Use of 1D convolution across overlapping sliding windows of trading data. There is a time-dependency in the data which I want to accentuate

Model summary

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
conv1d_1 (Conv1D)            (None, 33, 50)            25050
_________________________________________________________________
max_pooling1d_1 (MaxPooling1 (None, 6, 50)             0
_________________________________________________________________
lstm_1 (LSTM)                (None, 30)                9720
_________________________________________________________________
dense_1 (Dense)              (None, 200)               6200
_________________________________________________________________
dropout_1 (Dropout)          (None, 200)               0
_________________________________________________________________
dense_2 (Dense)              (None, 200)               40200
_________________________________________________________________
dense_3 (Dense)              (None, 20)                4020
_________________________________________________________________
dense_4 (Dense)              (None, 1)                 21
=================================================================

Trading Conv1D

from __future__ import print_function
import keras
from keras.datasets import cifar10
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential
from keras.layers import Bidirectional, Dense, Dropout, Activation, Flatten, GaussianDropout,GaussianNoise
from keras.layers import Conv1D, Conv2D, Reshape, MaxPooling1D, LSTM
from keras.utils.np_utils import to_categorical
import numpy as np
from keras import regularizers

import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import numpy as np


# train ==  (4525, 34) last col is binary label
dataset_train = np.loadtxt("train.csv",delimiter=",",skiprows=1)

# test == (1507, 34) last col is binary label
dataset_test = np.loadtxt("test.csv",delimiter=",",skiprows=1)

# window size
ws = 100;

def windowize( dataset, ws ):
	
	recs, features = dataset.shape
	signals = np.ndarray(shape=(recs-ws,features-1,ws), dtype=float, order='F')
	# we want to compute (samplesize-ws) rectangular overlapping windows. 
	# this means we can not make predictions on the first ws records

	for i in range(0,recs-ws):
		signals[i, :, :] = dataset[ i:i+ws, :-1].transpose()

	labels = dataset[ws:,-1].astype(int)
	
	return (signals , labels)

train_x, train_y = windowize(dataset_train, ws)
test_x, test_y = windowize(dataset_test, ws)

# looking good?
plt.show( plt.imshow(train_x[100,:,:],interpolation="none") )

model = Sequential()
model.add(Conv1D(filters = 50, 
			activation = "relu",
			kernel_size = (5), 
			dilation_rate = 2, 
			kernel_regularizer=regularizers.l2(0.01),
			activity_regularizer=regularizers.l1(0.01),
			padding='same',
			input_shape=train_x.shape[1:]))
model.add(MaxPooling1D(pool_size=5))
model.add(LSTM(30))
model.add(Dense(200, activation="relu"))
model.add(Dropout(0.25))
model.add(Dense(200, activation="relu"))
model.add(Dense(20, activation="relu"))
model.add(Dense(1, activation="sigmoid"))

model.summary()

model.compile(loss='binary_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])

model.fit(train_x, 
	train_y, 
	nb_epoch=1000, 
	batch_size=15, 
	verbose=1, 
	validation_data=(test_x,test_y))

scores = model.evaluate(test_x, test_y)

model.save_weights(filepath="weights.w")
print("%s: %.2f%%" % (model.metrics_names[1], scores[1]*100))