In [431]:
from sklearn.model_selection import train_test_split
import pandas as pd
import numpy as np
import tensorflow as tf
from tensorflow import keras
In [ ]:
def plot_learning_curves(loss, val_loss):
    plt.plot(np.arange(len(loss)) + 0.5, loss, "b.-", label="Training loss")
    plt.plot(np.arange(len(val_loss)) + 1, val_loss, "r.-", label="Validation loss")
    plt.gca().xaxis.set_major_locator(mpl.ticker.MaxNLocator(integer=True))
    plt.axis([1, 20, 0, 0.05])
    plt.legend(fontsize=14)
    plt.xlabel("Epochs")
    plt.ylabel("Loss")
    plt.grid(True)

training_data_sets

In [432]:
training_data_store = pd.read_csv('cut_training_data_file/cut_training_data.csv')
In [433]:
training_data_store.head()
Out[433]:
ax ay az gx gy gz class_
0 3.536667 -8.764157 8.219295 -0.426663 0.376293 0.468237 running
1 3.972804 -12.131095 11.188315 -0.133465 -0.126467 0.777247 running
2 3.418738 -14.802985 12.352308 0.195966 -0.441429 0.986390 running
3 1.820551 -16.844654 8.020269 0.370743 -0.536392 1.180035 running
4 0.771038 -16.254522 3.320959 -0.066549 -0.279340 1.371567 running
In [434]:
pd.api.types.is_string_dtype(training_data_store['class_'])
Out[434]:
True
In [435]:
for label, content in training_data_store.items():
    if pd.api.types.is_string_dtype(content):
        print(label)

class_
In [436]:
for label, content in training_data_store.items():
    if pd.api.types.is_string_dtype(content):
        training_data_store[label] = content.astype('category').cat.as_ordered()
In [437]:
training_data_store['class_'].cat.codes
Out[437]:
0        2
1        2
2        2
3        2
4        2
        ..
17820    0
17821    0
17822    0
17823    0
17824    0
Length: 17825, dtype: int8
In [438]:
training_data_store['class_'] = training_data_store['class_'].cat.codes
In [439]:
training_data_store
Out[439]:
ax ay az gx gy gz class_
0 3.536667 -8.764157 8.219295 -0.426663 0.376293 0.468237 2
1 3.972804 -12.131095 11.188315 -0.133465 -0.126467 0.777247 2
2 3.418738 -14.802985 12.352308 0.195966 -0.441429 0.986390 2
3 1.820551 -16.844654 8.020269 0.370743 -0.536392 1.180035 2
4 0.771038 -16.254522 3.320959 -0.066549 -0.279340 1.371567 2
... ... ... ... ... ... ... ...
17820 -4.225421 -7.347326 2.834419 1.315053 0.989951 0.788295 0
17821 -3.888998 -6.893213 2.058209 1.380154 1.288926 0.862489 0
17822 -4.683547 -6.589226 1.510789 1.444906 1.497039 0.967244 0
17823 -5.238657 -6.579218 0.651558 1.428796 1.569016 1.079469 0
17824 -5.052723 -6.786039 -0.916491 1.325403 1.399684 1.295959 0

17825 rows × 7 columns

Dense¶

In [440]:
X = training_data_store.drop(['class_'], axis='columns')
y = np.ravel(training_data_store['class_'])
In [441]:
X_train, X_temp, y_train, y_temp = train_test_split(X, y, test_size=0.4, random_state=0)
X_valid, X_test, y_valid, y_test = train_test_split(X_temp, y_temp, test_size=0.5, random_state=0)
In [442]:
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler().fit(X_train)
X_train = scaler.transform(X_train)
X_test = scaler.transform(X_test)
In [443]:
import tensorflow as tf
from keras.models import Sequential
import pandas as pd
from keras.layers import Dense

model = Sequential()
model.add(Dense(8, activation='relu', input_shape=(None, 1,6)))
model.add(Dense(8, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer='sgd',
metrics=['accuracy'])
history = model.fit(X_train, y_train,epochs=8, batch_size=1, verbose=1)

Epoch 1/8
WARNING:tensorflow:Model was constructed with shape (None, None, 1, 6) for input KerasTensor(type_spec=TensorSpec(shape=(None, None, 1, 6), dtype=tf.float32, name='dense_59_input'), name='dense_59_input', description="created by layer 'dense_59_input'"), but it was called on an input with incompatible shape (1, 6).
WARNING:tensorflow:Model was constructed with shape (None, None, 1, 6) for input KerasTensor(type_spec=TensorSpec(shape=(None, None, 1, 6), dtype=tf.float32, name='dense_59_input'), name='dense_59_input', description="created by layer 'dense_59_input'"), but it was called on an input with incompatible shape (1, 6).
10695/10695 [==============================] - 9s 777us/step - loss: 654609.3750 - accuracy: 0.2510
Epoch 2/8
10695/10695 [==============================] - 9s 827us/step - loss: -37.6062 - accuracy: 0.2511
Epoch 3/8
10695/10695 [==============================] - 9s 880us/step - loss: -62.5471 - accuracy: 0.2511
Epoch 4/8
10695/10695 [==============================] - 11s 1ms/step - loss: -87.4949 - accuracy: 0.2511
Epoch 5/8
10695/10695 [==============================] - 10s 954us/step - loss: -112.4436 - accuracy: 0.2511
Epoch 6/8
10695/10695 [==============================] - 9s 846us/step - loss: -137.3822 - accuracy: 0.2511
Epoch 7/8
10695/10695 [==============================] - 9s 865us/step - loss: -162.3128 - accuracy: 0.2511
Epoch 8/8
10695/10695 [==============================] - 9s 834us/step - loss: -187.2427 - accuracy: 0.2511
Out[443]:
<keras.callbacks.History at 0x1d2658156c0>
In [ ]:
plot_learning_curves(history.history["loss"], history.history["val_loss"])
plt.show()
In [ ]:
In [ ]:
In [444]:
from keras.models import Sequential
from keras.layers import LSTM
from keras.layers import Dense
from keras.layers import Dropout
In [445]:
X = training_data_store.drop(['class_'], axis='columns')
y = np.ravel(training_data_store['class_'])
In [446]:
X_train, X_temp, y_train, y_temp = train_test_split(X, y, test_size=0.4, random_state=0)
X_valid, X_test, y_valid, y_test = train_test_split(X_temp, y_temp, test_size=0.5, random_state=0)
In [447]:
x_train, y_train = X_train , y_train
x_test, y_test = X_test, y_test
batch_size = 64
epochs = 5
In [448]:
x_train.shape, y_train.shape
Out[448]:
((10695, 6), (10695,))

SimpleRNN¶

In [449]:
model = Sequential()
model.add(keras.Input(shape=(6,1)))
model.add(layers.SimpleRNN(128,return_sequences=False, activation='relu')) # N, 128
model.add(layers.Dense(10))
# model.add(LSTM(units = 50, return_sequences=True, input_shape=(x_train.shape[1], 1)))
In [450]:
# loss and optimizer
loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
optim = keras.optimizers.Adam(learning_rate=0.001)
metrics = ['accuracy']
In [451]:
model.compile(loss=loss, optimizer=optim, metrics=metrics)
In [452]:
model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=2)

Epoch 1/5
168/168 - 1s - loss: 1.0913 - accuracy: 0.5783 - 1s/epoch - 7ms/step
Epoch 2/5
168/168 - 0s - loss: 0.7169 - accuracy: 0.7406 - 402ms/epoch - 2ms/step
Epoch 3/5
168/168 - 0s - loss: 0.5904 - accuracy: 0.7821 - 355ms/epoch - 2ms/step
Epoch 4/5
168/168 - 0s - loss: 0.5399 - accuracy: 0.8013 - 359ms/epoch - 2ms/step
Epoch 5/5
168/168 - 0s - loss: 0.5042 - accuracy: 0.8162 - 410ms/epoch - 2ms/step
Out[452]:
<keras.callbacks.History at 0x1d265922290>
In [453]:
model.evaluate(x_test, y_test, batch_size=batch_size, verbose=2)

56/56 - 0s - loss: 0.4919 - accuracy: 0.8208 - 249ms/epoch - 4ms/step
Out[453]:
[0.49192091822624207, 0.8207573890686035]

GRU¶

In [454]:
model = keras.models.Sequential()
model.add(keras.Input(shape=(6, 1))) # seq_length, input_size
# model.add(layers.SimpleRNN(128,return_sequences=True, activation='relu')) # N, 28, 128
model.add(layers.GRU(128,return_sequences=False, activation='relu')) # N, 128

model.add(layers.Dense(10))

# loss and optimizer
loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
optim = keras.optimizers.Adam(learning_rate=0.001)
metrics = ['accuracy']

model.compile(loss=loss, optimizer=optim, metrics=metrics)

model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=2)

# evaluate
model.evaluate(x_test, y_test, batch_size=batch_size, verbose=2)

Epoch 1/5
168/168 - 2s - loss: 1.1838 - accuracy: 0.5154 - 2s/epoch - 11ms/step
Epoch 2/5
168/168 - 1s - loss: 0.7972 - accuracy: 0.6970 - 778ms/epoch - 5ms/step
Epoch 3/5
168/168 - 1s - loss: 0.6352 - accuracy: 0.7640 - 846ms/epoch - 5ms/step
Epoch 4/5
168/168 - 1s - loss: 0.5438 - accuracy: 0.8012 - 860ms/epoch - 5ms/step
Epoch 5/5
168/168 - 1s - loss: 0.4953 - accuracy: 0.8204 - 796ms/epoch - 5ms/step
56/56 - 0s - loss: 0.5216 - accuracy: 0.8098 - 337ms/epoch - 6ms/step
Out[454]:
[0.521604061126709, 0.8098176717758179]

LSTM¶

In [455]:
model = keras.models.Sequential()
model.add(keras.Input(shape=(6, 1))) # seq_length, input_size
# model.add(layers.SimpleRNN(128,return_sequences=True, activation='relu')) # N, 28, 128
model.add(layers.LSTM(128,return_sequences=False, activation='relu')) # N, 128

model.add(layers.Dense(10))

# loss and optimizer
loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
optim = keras.optimizers.Adam(learning_rate=0.001)
metrics = ['accuracy']

model.compile(loss=loss, optimizer=optim, metrics=metrics)

model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=2)

# evaluate
model.evaluate(x_test, y_test, batch_size=batch_size, verbose=2)

Epoch 1/5
168/168 - 2s - loss: 1.2294 - accuracy: 0.4914 - 2s/epoch - 11ms/step
Epoch 2/5
168/168 - 1s - loss: 0.9171 - accuracy: 0.6347 - 879ms/epoch - 5ms/step
Epoch 3/5
168/168 - 1s - loss: 0.7524 - accuracy: 0.7096 - 981ms/epoch - 6ms/step
Epoch 4/5
168/168 - 1s - loss: 0.6591 - accuracy: 0.7525 - 917ms/epoch - 5ms/step
Epoch 5/5
168/168 - 1s - loss: 0.6043 - accuracy: 0.7765 - 909ms/epoch - 5ms/step
56/56 - 0s - loss: 0.5745 - accuracy: 0.7835 - 334ms/epoch - 6ms/step
Out[455]:
[0.5744776725769043, 0.783450186252594]

Dense¶

In [456]:
model = tf.keras.Sequential([
    tf.keras.layers.Dense(16, activation='relu'), # if x <= 0, --> 0, x > 0 --> x
    tf.keras.layers.Dense(16, activation='relu'),
    tf.keras.layers.Dense(1, activation='sigmoid')
])
In [457]:
model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
             loss=tf.keras.losses.BinaryCrossentropy(),
             metrics=['accuracy'])
In [458]:
model.evaluate(X_train, y_train)

335/335 [==============================] - 0s 809us/step - loss: 8.8606 - accuracy: 0.2745
Out[458]:
[8.860629081726074, 0.2745208144187927]
In [459]:
model.evaluate(X_valid, y_valid)

112/112 [==============================] - 0s 1ms/step - loss: 8.8132 - accuracy: 0.2805
Out[459]:
[8.813237190246582, 0.28050491213798523]
In [460]:
model.fit(X_train, y_train, batch_size=16, epochs=20, validation_data=(X_valid, y_valid))

Epoch 1/20
669/669 [==============================] - 4s 6ms/step - loss: -235.1996 - accuracy: 0.2428 - val_loss: -1019.5363 - val_accuracy: 0.2496
Epoch 2/20
669/669 [==============================] - 1s 1ms/step - loss: -4164.0693 - accuracy: 0.2443 - val_loss: -8750.5283 - val_accuracy: 0.2415
Epoch 3/20
669/669 [==============================] - 1s 1ms/step - loss: -17297.6328 - accuracy: 0.2389 - val_loss: -27263.3320 - val_accuracy: 0.2384
Epoch 4/20
669/669 [==============================] - 1s 1ms/step - loss: -42248.3164 - accuracy: 0.2367 - val_loss: -58239.5977 - val_accuracy: 0.2359
Epoch 5/20
669/669 [==============================] - 1s 1ms/step - loss: -80429.0391 - accuracy: 0.2379 - val_loss: -102664.6016 - val_accuracy: 0.2432
Epoch 6/20
669/669 [==============================] - 1s 1ms/step - loss: -132948.6719 - accuracy: 0.2433 - val_loss: -161977.0156 - val_accuracy: 0.2457
Epoch 7/20
669/669 [==============================] - 1s 1ms/step - loss: -201032.5938 - accuracy: 0.2434 - val_loss: -237127.0938 - val_accuracy: 0.2446
Epoch 8/20
669/669 [==============================] - 1s 1ms/step - loss: -285669.6875 - accuracy: 0.2455 - val_loss: -328911.4688 - val_accuracy: 0.2435
Epoch 9/20
669/669 [==============================] - 1s 1ms/step - loss: -387987.5625 - accuracy: 0.2453 - val_loss: -438936.5000 - val_accuracy: 0.2435
Epoch 10/20
669/669 [==============================] - 1s 1ms/step - loss: -508355.7812 - accuracy: 0.2449 - val_loss: -566644.9375 - val_accuracy: 0.2424
Epoch 11/20
669/669 [==============================] - 1s 1ms/step - loss: -649278.3750 - accuracy: 0.2449 - val_loss: -715668.8750 - val_accuracy: 0.2449
Epoch 12/20
669/669 [==============================] - 1s 1ms/step - loss: -811134.1250 - accuracy: 0.2456 - val_loss: -885734.7500 - val_accuracy: 0.2496
Epoch 13/20
669/669 [==============================] - 1s 1ms/step - loss: -994793.5625 - accuracy: 0.2453 - val_loss: -1077977.1250 - val_accuracy: 0.2496
Epoch 14/20
669/669 [==============================] - 1s 1ms/step - loss: -1200795.3750 - accuracy: 0.2459 - val_loss: -1291846.5000 - val_accuracy: 0.2454
Epoch 15/20
669/669 [==============================] - 1s 1ms/step - loss: -1430086.2500 - accuracy: 0.2457 - val_loss: -1530435.3750 - val_accuracy: 0.2454
Epoch 16/20
669/669 [==============================] - 1s 1ms/step - loss: -1684317.0000 - accuracy: 0.2459 - val_loss: -1793811.7500 - val_accuracy: 0.2452
Epoch 17/20
669/669 [==============================] - 1s 1ms/step - loss: -1967338.5000 - accuracy: 0.2460 - val_loss: -2087851.1250 - val_accuracy: 0.2460
Epoch 18/20
669/669 [==============================] - 1s 1ms/step - loss: -2279071.5000 - accuracy: 0.2464 - val_loss: -2408506.0000 - val_accuracy: 0.2452
Epoch 19/20
669/669 [==============================] - 1s 1ms/step - loss: -2620246.5000 - accuracy: 0.2453 - val_loss: -2758454.7500 - val_accuracy: 0.2471
Epoch 20/20
669/669 [==============================] - 1s 1ms/step - loss: -2989409.2500 - accuracy: 0.2468 - val_loss: -3137055.0000 - val_accuracy: 0.2482
Out[460]:
<keras.callbacks.History at 0x1d26d9c8c10>
In [461]:
model.evaluate(X_test, y_test)

112/112 [==============================] - 0s 906us/step - loss: -3185349.7500 - accuracy: 0.2494
Out[461]:
[-3185349.75, 0.24936886131763458]
In [ ]:
In [ ]:
In [ ]:

testing_data_sets

In [467]:
testing_data_store = pd.read_csv('cut_testing_data_file/cut_testing_data.csv')
In [468]:
testing_data_store.head()
Out[468]:
ax ay az gx gy gz class_
0 -1.303614 -6.411490 4.077556 -0.421829 -0.733195 -0.274802 running
1 -0.998066 -6.549138 4.557804 -0.496127 -0.755675 -0.248011 running
2 -0.687327 -6.553696 4.994870 -0.511329 -0.774472 -0.243334 running
3 -0.329969 -6.566999 6.440067 -0.487907 -0.773408 -0.255150 running
4 -0.195836 -6.753056 6.938913 -0.338978 -0.745797 -0.273982 running
In [469]:
pd.api.types.is_string_dtype(testing_data_store['class_'])
Out[469]:
True
In [470]:
for label, content in testing_data_store.items():
    if pd.api.types.is_string_dtype(content):
        print(label)

class_
In [471]:
for label, content in testing_data_store.items():
    if pd.api.types.is_string_dtype(content):
        testing_data_store[label] = content.astype('category').cat.as_ordered()
In [472]:
testing_data_store['class_'].cat.codes
Out[472]:
0       2
1       2
2       2
3       2
4       2
       ..
3595    0
3596    0
3597    0
3598    0
3599    0
Length: 3600, dtype: int8
In [473]:
testing_data_store['class_'] = testing_data_store['class_'].cat.codes
In [477]:
X = testing_data_store.drop(['class_'], axis='columns')
y = np.ravel(testing_data_store['class_'])
In [478]:
X_train, X_temp, y_train, y_temp = train_test_split(X, y, test_size=0.4, random_state=0)
X_valid, X_test, y_valid, y_test = train_test_split(X_temp, y_temp, test_size=0.5, random_state=0)
In [479]:
x_train, y_train = X_train , y_train
x_test, y_test = X_test, y_test
batch_size = 64
epochs = 5
In [480]:
x_train.shape, y_train.shape
Out[480]:
((2160, 6), (2160,))

SimpleRNN¶

In [481]:
model = Sequential()
model.add(keras.Input(shape=(6,1)))
model.add(layers.SimpleRNN(128,return_sequences=False, activation='relu')) # N, 128
model.add(layers.Dense(10))


# loss and optimizer
loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
optim = keras.optimizers.Adam(learning_rate=0.001)
metrics = ['accuracy']

model.compile(loss=loss, optimizer=optim, metrics=metrics)

model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=2)

model.evaluate(x_test, y_test, batch_size=batch_size, verbose=2)

Epoch 1/5
34/34 - 1s - loss: 1.3755 - accuracy: 0.4579 - 689ms/epoch - 20ms/step
Epoch 2/5
34/34 - 0s - loss: 1.0177 - accuracy: 0.5944 - 101ms/epoch - 3ms/step
Epoch 3/5
34/34 - 0s - loss: 0.9299 - accuracy: 0.6514 - 92ms/epoch - 3ms/step
Epoch 4/5
34/34 - 0s - loss: 0.8803 - accuracy: 0.6662 - 110ms/epoch - 3ms/step
Epoch 5/5
34/34 - 0s - loss: 0.8263 - accuracy: 0.6972 - 92ms/epoch - 3ms/step
12/12 - 0s - loss: 0.8438 - accuracy: 0.6694 - 178ms/epoch - 15ms/step
Out[481]:
[0.8437779545783997, 0.6694444417953491]

GRU¶

In [482]:
model = keras.models.Sequential()
model.add(keras.Input(shape=(6, 1))) # seq_length, input_size
# model.add(layers.SimpleRNN(128,return_sequences=True, activation='relu')) # N, 28, 128
model.add(layers.GRU(128,return_sequences=False, activation='relu')) # N, 128

model.add(layers.Dense(10))

# loss and optimizer
loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
optim = keras.optimizers.Adam(learning_rate=0.001)
metrics = ['accuracy']

model.compile(loss=loss, optimizer=optim, metrics=metrics)

model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=2)

# evaluate
model.evaluate(x_test, y_test, batch_size=batch_size, verbose=2)

Epoch 1/5
34/34 - 1s - loss: 1.7664 - accuracy: 0.3546 - 1s/epoch - 37ms/step
Epoch 2/5
34/34 - 0s - loss: 1.1162 - accuracy: 0.4995 - 192ms/epoch - 6ms/step
Epoch 3/5
34/34 - 0s - loss: 0.9772 - accuracy: 0.5940 - 195ms/epoch - 6ms/step
Epoch 4/5
34/34 - 0s - loss: 0.9165 - accuracy: 0.6315 - 194ms/epoch - 6ms/step
Epoch 5/5
34/34 - 0s - loss: 0.8649 - accuracy: 0.6579 - 207ms/epoch - 6ms/step
12/12 - 0s - loss: 0.8343 - accuracy: 0.6625 - 245ms/epoch - 20ms/step
Out[482]:
[0.8342746496200562, 0.6625000238418579]

LSTM¶

In [483]:
model = keras.models.Sequential()
model.add(keras.Input(shape=(6, 1))) # seq_length, input_size
# model.add(layers.SimpleRNN(128,return_sequences=True, activation='relu')) # N, 28, 128
model.add(layers.LSTM(128,return_sequences=False, activation='relu')) # N, 128

model.add(layers.Dense(10))

# loss and optimizer
loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
optim = keras.optimizers.Adam(learning_rate=0.001)
metrics = ['accuracy']

model.compile(loss=loss, optimizer=optim, metrics=metrics)

model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=2)

# evaluate
model.evaluate(x_test, y_test, batch_size=batch_size, verbose=2)

Epoch 1/5
34/34 - 1s - loss: 1.6878 - accuracy: 0.3597 - 1s/epoch - 34ms/step
Epoch 2/5
34/34 - 0s - loss: 1.1969 - accuracy: 0.4644 - 213ms/epoch - 6ms/step
Epoch 3/5
34/34 - 0s - loss: 1.0142 - accuracy: 0.5681 - 212ms/epoch - 6ms/step
Epoch 4/5
34/34 - 0s - loss: 0.9779 - accuracy: 0.6014 - 220ms/epoch - 6ms/step
Epoch 5/5
34/34 - 0s - loss: 0.9279 - accuracy: 0.6352 - 242ms/epoch - 7ms/step
12/12 - 0s - loss: 0.9007 - accuracy: 0.6472 - 222ms/epoch - 19ms/step
Out[483]:
[0.9006998538970947, 0.6472222208976746]
In [ ]:
In [ ]:
In [ ]: