Conv1D and batch_size questions - keras

Question

I face two problems when I implement 1D convnet for multi-channel sequential data.
(224 samples x 300 time sequential x 19 channels)
1) I set batch_size as 7 but it jumps with 5 times of that.
not 7 14 21 28, but 7, 56, 105, 147... what's wrong with mine?
2) when I look at the records of accuracy, it looks like to learn NOTHING.
is it impossible implement classifier for multi-channel sequential data with Conv1D?
If possible can you give me some advice from my code?
#result
x_train shape: (224, 300, 19)
224 train samples
28 test samples
Train on 224 samples, validate on 28 samples
Epoch 1/50
7/224 [..............................] - ETA: 68s - loss: 0.6945 - acc: 0.5714
56/224 [======>.......................] - ETA: 6s - loss: 0.6993 - acc: 0.4464
105/224 [=============>................] - ETA: 2s - loss: 0.6979 - acc: 0.4381
147/224 [==================>...........] - ETA: 1s - loss: 0.6968 - acc: 0.4422
189/224 [========================>.....] - ETA: 0s - loss: 0.6953 - acc: 0.4444
224/224 [==============================] - 2s - loss: 0.6953 - acc: 0.4420 - val_loss: 0.6956 - val_acc: 0.5000
Epoch 2/50
7/224 [..............................] - ETA: 0s - loss: 0.6759 - acc: 0.5714
63/224 [=======>......................] - ETA: 0s - loss: 0.6924 - acc: 0.5556
133/224 [================>.............] - ETA: 0s - loss: 0.6905 - acc: 0.5338
203/224 [==========================>...] - ETA: 0s - loss: 0.6903 - acc: 0.5567
224/224 [==============================] - 0s - loss: 0.6923 - acc: 0.5357 - val_loss: 0.6968 - val_acc: 0.5000
# code
from __future__ import print_function
import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten, Activation
from keras.layers import Conv2D, MaxPooling2D, Conv1D, MaxPooling1D
import numpy as np
batch_size = 7
num_classes = 2
epochs = 50
# input data dimensions : 300 sequential x 19 channels
eeg_rows, num_ch = 300, 19
x_train = np.load('eeg_train.npy')
y_train = np.load('label_train.npy')
x_test = np.load('eeg_test.npy')
y_test = np.load('label_test.npy')
x_valid = np.load('eeg_valid.npy')
y_valid = np.load('label_valid.npy')
x_train = x_train.reshape(x_train.shape[0], eeg_rows, num_ch)
x_test = x_test.reshape(x_test.shape[0], eeg_rows,num_ch)
x_valid = x_valid.reshape(x_valid.shape[0], eeg_rows, num_ch)
input_shape = (eeg_rows, num_ch)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_valid = x_test.astype('float32')
x_train /= 100
x_test /= 100
x_valid /= 100
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
# model
conv = Sequential()
conv.add(Conv1D(32, 3, input_shape=input_shape, activation='relu', padding='same'))
conv.add(Conv1D(32, 3, activation='relu', padding='same'))
conv.add(MaxPooling1D(pool_size=2, strides=2, padding='same'))
conv.add(Dropout(0.2))
conv.add(Flatten())
conv.add(Dense(16, activation='relu'))
conv.add(Dropout(0.5))
conv.add(Dense(2, activation='softmax'))
conv.compile(loss='categorical_crossentropy',
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
# train
conv.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(x_valid, y_valid))
score = conv.evaluate(x_valid, y_valid, verbose=0)
print(conv.summary())
print(conv.input_shape)
print(conv.output_shape)
print('Test loss:', score[0])
print('Test accuracy:', score[1])