def train_rnn_tensorflow(window, data_h5):
import tensorflow as tf
# mf = pd.DataFrame(columns=['Window', 'lstm_dim', 'Test_score', 'Test_accuracy'])
df = misc2.create_dataframe_with_window(data_h5, window)
for n in range(df["Windowed_poses"].shape[0]):
df["Windowed_poses"].iloc[n] = np.hstack(
df["Windowed_poses"].iloc[n]).reshape((window, 14))
data = np.hstack(df["Windowed_poses"]).reshape(
(df["Windowed_poses"].shape[0], window, 14))
labels = df["Class"].values.flatten()
#Change if binary
#labels = labels.reshape((df["Class"].shape[0],1))
#labels = to_categorical(labels)
print("shape of data: ", data.shape)
print("shape of values: ", labels.shape)
print(data[0])
X_train, X_test, y_train, y_test = train_test_split(data,
labels,
test_size=0.2,
shuffle=False)
n_steps = window
n_inputs = 14
n_neurons = 150
n_outputs = 3
learning_rate = 0.001
X = tf.placeholder(tf.float32, [None, n_steps, n_inputs])
y = tf.placeholder(tf.int32, [None])
basic_cell = tf.contrib.rnn.BasicRNNCell(num_units=n_neurons)
outputs, states = tf.nn.dynamic_rnn(basic_cell, X, dtype=tf.float32)
logits = tf.layers.dense(states, n_outputs)
xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y,
logits=logits)
loss = tf.reduce_mean(xentropy)
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
training_op = optimizer.minimize(loss)
correct = tf.nn.in_top_k(logits, y, 1)
accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))
init = tf.global_variables_initializer()
n_epochs = 100
batch_size = 150
cc = 0
with tf.Session() as sess:
init.run()
for epoch in range(n_epochs):
for iteration in range(X_train.shape[0] // batch_size):
X_batch = X_train[cc:(cc + batch_size)]
y_batch = y_train[cc:(cc + batch_size)]
sess.run(training_op, feed_dict={X: X_batch, y: y_batch})
cc = cc + 150
acc_train = accuracy.eval(feed_dict={X: X_batch, y: y_batch})
acc_test = accuracy.eval(feed_dict={X: X_test, y: y_test})
print(epoch, "Train accuracy: ", acc_train, "Test_accuracy: ",
acc_test)
def lstm_train(window, data_h5):
#lstm_dim = [8,16,32,64,128,256]
lstm_dim = [16, 32]
mf = pd.DataFrame(
columns=['Window', 'lstm_dim', 'Test_score', 'Test_accuracy'])
df = misc2.create_dataframe_with_window(data_h5, window)
#print(df)
for n in range(df["Windowed_poses"].shape[0]):
df["Windowed_poses"].iloc[n] = np.hstack(
df["Windowed_poses"].iloc[n]).reshape((window, 14))
data = np.hstack(df["Windowed_poses"]).reshape(
(df["Windowed_poses"].shape[0], window, 14))
labels = df["Class"].values.flatten()
#Change if binary
#labels = labels.reshape((df["Class"].shape[0],1))
labels = to_categorical(labels, num_classes=3)
# my_labels = []
# for ii in range(labels.shape[0]):
# my_labels.append(labels[ii][0])
print("shape of data: ", data.shape)
print("shape of values: ", labels.shape)
X_train, X_test, y_train, y_test = train_test_split(data,
labels,
test_size=0.2,
shuffle=True,
random_state=42,
stratify=labels)
for l in lstm_dim:
model = Sequential()
#model.add(BatchNormalization(input_shape=(window,14)))
#lstm 128 works 88%
model.add(
Bidirectional(
LSTM(l,
input_shape=(window, 14),
activation='tanh',
return_sequences=False)))
#model.add(SimpleRNN(32, activation='relu', return_sequences=False))
#model.add(Dense(5,activation='tanh'))
model.add(Dense(labels.shape[1], activation='sigmoid'))
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
#early stop
es = EarlyStopping(monitor='val_acc', mode='max', verbose=0)
# 75 epochs seems the best one. with lstm 128
history = model.fit(X_train,
y_train,
batch_size=32,
epochs=128,
verbose=0,
shuffle=False,
validation_split=0.2,
callbacks=[es])
score, acc = model.evaluate(X_test, y_test, batch_size=1)
model.save(
'./keras_models/milti_class_classification_window_%s_lstm_%s_acc_%.2f.h5'
% (window, l, acc))
#y_pred = model.predict_classes(X_test, batch_size=1)
# confusion_mat = confusion_matrix(y_test, y_pred)
# plot_confusion_matrix(confusion_mat)
# target_names = ['Class-0','Class-1','Class-2']
# print(classification_report(y_test, y_pred, target_names=target_names))
mf = mf.append(
pd.DataFrame(
{
"Window": window,
"lstm_dim": l,
"Test_score": score,
"Test_accuracy": acc * 100
},
index=[0]))
return (mf, X_test, y_test)
def lstm_train_binary(window, data_h5):
lstm_dim = [64, 128, 256]
from sklearn import metrics
from sklearn.metrics import roc_auc_score
mf = pd.DataFrame(
columns=['Window', 'lstm_dim', 'Test_score', 'Test_accuracy'])
df = misc2.create_dataframe_with_window(data_h5, window)
## other approach
# df["Windowed_poses"] = pd.Series(df["Windowed_poses"]).apply(np.asarray)
# max_sequence_length = 14
# X_init = np.asarray(df.Windowed_poses)
# # Use hstack to and reshape to make the inputs a 3d vector
# X_t = np.hstack(X_init).reshape(df["Windowed_poses"].iloc[n],window,14)
# #y_t = np.hstack(np.asarray(original_data.Handeness)).reshape(264,)
#data = X_t
##
for n in range(df["Windowed_poses"].shape[0]):
df["Windowed_poses"].iloc[n] = np.hstack(
df["Windowed_poses"].iloc[n]).reshape((window, 14))
data = np.hstack(df["Windowed_poses"]).reshape(
(df["Windowed_poses"].shape[0], window, 14))
labels = df["Class"].values.flatten()
#Change if binary
#labels = labels.reshape((df["Class"].shape[0],1))
labels = to_categorical(labels)
print("shape of data: ", data.shape)
print("shape of values: ", labels.shape)
X_train, X_test, y_train, y_test = train_test_split(data,
labels,
test_size=0.2,
shuffle=True)
# class_weights = class_weight.compute_class_weight('balanced',
# np.unique(y_train),
# y_train[:,0])
for l in lstm_dim:
model = Sequential()
model.add(BatchNormalization(input_shape=(window, 14)))
#lstm 128 works 88%
model.add(
Bidirectional(
LSTM(l,
dropout=0.4,
recurrent_dropout=0.4,
activation='relu',
return_sequences=False)))
model.add(Dense(7, activation='sigmoid'))
model.add(Dense(labels.shape[1], activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
# 75 epochs seems the best one. with lstm 128
history = model.fit(X_train,
y_train,
batch_size=64,
epochs=64,
verbose=0,
shuffle=False,
validation_split=0.1)
score, acc = model.evaluate(X_test, y_test, batch_size=16)
probs = model.predict(X_test, batch_size=1)
#print(y_test.dtype)
roc = roc_auc_score(y_test, probs)
mf = mf.append(
pd.DataFrame(
{
"Window": window,
"lstm_dim": l,
"Test_score": acc,
"Test_accuracy": roc
},
index=[0]))
model.save(
'./keras_models/binary_classification_window_%s_lstm_%s_roc_%.2f.h5'
% (window, l, roc))
return (mf)