 - val_loss: 0.0038 - val_mae: 0.0386 - val_mse: 0.0038
Epoch 19/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0032 - mae: 0.0372 - mse: 0.0032 - val_loss: 0.0039 - val_mae: 0.0427 - val_mse: 0.0039
Epoch 20/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0031 - mae: 0.0372 - mse: 0.0031 - val_loss: 0.0039 - val_mae: 0.0397 - val_mse: 0.0039
Epoch 21/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0032 - mae: 0.0368 - mse: 0.0032 - val_loss: 0.0039 - val_mae: 0.0407 - val_mse: 0.0039
Epoch 22/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0029 - mae: 0.0358 - mse: 0.0029 - val_loss: 0.0038 - val_mae: 0.0393 - val_mse: 0.0038
Epoch 23/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0029 - mae: 0.0359 - mse: 0.0029 - val_loss: 0.0035 - val_mae: 0.0384 - val_mse: 0.0035
Epoch 24/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0029 - mae: 0.0352 - mse: 0.0029 - val_loss: 0.0037 - val_mae: 0.0403 - val_mse: 0.0037
Epoch 25/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0028 - mae: 0.0349 - mse: 0.0028 - val_loss: 0.0035 - val_mae: 0.0379 - val_mse: 0.0035
Epoch 26/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0029 - mae: 0.0351 - mse: 0.0029 - val_loss: 0.0039 - val_mae: 0.0395 - val_mse: 0.0039
Epoch 27/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0028 - mae: 0.0349 - mse: 0.0028 - val_loss: 0.0035 - val_mae: 0.0362 - val_mse: 0.0035
Epoch 28/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0027 - mae: 0.0342 - mse: 0.0027 - val_loss: 0.0042 - val_mae: 0.0387 - val_mse: 0.0042
Epoch 29/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0029 - mae: 0.0354 - mse: 0.0029 - val_loss: 0.0037 - val_mae: 0.0376 - val_mse: 0.0037
Epoch 30/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0027 - mae: 0.0344 - mse: 0.0027 - val_loss: 0.0037 - val_mae: 0.0397 - val_mse: 0.0037
Epoch 31/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0027 - mae: 0.0342 - mse: 0.0027 - val_loss: 0.0036 - val_mae: 0.0371 - val_mse: 0.0036
Epoch 32/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0026 - mae: 0.0339 - mse: 0.0026 - val_loss: 0.0037 - val_mae: 0.0365 - val_mse: 0.0037
Epoch 33/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0026 - mae: 0.0337 - mse: 0.0026 - val_loss: 0.0036 - val_mae: 0.0364 - val_mse: 0.0036
Epoch 34/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0024 - mae: 0.0326 - mse: 0.0024 - val_loss: 0.0036 - val_mae: 0.0373 - val_mse: 0.0036
Epoch 35/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0025 - mae: 0.0329 - mse: 0.0025 - val_loss: 0.0036 - val_mae: 0.0367 - val_mse: 0.0036
Epoch 36/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0024 - mae: 0.0321 - mse: 0.0024 - val_loss: 0.0036 - val_mae: 0.0363 - val_mse: 0.0036
Epoch 37/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0024 - mae: 0.0322 - mse: 0.0024 - val_loss: 0.0034 - val_mae: 0.0363 - val_mse: 0.0034
Epoch 38/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0024 - mae: 0.0323 - mse: 0.0024 - val_loss: 0.0036 - val_mae: 0.0354 - val_mse: 0.0036
Epoch 39/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0025 - mae: 0.0327 - mse: 0.0025 - val_loss: 0.0036 - val_mae: 0.0360 - val_mse: 0.0036
Epoch 40/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0023 - mae: 0.0319 - mse: 0.0023 - val_loss: 0.0034 - val_mae: 0.0360 - val_mse: 0.0034
Epoch 41/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0024 - mae: 0.0318 - mse: 0.0024 - val_loss: 0.0034 - val_mae: 0.0358 - val_mse: 0.0034
Epoch 42/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0023 - mae: 0.0314 - mse: 0.0023 - val_loss: 0.0036 - val_mae: 0.0366 - val_mse: 0.0036
Epoch 43/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0022 - mae: 0.0312 - mse: 0.0022 - val_loss: 0.0034 - val_mae: 0.0361 - val_mse: 0.0034
Epoch 44/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0022 - mae: 0.0312 - mse: 0.0022 - val_loss: 0.0035 - val_mae: 0.0354 - val_mse: 0.0035
Epoch 45/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0023 - mae: 0.0314 - mse: 0.0023 - val_loss: 0.0035 - val_mae: 0.0386 - val_mse: 0.0035
Epoch 46/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0022 - mae: 0.0310 - mse: 0.0022 - val_loss: 0.0035 - val_mae: 0.0359 - val_mse: 0.0035
Epoch 47/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0022 - mae: 0.0313 - mse: 0.0022 - val_loss: 0.0038 - val_mae: 0.0374 - val_mse: 0.0038
Epoch 48/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0022 - mae: 0.0308 - mse: 0.0022 - val_loss: 0.0034 - val_mae: 0.0350 - val_mse: 0.0034
Epoch 49/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0022 - mae: 0.0307 - mse: 0.0022 - val_loss: 0.0035 - val_mae: 0.0357 - val_mse: 0.0035
Epoch 50/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0021 - mae: 0.0305 - mse: 0.0021 - val_loss: 0.0035 - val_mae: 0.0381 - val_mse: 0.0035
Epoch 51/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0022 - mae: 0.0309 - mse: 0.0022 - val_loss: 0.0034 - val_mae: 0.0357 - val_mse: 0.0034
Epoch 52/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0021 - mae: 0.0304 - mse: 0.0021 - val_loss: 0.0034 - val_mae: 0.0347 - val_mse: 0.0034
Epoch 53/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0022 - mae: 0.0304 - mse: 0.0022 - val_loss: 0.0035 - val_mae: 0.0354 - val_mse: 0.0035
Epoch 54/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0021 - mae: 0.0301 - mse: 0.0021 - val_loss: 0.0033 - val_mae: 0.0366 - val_mse: 0.0033
Epoch 55/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0021 - mae: 0.0301 - mse: 0.0021 - val_loss: 0.0034 - val_mae: 0.0348 - val_mse: 0.0034
Epoch 56/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0019 - mae: 0.0288 - mse: 0.0019 - val_loss: 0.0033 - val_mae: 0.0347 - val_mse: 0.0033
Epoch 57/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0020 - mae: 0.0298 - mse: 0.0020 - val_loss: 0.0035 - val_mae: 0.0361 - val_mse: 0.0035
Epoch 58/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0021 - mae: 0.0299 - mse: 0.0021 - val_loss: 0.0034 - val_mae: 0.0353 - val_mse: 0.0034
Epoch 59/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0021 - mae: 0.0299 - mse: 0.0021 - val_loss: 0.0032 - val_mae: 0.0361 - val_mse: 0.0032
Epoch 60/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0021 - mae: 0.0296 - mse: 0.0021 - val_loss: 0.0033 - val_mae: 0.0344 - val_mse: 0.0033
Epoch 61/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0020 - mae: 0.0290 - mse: 0.0020 - val_loss: 0.0033 - val_mae: 0.0346 - val_mse: 0.0033
Epoch 62/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0020 - mae: 0.0297 - mse: 0.0020 - val_loss: 0.0033 - val_mae: 0.0351 - val_mse: 0.0033
Epoch 63/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0020 - mae: 0.0294 - mse: 0.0020 - val_loss: 0.0034 - val_mae: 0.0352 - val_mse: 0.0034
Epoch 64/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0020 - mae: 0.0294 - mse: 0.0020 - val_loss: 0.0034 - val_mae: 0.0360 - val_mse: 0.0034
Epoch 65/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0020 - mae: 0.0292 - mse: 0.0020 - val_loss: 0.0035 - val_mae: 0.0383 - val_mse: 0.0035
Epoch 66/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0020 - mae: 0.0295 - mse: 0.0020 - val_loss: 0.0034 - val_mae: 0.0348 - val_mse: 0.0034
Epoch 67/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0020 - mae: 0.0292 - mse: 0.0020 - val_loss: 0.0034 - val_mae: 0.0346 - val_mse: 0.0034
Epoch 68/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0019 - mae: 0.0287 - mse: 0.0019 - val_loss: 0.0033 - val_mae: 0.0347 - val_mse: 0.0033
Epoch 69/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0284 - mse: 0.0018 - val_loss: 0.0032 - val_mae: 0.0348 - val_mse: 0.0032
Epoch 70/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0019 - mae: 0.0289 - mse: 0.0019 - val_loss: 0.0034 - val_mae: 0.0355 - val_mse: 0.0034
Epoch 71/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0019 - mae: 0.0290 - mse: 0.0019 - val_loss: 0.0035 - val_mae: 0.0372 - val_mse: 0.0035
Epoch 72/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0019 - mae: 0.0284 - mse: 0.0019 - val_loss: 0.0034 - val_mae: 0.0343 - val_mse: 0.0034
Epoch 73/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0283 - mse: 0.0018 - val_loss: 0.0033 - val_mae: 0.0364 - val_mse: 0.0033
Epoch 74/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0019 - mae: 0.0289 - mse: 0.0019 - val_loss: 0.0033 - val_mae: 0.0364 - val_mse: 0.0033
Epoch 75/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0282 - mse: 0.0018 - val_loss: 0.0034 - val_mae: 0.0345 - val_mse: 0.0034
Epoch 76/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0019 - mae: 0.0285 - mse: 0.0019 - val_loss: 0.0033 - val_mae: 0.0344 - val_mse: 0.0033
Epoch 77/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0019 - mae: 0.0283 - mse: 0.0019 - val_loss: 0.0034 - val_mae: 0.0352 - val_mse: 0.0034
Epoch 78/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0278 - mse: 0.0018 - val_loss: 0.0034 - val_mae: 0.0355 - val_mse: 0.0034
Epoch 79/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0277 - mse: 0.0018 - val_loss: 0.0034 - val_mae: 0.0362 - val_mse: 0.0034
Epoch 80/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0017 - mae: 0.0277 - mse: 0.0017 - val_loss: 0.0033 - val_mae: 0.0361 - val_mse: 0.0033
Epoch 81/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0277 - mse: 0.0018 - val_loss: 0.0034 - val_mae: 0.0369 - val_mse: 0.0034
Epoch 82/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0282 - mse: 0.0018 - val_loss: 0.0034 - val_mae: 0.0352 - val_mse: 0.0034
Epoch 83/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0276 - mse: 0.0018 - val_loss: 0.0035 - val_mae: 0.0354 - val_mse: 0.0035
Epoch 84/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0282 - mse: 0.0018 - val_loss: 0.0034 - val_mae: 0.0343 - val_mse: 0.0034
Epoch 85/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0019 - mae: 0.0280 - mse: 0.0019 - val_loss: 0.0035 - val_mae: 0.0375 - val_mse: 0.0035
Epoch 86/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0017 - mae: 0.0273 - mse: 0.0017 - val_loss: 0.0035 - val_mae: 0.0352 - val_mse: 0.0035
Epoch 87/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0277 - mse: 0.0018 - val_loss: 0.0034 - val_mae: 0.0348 - val_mse: 0.0034
Epoch 88/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0017 - mae: 0.0273 - mse: 0.0017 - val_loss: 0.0033 - val_mae: 0.0346 - val_mse: 0.0033
Epoch 89/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.0018 - mae: 0.0276 - mse: 0.0018 - val_loss: 0.0034 - val_mae: 0.0354 - val_mse: 0.0034

png

png

y_pred = model.predict(X_test)
err=mean_squared_error(y_pred,y_test)
y_pred_ori = scaler.inverse_transform(y_pred)
y_test_ori = scaler.inverse_transform(y_test)
print(err)
print(y_pred_ori[:10])
print(y_test_ori[:10])

0.0023450270504444958
[[ 7.5200114]
 [36.162247 ]
 [-1.5508894]
 [11.4153385]
 [30.12201  ]
 [ 4.0796814]
 [ 2.721789 ]
 [43.332848 ]
 [23.49451  ]
 [47.286026 ]]
[[10.]
 [30.]
 [ 0.]
 [15.]
 [15.]
 [ 5.]
 [ 5.]
 [60.]
 [10.]
 [50.]]

diff = y_test_ori - y_pred_ori
plt.figure(figsize=(20, 10))
plt.plot(y_pred_ori, label="Prediction")
plt.plot(y_test_ori, label="True")
plt.plot(diff, label="Difference")
plt.ylabel('Waiting time')
plt.xlabel('Day')
plt.legend()
plt.savefig('DNN_regresion.png')

png

Training on category

in this case, we will pre-process the wait time into 3 different categories.

  • [1, 0, 0]: Wait time < 30 mins

  • [0, 1, 0]: Wait time >= 30 and <= 70

cat_vars=['Name', 'IsWeekend','IsHoliday','Hour modify', 'Weather']
num_vars=['Temperature', 'Pressure', 'Humidity', 'Cloud', 'Wind degree', ]

X_train,X_test,y_train,y_test = generate_training_data(train, 'Is busy', cat_vars=cat_vars, num_vars=num_vars, should_reshape=False)
print(y_train.shape)
print(X_train.shape)

(14328, 3)
(14328, 44)

early_stop = keras.callbacks.EarlyStopping(monitor='val_loss', patience=20)
model = keras.Sequential(
    [
        layers.Dense(64, activation="relu",  
            input_shape=(X_train.shape[1],),),
        layers.Dense(512, activation="relu",  
                     input_shape=(X_train.shape[1],),),
        layers.Dropout(0.5),
        layers.Dense(256, activation="relu",  
                     input_shape=(X_train.shape[1],),),
        layers.Dropout(0.5),
        layers.Dense(128, activation="relu",  
                     input_shape=(X_train.shape[1],),),
        layers.Dense(y_train.shape[1], activation='softmax' ),
    ]
)

opt = keras.optimizers.Adam(learning_rate=0.001)

model.compile(loss='categorical_crossentropy',
              optimizer=opt, metrics=['acc', 'mae', 'mse'])

history = model.fit(X_train, y_train, epochs=1000, callbacks=[early_stop], validation_split=0.2)
plot_history(history)

Epoch 1/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2155 - acc: 0.9014 - mae: 0.0904 - mse: 0.0453 - val_loss: 0.4462 - val_acc: 0.8472 - val_mae: 0.1254 - val_mse: 0.0757
Epoch 2/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2120 - acc: 0.9025 - mae: 0.0880 - mse: 0.0441 - val_loss: 0.4410 - val_acc: 0.8381 - val_mae: 0.1258 - val_mse: 0.0754
Epoch 3/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2112 - acc: 0.9035 - mae: 0.0883 - mse: 0.0443 - val_loss: 0.4480 - val_acc: 0.8416 - val_mae: 0.1229 - val_mse: 0.0752
Epoch 4/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2063 - acc: 0.9074 - mae: 0.0856 - mse: 0.0429 - val_loss: 0.4622 - val_acc: 0.8458 - val_mae: 0.1219 - val_mse: 0.0739
Epoch 5/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2055 - acc: 0.9060 - mae: 0.0859 - mse: 0.0430 - val_loss: 0.4529 - val_acc: 0.8423 - val_mae: 0.1225 - val_mse: 0.0752
Epoch 6/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1968 - acc: 0.9093 - mae: 0.0822 - mse: 0.0413 - val_loss: 0.4663 - val_acc: 0.8458 - val_mae: 0.1188 - val_mse: 0.0763
Epoch 7/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2024 - acc: 0.9108 - mae: 0.0838 - mse: 0.0419 - val_loss: 0.5039 - val_acc: 0.8503 - val_mae: 0.1188 - val_mse: 0.0762
Epoch 8/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2025 - acc: 0.9101 - mae: 0.0841 - mse: 0.0419 - val_loss: 0.5043 - val_acc: 0.8486 - val_mae: 0.1139 - val_mse: 0.0758
Epoch 9/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2047 - acc: 0.9053 - mae: 0.0851 - mse: 0.0432 - val_loss: 0.4379 - val_acc: 0.8514 - val_mae: 0.1199 - val_mse: 0.0717
Epoch 10/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1961 - acc: 0.9114 - mae: 0.0826 - mse: 0.0412 - val_loss: 0.4864 - val_acc: 0.8479 - val_mae: 0.1186 - val_mse: 0.0759
Epoch 11/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1993 - acc: 0.9097 - mae: 0.0819 - mse: 0.0412 - val_loss: 0.4532 - val_acc: 0.8465 - val_mae: 0.1209 - val_mse: 0.0736
Epoch 12/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1934 - acc: 0.9108 - mae: 0.0816 - mse: 0.0409 - val_loss: 0.4550 - val_acc: 0.8548 - val_mae: 0.1178 - val_mse: 0.0731
Epoch 13/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1870 - acc: 0.9135 - mae: 0.0788 - mse: 0.0394 - val_loss: 0.4832 - val_acc: 0.8416 - val_mae: 0.1207 - val_mse: 0.0762
Epoch 14/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1905 - acc: 0.9126 - mae: 0.0793 - mse: 0.0398 - val_loss: 0.5012 - val_acc: 0.8486 - val_mae: 0.1182 - val_mse: 0.0759
Epoch 15/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1882 - acc: 0.9163 - mae: 0.0781 - mse: 0.0393 - val_loss: 0.5033 - val_acc: 0.8503 - val_mae: 0.1180 - val_mse: 0.0750
Epoch 16/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1890 - acc: 0.9164 - mae: 0.0788 - mse: 0.0395 - val_loss: 0.4769 - val_acc: 0.8500 - val_mae: 0.1179 - val_mse: 0.0747
Epoch 17/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1904 - acc: 0.9094 - mae: 0.0799 - mse: 0.0402 - val_loss: 0.5425 - val_acc: 0.8524 - val_mae: 0.1129 - val_mse: 0.0751
Epoch 18/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1865 - acc: 0.9167 - mae: 0.0763 - mse: 0.0386 - val_loss: 0.4673 - val_acc: 0.8510 - val_mae: 0.1175 - val_mse: 0.0738
Epoch 19/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1857 - acc: 0.9156 - mae: 0.0769 - mse: 0.0386 - val_loss: 0.4718 - val_acc: 0.8458 - val_mae: 0.1176 - val_mse: 0.0735
Epoch 20/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1864 - acc: 0.9157 - mae: 0.0776 - mse: 0.0390 - val_loss: 0.5050 - val_acc: 0.8423 - val_mae: 0.1175 - val_mse: 0.0760
Epoch 21/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1795 - acc: 0.9187 - mae: 0.0753 - mse: 0.0378 - val_loss: 0.4781 - val_acc: 0.8514 - val_mae: 0.1176 - val_mse: 0.0739
Epoch 22/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1805 - acc: 0.9169 - mae: 0.0750 - mse: 0.0379 - val_loss: 0.4897 - val_acc: 0.8489 - val_mae: 0.1166 - val_mse: 0.0754
Epoch 23/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1818 - acc: 0.9181 - mae: 0.0757 - mse: 0.0378 - val_loss: 0.5106 - val_acc: 0.8510 - val_mae: 0.1143 - val_mse: 0.0748
Epoch 24/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1819 - acc: 0.9193 - mae: 0.0748 - mse: 0.0378 - val_loss: 0.5458 - val_acc: 0.8440 - val_mae: 0.1169 - val_mse: 0.0771
Epoch 25/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1768 - acc: 0.9211 - mae: 0.0736 - mse: 0.0367 - val_loss: 0.5249 - val_acc: 0.8419 - val_mae: 0.1169 - val_mse: 0.0773
Epoch 26/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1752 - acc: 0.9232 - mae: 0.0721 - mse: 0.0362 - val_loss: 0.5045 - val_acc: 0.8475 - val_mae: 0.1176 - val_mse: 0.0743
Epoch 27/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1733 - acc: 0.9235 - mae: 0.0722 - mse: 0.0361 - val_loss: 0.5165 - val_acc: 0.8444 - val_mae: 0.1163 - val_mse: 0.0761
Epoch 28/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1762 - acc: 0.9205 - mae: 0.0729 - mse: 0.0366 - val_loss: 0.4968 - val_acc: 0.8479 - val_mae: 0.1153 - val_mse: 0.0741
Epoch 29/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.1708 - acc: 0.9238 - mae: 0.0709 - mse: 0.0353 - val_loss: 0.5015 - val_acc: 0.8454 - val_mae: 0.1162 - val_mse: 0.0761

png

png

<Figure size 432x288 with 0 Axes>

y_pred = model.predict(X_test)

def print_acc(pred, true):
    right = 0
    for i in range(len(pred)):
        p = np.argmax(pred[i])
        t = np.argmax(true[i])
        if p == t:
            right += 1

    print(f"acc: {right}/{len(pred)}, {right/len(pred)}")


print_acc(y_pred, y_test)

acc: 250/293, 0.8532423208191127

Training by using category

In this example, we will use average-wait-time to compute their waiting time's category

  1. If wait-time is less than the average-wait time of that facility, then it is not busy.

  2. If wait-time is around average wait-time, then it is in a normal situation

  3. If wait-time is above average-wait-time, then it is busy.

In this model, we are basically assume facilities are different. Some will have more popularity than others. And we will use the model to predict whether a model is above its average or not.

def is_busy_2(row):
    if row[3] < row[1] - 5:
        return [1, 0, 0]
    elif row[3] >= row[1] - 5 and row[3] <= row[1] + 5:
        return [0, 1, 0]
    else:
        return [0, 0, 1]


def calculate_average_wait_time(df: pd.DataFrame)-> pd.DataFrame:
    new_df = df.copy()
    new_df2 = df.copy()
    new_df['tmp_date'] = new_df['Time'].apply(lambda x: x.date())
    new_df = new_df[['tmp_date', 'Wait time', 'Name']]
    new_df = new_df.groupby(['Name']).mean()
    new_df = new_df.rename(columns={'Wait time': 'Average wait time'})

    new_df = new_df.merge(new_df2, on='Name', how='right')
    return new_df


average_sampled_train = calculate_average_wait_time(train)
average_sampled_train['Is busy'] = average_sampled_train.apply(is_busy_2, axis=1)
average_sampled_train.sample(20)
Name Average wait time Time Wait time Weather Temperature Max temperature Min temperature Pressure Humidity Wind degree Wind speed Cloud id IsWeekend IsHoliday Hour modify Is busy
6302 Dumbo the Flying Elephant 24.476950 2020-10-03 17:19:47.471631 40.0 Clouds 23.73 24.00 24.00 1012 78 90.0 6.0 75 12 1 1 Evening [0, 0, 1]
10102 Siren's Revenge 5.000000 2020-10-21 13:22:34.657099 5.0 Clouds 18.36 19.44 19.44 1018 88 340.0 5.0 90 18 0 0 Afternoon [0, 1, 0]
5149 The Many Adventures of Winnie the Pooh 21.727129 2020-10-08 13:17:12.792022 20.0 Clouds 23.83 25.00 25.00 1022 53 20.0 9.0 75 10 0 1 Afternoon [0, 1, 0]
8802 Jet Packs 34.943639 2020-10-17 13:21:49.642192 40.0 Clouds 21.54 22.00 22.00 1023 56 10.0 5.0 20 16 1 0 Afternoon [0, 0, 1]
12292 Rex’s Racer 56.937984 2020-09-04 13:17:48.730671 90.0 Clouds 33.22 36.67 36.67 1012 45 170.0 6.0 91 28 0 0 Afternoon [0, 0, 1]
4527 Seven Dwarfs Mine Train 52.788310 2020-10-09 11:21:36.066646 45.0 Clouds 23.20 24.00 24.00 1021 53 10.0 6.0 23 9 0 0 Morning [1, 0, 0]
13143 Slinky Dog Spin 24.613601 2020-09-22 16:21:22.856577 20.0 Clouds 24.80 25.00 25.00 1014 61 60.0 6.0 40 29 0 0 Evening [0, 1, 0]
5309 The Many Adventures of Winnie the Pooh 21.727129 2020-10-22 17:23:25.739682 40.0 Clouds 19.46 20.00 20.00 1016 68 10.0 8.0 40 10 0 0 Evening [0, 0, 1]
5307 The Many Adventures of Winnie the Pooh 21.727129 2020-10-22 15:23:48.253335 10.0 Clouds 22.51 23.33 23.33 1016 56 340.0 5.0 13 10 0 0 Afternoon [1, 0, 0]
12042 Roaring Rapids 31.629555 2020-09-25 21:18:18.834521 5.0 Rain 22.66 23.00 23.00 1014 69 10.0 5.0 100 23 0 0 Night [1, 0, 0]
5539 Voyage to the Crystal Grotto 17.301459 2020-09-18 17:18:57.544340 40.0 Clouds 19.43 20.00 20.00 1018 88 270.0 3.0 40 11 0 0 Evening [0, 0, 1]
11842 Roaring Rapids 31.629555 2020-09-08 14:21:00.395020 20.0 Clouds 33.61 35.00 35.00 1010 49 180.0 8.0 20 23 0 0 Afternoon [1, 0, 0]
6323 Dumbo the Flying Elephant 24.476950 2020-10-05 17:20:52.785435 30.0 Clouds 20.67 21.11 21.11 1020 52 10.0 8.0 36 12 0 1 Evening [0, 0, 1]
11943 Roaring Rapids 31.629555 2020-09-17 16:20:41.428588 25.0 Rain 21.05 21.67 21.67 1010 100 350.0 7.0 75 23 0 0 Evening [1, 0, 0]
5281 The Many Adventures of Winnie the Pooh 21.727129 2020-10-20 10:24:19.155382 20.0 Clouds 21.59 22.00 22.00 1025 53 50.0 6.0 40 10 0 0 Morning [0, 1, 0]
9435 Buzz Lightyear Planet Rescue 7.097039 2020-10-19 15:24:32.051678 10.0 Clouds 21.24 23.33 23.33 1023 56 30.0 5.0 40 17 0 0 Afternoon [0, 1, 0]
1890 “Once Upon a Time” Adventure 7.246127 2020-09-17 10:21:16.322591 5.0 Rain 22.64 23.00 23.00 1010 88 40.0 7.0 75 4 0 0 Morning [0, 1, 0]
9942 Siren's Revenge 5.000000 2020-10-07 17:21:19.897009 5.0 Clouds 21.37 22.00 22.00 1021 56 20.0 8.0 20 18 0 1 Evening [0, 1, 0]
2997 Hunny Pot Spin 9.258114 2020-09-12 09:22:28.558742 5.0 Clouds 26.08 27.78 27.78 1013 78 330.0 3.0 20 7 1 0 Morning [0, 1, 0]
13470 Slinky Dog Spin 24.613601 2020-10-19 12:25:41.694352 40.0 Clouds 20.87 22.78 22.78 1024 56 20.0 6.0 40 29 0 0 Afternoon [0, 0, 1] 


cat_vars=['Name', 'IsWeekend','IsHoliday','Hour modify', 'Weather']
num_vars=['Temperature', 'Pressure', 'Humidity', 'Cloud', 'Wind degree', ]

X_train,X_test,y_train,y_test = generate_training_data(average_sampled_train, 'Is busy', cat_vars=cat_vars, num_vars=num_vars, should_reshape=False)

early_stop = keras.callbacks.EarlyStopping(monitor='val_loss', patience=20)
model = keras.Sequential(
    [
        layers.Dense(64, activation="relu",  
            input_shape=(X_train.shape[1],),),
        layers.Dense(512, activation="relu",  
                     input_shape=(X_train.shape[1],),),
        layers.Dropout(0.5),
        layers.Dense(256, activation="relu",  
                     input_shape=(X_train.shape[1],),),
        layers.Dropout(0.5),
        # layers.Dense(512, activation="relu",  
        #              input_shape=(X_train.shape[1],),),
        # layers.Dropout(0.5),
        # layers.Dense(512, activation="relu",  
        #              input_shape=(X_train.shape[1],),),
        # layers.Dropout(0.5),
        # layers.Dense(256, activation="relu",  
        #              input_shape=(X_train.shape[1],),),
        # layers.Dropout(0.5),
        layers.Dense(128, activation="relu",  
                     input_shape=(X_train.shape[1],),),
        layers.Dense(y_train.shape[1], activation='softmax' ),
    ]
)

opt = keras.optimizers.Adam(learning_rate=0.001)

model.compile(loss='categorical_crossentropy',
              optimizer=opt, metrics=['acc', 'mae', 'mse'])

history = model.fit(X_train, y_train, epochs=1000, callbacks=[early_stop], validation_split=0.2)
plot_history(history)

Epoch 1/1000
359/359 [==============================] - 1s 4ms/step - loss: 0.5742 - acc: 0.7383 - mae: 0.2304 - mse: 0.1161 - val_loss: 0.4259 - val_acc: 0.8102 - val_mae: 0.1846 - val_mse: 0.0883
Epoch 2/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.4323 - acc: 0.7988 - mae: 0.1772 - mse: 0.0895 - val_loss: 0.4156 - val_acc: 0.8123 - val_mae: 0.1713 - val_mse: 0.0858
Epoch 3/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.4067 - acc: 0.8137 - mae: 0.1683 - mse: 0.0848 - val_loss: 0.4012 - val_acc: 0.8158 - val_mae: 0.1741 - val_mse: 0.0836
Epoch 4/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3872 - acc: 0.8274 - mae: 0.1599 - mse: 0.0805 - val_loss: 0.3918 - val_acc: 0.8266 - val_mae: 0.1623 - val_mse: 0.0805
Epoch 5/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3733 - acc: 0.8318 - mae: 0.1553 - mse: 0.0777 - val_loss: 0.3877 - val_acc: 0.8200 - val_mae: 0.1607 - val_mse: 0.0801
Epoch 6/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3590 - acc: 0.8396 - mae: 0.1489 - mse: 0.0750 - val_loss: 0.3819 - val_acc: 0.8248 - val_mae: 0.1600 - val_mse: 0.0793
Epoch 7/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3567 - acc: 0.8396 - mae: 0.1478 - mse: 0.0742 - val_loss: 0.3821 - val_acc: 0.8287 - val_mae: 0.1570 - val_mse: 0.0788
Epoch 8/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3472 - acc: 0.8437 - mae: 0.1445 - mse: 0.0725 - val_loss: 0.3714 - val_acc: 0.8339 - val_mae: 0.1493 - val_mse: 0.0763
Epoch 9/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3415 - acc: 0.8475 - mae: 0.1421 - mse: 0.0711 - val_loss: 0.3853 - val_acc: 0.8308 - val_mae: 0.1499 - val_mse: 0.0786
Epoch 10/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3323 - acc: 0.8504 - mae: 0.1386 - mse: 0.0694 - val_loss: 0.3726 - val_acc: 0.8374 - val_mae: 0.1404 - val_mse: 0.0752
Epoch 11/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3264 - acc: 0.8540 - mae: 0.1360 - mse: 0.0681 - val_loss: 0.3828 - val_acc: 0.8381 - val_mae: 0.1456 - val_mse: 0.0760
Epoch 12/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3233 - acc: 0.8567 - mae: 0.1334 - mse: 0.0671 - val_loss: 0.3683 - val_acc: 0.8353 - val_mae: 0.1471 - val_mse: 0.0748
Epoch 13/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3184 - acc: 0.8574 - mae: 0.1322 - mse: 0.0663 - val_loss: 0.3639 - val_acc: 0.8454 - val_mae: 0.1422 - val_mse: 0.0734
Epoch 14/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3033 - acc: 0.8622 - mae: 0.1269 - mse: 0.0631 - val_loss: 0.3802 - val_acc: 0.8398 - val_mae: 0.1372 - val_mse: 0.0756
Epoch 15/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.3044 - acc: 0.8618 - mae: 0.1260 - mse: 0.0634 - val_loss: 0.3642 - val_acc: 0.8391 - val_mae: 0.1368 - val_mse: 0.0728
Epoch 16/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2985 - acc: 0.8649 - mae: 0.1247 - mse: 0.0624 - val_loss: 0.3836 - val_acc: 0.8364 - val_mae: 0.1338 - val_mse: 0.0762
Epoch 17/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2926 - acc: 0.8694 - mae: 0.1215 - mse: 0.0611 - val_loss: 0.3739 - val_acc: 0.8423 - val_mae: 0.1384 - val_mse: 0.0740
Epoch 18/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2936 - acc: 0.8684 - mae: 0.1221 - mse: 0.0611 - val_loss: 0.3752 - val_acc: 0.8385 - val_mae: 0.1329 - val_mse: 0.0732
Epoch 19/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2824 - acc: 0.8709 - mae: 0.1180 - mse: 0.0591 - val_loss: 0.3726 - val_acc: 0.8437 - val_mae: 0.1309 - val_mse: 0.0733
Epoch 20/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2793 - acc: 0.8775 - mae: 0.1155 - mse: 0.0581 - val_loss: 0.3680 - val_acc: 0.8378 - val_mae: 0.1411 - val_mse: 0.0732
Epoch 21/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2773 - acc: 0.8749 - mae: 0.1157 - mse: 0.0577 - val_loss: 0.3865 - val_acc: 0.8451 - val_mae: 0.1316 - val_mse: 0.0735
Epoch 22/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2718 - acc: 0.8777 - mae: 0.1130 - mse: 0.0566 - val_loss: 0.3870 - val_acc: 0.8398 - val_mae: 0.1293 - val_mse: 0.0748
Epoch 23/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2668 - acc: 0.8810 - mae: 0.1110 - mse: 0.0558 - val_loss: 0.3765 - val_acc: 0.8465 - val_mae: 0.1308 - val_mse: 0.0730
Epoch 24/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2646 - acc: 0.8838 - mae: 0.1099 - mse: 0.0550 - val_loss: 0.3926 - val_acc: 0.8364 - val_mae: 0.1301 - val_mse: 0.0740
Epoch 25/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2581 - acc: 0.8841 - mae: 0.1073 - mse: 0.0538 - val_loss: 0.3855 - val_acc: 0.8493 - val_mae: 0.1284 - val_mse: 0.0728
Epoch 26/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2576 - acc: 0.8854 - mae: 0.1070 - mse: 0.0537 - val_loss: 0.3911 - val_acc: 0.8440 - val_mae: 0.1263 - val_mse: 0.0743
Epoch 27/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2541 - acc: 0.8856 - mae: 0.1053 - mse: 0.0531 - val_loss: 0.3986 - val_acc: 0.8447 - val_mae: 0.1278 - val_mse: 0.0740
Epoch 28/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2489 - acc: 0.8878 - mae: 0.1038 - mse: 0.0521 - val_loss: 0.4028 - val_acc: 0.8447 - val_mae: 0.1259 - val_mse: 0.0744
Epoch 29/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2455 - acc: 0.8897 - mae: 0.1019 - mse: 0.0513 - val_loss: 0.4089 - val_acc: 0.8468 - val_mae: 0.1248 - val_mse: 0.0732
Epoch 30/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2400 - acc: 0.8934 - mae: 0.1000 - mse: 0.0500 - val_loss: 0.4043 - val_acc: 0.8409 - val_mae: 0.1237 - val_mse: 0.0741
Epoch 31/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2405 - acc: 0.8936 - mae: 0.0993 - mse: 0.0499 - val_loss: 0.3976 - val_acc: 0.8430 - val_mae: 0.1284 - val_mse: 0.0748
Epoch 32/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2382 - acc: 0.8943 - mae: 0.0989 - mse: 0.0496 - val_loss: 0.4170 - val_acc: 0.8402 - val_mae: 0.1280 - val_mse: 0.0755
Epoch 33/1000
359/359 [==============================] - 1s 3ms/step - loss: 0.2382 - acc: 0.8934 - mae: 0.0989 - mse: 0.0498 - val_loss: 0.3866 - val_acc: 0.8461 - val_mae: 0.1285 - val_mse: 0.0718

png

png

<Figure size 432x288 with 0 Axes>

y_pred = model.predict(X_test)

print_acc(y_pred, y_test)

acc: 243/293, 0.8293515358361775

LSTM Training

Window generator

To make a single prediction 24h into the future, given 24h of history you might define a window like this:

image

class WindowGenerator():
    def __init__(self, input_width, offset, data, train_split):
        self.data = data
        self.input_width = input_width
        self.offset = offset
        self.train_split = train_split

    def to_sequences(self):
        """
        Return both data and label
        """
        data_len = len(self.data)
        ret = []
        ret_label = []
        for i in range(data_len - self.offset - self.input_width + 1):
            tmp = self.data[i : i + self.input_width ]
            tmp_label = self.data[i + self.input_width + self.offset - 1]
            ret.append(tmp)
            ret_label.append(tmp_label)

        return np.array(ret), np.array(ret_label)

    def split(self):
        x, y = self.to_sequences()
        num_train = int((1 - self.train_split) * x.shape[0])
        X_train = x[:num_train]
        y_train = y[:num_train]
        X_test = x[num_train:]
        y_test = y[num_train:]
        return X_train, y_train, X_test, y_test

SEQ_LEN = 10
cat_vars=['IsWeekend','IsHoliday']
num_vars=['Temperature', 'Pressure', 'Humidity', 'Cloud', 'Wind degree' ]

from tensorflow.keras.layers import Bidirectional, Dropout, LSTM, Dense, Activation

Preprocess data

average_sampled_train['temp_date'] = average_sampled_train['Time'].apply(lambda x: x.date())
grouped_average = average_sampled_train.groupby('temp_date').mean()
grouped_average = grouped_average[['Temperature', 'Max temperature', 'Min temperature', 'Wind degree', 'Humidity' , 'Wind speed', 'Cloud', 'IsWeekend', 'IsHoliday', "Wait time", "Pressure"]]
grouped_average.columns

Index(['Temperature', 'Max temperature', 'Min temperature', 'Wind degree',
       'Humidity', 'Wind speed', 'Cloud', 'IsWeekend', 'IsHoliday',
       'Wait time', 'Pressure'],
      dtype='object')

numeric_transformer=Pipeline(steps=[
('scaler', RobustScaler())])
categorical_transformer=Pipeline(steps=[
('oneHot',OneHotEncoder(sparse=False))])

preprocessor=ColumnTransformer(transformers=[
('num',numeric_transformer,num_vars),
('cat',categorical_transformer,cat_vars)])

data_transformed=preprocessor.fit_transform(grouped_average)
print(data_transformed.shape)

---------------------------------------------------------------------------

ValueError                                Traceback (most recent call last)

/usr/local/lib/python3.6/dist-packages/sklearn/utils/__init__.py in _get_column_indices(X, key)
    462         try:
--> 463             column_indices = [all_columns.index(col) for col in columns]
    464         except ValueError as e:


/usr/local/lib/python3.6/dist-packages/sklearn/utils/__init__.py in <listcomp>(.0)
    462         try:
--> 463             column_indices = [all_columns.index(col) for col in columns]
    464         except ValueError as e:


ValueError: 'Name' is not in list


The above exception was the direct cause of the following exception:


ValueError                                Traceback (most recent call last)

<ipython-input-38-c2d6048db054> in <module>()
      8 ('cat',categorical_transformer,cat_vars)])
      9 
---> 10 data_transformed=preprocessor.fit_transform(grouped_average)
     11 print(data_transformed.shape)


/usr/local/lib/python3.6/dist-packages/sklearn/compose/_column_transformer.py in fit_transform(self, X, y)
    514         self._validate_transformers()
    515         self._validate_column_callables(X)
--> 516         self._validate_remainder(X)
    517 
    518         result = self._fit_transform(X, y, _fit_transform_one)


/usr/local/lib/python3.6/dist-packages/sklearn/compose/_column_transformer.py in _validate_remainder(self, X)
    322         cols = []
    323         for columns in self._columns:
--> 324             cols.extend(_get_column_indices(X, columns))
    325         remaining_idx = list(set(range(self._n_features)) - set(cols))
    326         remaining_idx = sorted(remaining_idx) or None


/usr/local/lib/python3.6/dist-packages/sklearn/utils/__init__.py in _get_column_indices(X, key)
    466                 raise ValueError(
    467                     "A given column is not a column of the dataframe"
--> 468                 ) from e
    469             raise
    470


ValueError: A given column is not a column of the dataframe

wg = WindowGenerator(data=data_transformed, input_width=SEQ_LEN, offset=0, train_split=0.1)
wg_2 = WindowGenerator(data=grouped_average['Wait time'].to_numpy(), input_width=SEQ_LEN, offset=0, train_split=0.1)
X_train, _, X_test, _ = wg.split()
_, y_train, _, y_test = wg_2.split()
print(X_train.shape)
print(y_train.shape)

---------------------------------------------------------------------------

NameError                                 Traceback (most recent call last)

<ipython-input-39-3e9d8c3f67d6> in <module>()
----> 1 wg = WindowGenerator(data=data_transformed, input_width=SEQ_LEN, offset=0, train_split=0.1)
      2 wg_2 = WindowGenerator(data=grouped_average['Wait time'].to_numpy(), input_width=SEQ_LEN, offset=0, train_split=0.1)
      3 X_train, _, X_test, _ = wg.split()
      4 _, y_train, _, y_test = wg_2.split()
      5 print(X_train.shape)


NameError: name 'WindowGenerator' is not defined

WINDOW_SIZE = SEQ_LEN

model = keras.Sequential()
# Input layer
model.add(Bidirectional(LSTM(WINDOW_SIZE, return_sequences=True), input_shape=(WINDOW_SIZE, X_train.shape[-1])))
"""Bidirectional RNNs allows to train on the sequence data in forward and backward direction."""
model.add(Dropout(rate=0.2))
# 1st Hidden layer
model.add(Bidirectional(LSTM((WINDOW_SIZE * 2), return_sequences = True)))
model.add(Dropout(rate=0.2))
# 2nd Hidden layer
model.add(Bidirectional(LSTM(WINDOW_SIZE, return_sequences=False)))
# output layer
model.add(Dense(units=1))
model.add(Activation('linear'))

model.compile(loss='mean_squared_error', optimizer='adam', metrics=['mae', 'mse'])
model.summary()

history = model.fit(X_train, y_train, epochs=1000, shuffle=False, validation_split=0.1, callbacks=[early_stop]) 

Epoch 1/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2527 - acc: 0.8907 - mae: 0.1016 - mse: 0.0513 - val_loss: 0.3764 - val_acc: 0.8332 - val_mae: 0.1546 - val_mse: 0.0773
Epoch 2/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2506 - acc: 0.8893 - mae: 0.1032 - mse: 0.0517 - val_loss: 0.3732 - val_acc: 0.8353 - val_mae: 0.1477 - val_mse: 0.0752
Epoch 3/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2428 - acc: 0.8924 - mae: 0.1001 - mse: 0.0501 - val_loss: 0.3711 - val_acc: 0.8276 - val_mae: 0.1433 - val_mse: 0.0750
Epoch 4/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2407 - acc: 0.8952 - mae: 0.0993 - mse: 0.0495 - val_loss: 0.3701 - val_acc: 0.8332 - val_mae: 0.1468 - val_mse: 0.0752
Epoch 5/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2404 - acc: 0.8929 - mae: 0.0995 - mse: 0.0496 - val_loss: 0.3691 - val_acc: 0.8409 - val_mae: 0.1387 - val_mse: 0.0740
Epoch 6/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2375 - acc: 0.8962 - mae: 0.0976 - mse: 0.0488 - val_loss: 0.3705 - val_acc: 0.8367 - val_mae: 0.1407 - val_mse: 0.0743
Epoch 7/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2360 - acc: 0.8923 - mae: 0.0981 - mse: 0.0491 - val_loss: 0.3755 - val_acc: 0.8360 - val_mae: 0.1370 - val_mse: 0.0748
Epoch 8/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2347 - acc: 0.8962 - mae: 0.0964 - mse: 0.0482 - val_loss: 0.3740 - val_acc: 0.8451 - val_mae: 0.1411 - val_mse: 0.0746
Epoch 9/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2283 - acc: 0.8962 - mae: 0.0954 - mse: 0.0478 - val_loss: 0.3884 - val_acc: 0.8353 - val_mae: 0.1369 - val_mse: 0.0751
Epoch 10/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2257 - acc: 0.9017 - mae: 0.0935 - mse: 0.0469 - val_loss: 0.3705 - val_acc: 0.8395 - val_mae: 0.1339 - val_mse: 0.0737
Epoch 11/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2231 - acc: 0.8997 - mae: 0.0924 - mse: 0.0463 - val_loss: 0.3835 - val_acc: 0.8416 - val_mae: 0.1296 - val_mse: 0.0732
Epoch 12/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2272 - acc: 0.8972 - mae: 0.0946 - mse: 0.0475 - val_loss: 0.3905 - val_acc: 0.8430 - val_mae: 0.1307 - val_mse: 0.0745
Epoch 13/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2156 - acc: 0.9003 - mae: 0.0904 - mse: 0.0453 - val_loss: 0.3855 - val_acc: 0.8465 - val_mae: 0.1280 - val_mse: 0.0729
Epoch 14/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2194 - acc: 0.9041 - mae: 0.0901 - mse: 0.0452 - val_loss: 0.3853 - val_acc: 0.8381 - val_mae: 0.1290 - val_mse: 0.0740
Epoch 15/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2144 - acc: 0.9045 - mae: 0.0893 - mse: 0.0446 - val_loss: 0.3937 - val_acc: 0.8472 - val_mae: 0.1296 - val_mse: 0.0745
Epoch 16/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2119 - acc: 0.9065 - mae: 0.0874 - mse: 0.0438 - val_loss: 0.3989 - val_acc: 0.8458 - val_mae: 0.1282 - val_mse: 0.0749
Epoch 17/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2132 - acc: 0.9031 - mae: 0.0889 - mse: 0.0446 - val_loss: 0.4087 - val_acc: 0.8458 - val_mae: 0.1235 - val_mse: 0.0749
Epoch 18/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2098 - acc: 0.9048 - mae: 0.0872 - mse: 0.0438 - val_loss: 0.3939 - val_acc: 0.8395 - val_mae: 0.1307 - val_mse: 0.0753
Epoch 19/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2060 - acc: 0.9079 - mae: 0.0858 - mse: 0.0429 - val_loss: 0.3988 - val_acc: 0.8493 - val_mae: 0.1234 - val_mse: 0.0736
Epoch 20/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2082 - acc: 0.9059 - mae: 0.0865 - mse: 0.0435 - val_loss: 0.4140 - val_acc: 0.8507 - val_mae: 0.1226 - val_mse: 0.0741
Epoch 21/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2008 - acc: 0.9093 - mae: 0.0833 - mse: 0.0419 - val_loss: 0.4228 - val_acc: 0.8388 - val_mae: 0.1248 - val_mse: 0.0767
Epoch 22/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2061 - acc: 0.9080 - mae: 0.0849 - mse: 0.0426 - val_loss: 0.4233 - val_acc: 0.8500 - val_mae: 0.1232 - val_mse: 0.0751
Epoch 23/1000
403/403 [==============================] - 1s 3ms/step - loss: 0.2051 - acc: 0.9044 - mae: 0.0854 - mse: 0.0431 - val_loss: 0.4062 - val_acc: 0.8493 - val_mae: 0.1255 - val_mse: 0.0747
Epoch 24/1000
 21/403 [>.............................] - ETA: 0s - loss: 0.2305 - acc: 0.9033 - mae: 0.0996 - mse: 0.0482


---------------------------------------------------------------------------

KeyboardInterrupt                         Traceback (most recent call last)

<ipython-input-40-52ca1206cf0b> in <module>()
----> 1 history = model.fit(X_train, y_train, epochs=1000, shuffle=False, validation_split=0.1, callbacks=[early_stop])


/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/training.py in _method_wrapper(self, *args, **kwargs)
    106   def _method_wrapper(self, *args, **kwargs):
    107     if not self._in_multi_worker_mode():  # pylint: disable=protected-access
--> 108       return method(self, *args, **kwargs)
    109 
    110     # Running inside `run_distribute_coordinator` already.


/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)
   1096                 batch_size=batch_size):
   1097               callbacks.on_train_batch_begin(step)
-> 1098               tmp_logs = train_function(iterator)
   1099               if data_handler.should_sync:
   1100                 context.async_wait()


/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/def_function.py in __call__(self, *args, **kwds)
    778       else:
    779         compiler = "nonXla"
--> 780         result = self._call(*args, **kwds)
    781 
    782       new_tracing_count = self._get_tracing_count()


/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/def_function.py in _call(self, *args, **kwds)
    805       # In this case we have created variables on the first call, so we run the
    806       # defunned version which is guaranteed to never create variables.
--> 807       return self._stateless_fn(*args, **kwds)  # pylint: disable=not-callable
    808     elif self._stateful_fn is not None:
    809       # Release the lock early so that multiple threads can perform the call


/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/function.py in __call__(self, *args, **kwargs)
   2827     with self._lock:
   2828       graph_function, args, kwargs = self._maybe_define_function(args, kwargs)
-> 2829     return graph_function._filtered_call(args, kwargs)  # pylint: disable=protected-access
   2830 
   2831   @property


/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/function.py in _filtered_call(self, args, kwargs, cancellation_manager)
   1846                            resource_variable_ops.BaseResourceVariable))],
   1847         captured_inputs=self.captured_inputs,
-> 1848         cancellation_manager=cancellation_manager)
   1849 
   1850   def _call_flat(self, args, captured_inputs, cancellation_manager=None):


/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/function.py in _call_flat(self, args, captured_inputs, cancellation_manager)
   1922       # No tape is watching; skip to running the function.
   1923       return self._build_call_outputs(self._inference_function.call(
-> 1924           ctx, args, cancellation_manager=cancellation_manager))
   1925     forward_backward = self._select_forward_and_backward_functions(
   1926         args,


/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/function.py in call(self, ctx, args, cancellation_manager)
    548               inputs=args,
    549               attrs=attrs,
--> 550               ctx=ctx)
    551         else:
    552           outputs = execute.execute_with_cancellation(


/usr/local/lib/python3.6/dist-packages/tensorflow/python/eager/execute.py in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
     58     ctx.ensure_initialized()
     59     tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
---> 60                                         inputs, attrs, num_outputs)
     61   except core._NotOkStatusException as e:
     62     if name is not None:


KeyboardInterrupt:

 plot_history(history)

y_pred = model.predict(X_test)
print()
print(y_test[:10])
plt.plot(y_test)
plt.plot(y_pred.flatten())

[15.10416667 25.24390244 20.91836735 37.47126437 19.53846154]





[<matplotlib.lines.Line2D at 0x7f4ce57342e8>]

png

Conclusion

Because we are using averge daily data for LSTM model, we found that it may not be able to predict accurate results based on the current volume of dataset. However, DNN model will provide good prediction for both category results as well as regression results