train.py

# -*- coding: utf-8 -*-
"""
Created on Sun Feb 16 22:30:12 2020

@author: Tanmay Thakur
"""
import pandas as pd
import numpy as np

from tensorflow.keras.utils import to_categorical
from model import *


data = pd.read_csv("data.csv")
dataset = np.array(data)
np.random.shuffle(dataset)
X = dataset
Y = dataset
X = X[:, 0:1024]
Y = Y[:, 1024]

image_x = 32
image_y = 32

Y = to_categorical(Y)
X = np.reshape(X, (X.shape[0], image_x, image_y, 1))

# Only uncomment if you want to retrain

# Vanilla model has optimal speed whereas separable conv model has optimal accuracy
# Introducing additional complexity causes mode collapse

"""
model_new = vanilla_conv(image_x, image_y)

model_new.fit(X, Y, batch_size = 32, epochs = 10, validation_split =  0.25)

model_new.save("vanilla_conv.h5")


model_sc = sc_conv(image_x, image_y)

model_sc.fit(X, Y, batch_size = 32, epochs = 10, validation_split =  0.25)

model_sc.save("sc_conv.h5")


model_complex = sc_conv_complex(image_x, image_y)

model_complex.fit(X, Y, batch_size = 32, epochs = 10, validation_split =  0.25)

model_complex.save("sc_conv_complex.h5")
"""