This python script is a part of the model development process.
It acts to instantiate and train 3
classification models which use Gradient Descent as the
learning algorithm, enabling mini-batch training.
"""
from ML_DataPipeline import DataPipeline
from sklearn.linear_model import SGDClassifier
import _pickle as pickle
labels = [0, 1]
epochs = 200
seed = 42
pipeline = DataPipeline("SGD", batch_size=96)
print("Support Vector Machine")
print("----------------------")
svm = SGDClassifier(verbose=0, random_state=seed)
class_flag = True
for epoch in range(epochs):
if epoch % 20 == 0:
print("Epoch: %d" % (epoch + 1))
if epoch == (epochs - 1):
print("Epoch: %d" % epochs)
iterations = pipeline.get_iterations()
for iters in range(iterations):
x_train, y_train = pipeline.get_training_batch(iters)
if class_flag:
#mc= ModelCheckpoint('yourdirectory/your_model.h5', monitor='val_loss',
#mode='min', verbose=1, save_best_only=True)
#cb_list=[es,mc]
# compile model
my_model.compile(optimizer='sgd',
loss='binary_crossentropy',
metrics=['accuracy'])
path = r"/Users/michaelperez/Desktop/Compilation/"
train_path = os.path.join(path, "train_images.pickle")
test_path = os.path.join(path, "test_images.pickle")
epochs = 1 # varied from 10-30 by 10
pipeline = DataPipeline("CNN",
batch_size=32,
train_file_path=train_path,
test_file_path=test_path)
for epoch in range(epochs):
if (epoch % 2) == 0:
print("Epoch %d" % (epoch + 1))
if epoch + 1 == epochs:
print("Epoch %d" % (epochs))
iterations = pipeline.get_iterations()
for iters in range(iterations):
x_train, y_train = pipeline.get_training_batch(iters)
my_model.fit(x_train, y_train, epochs=1, batch_size=32)
my_model.save(
'CNNClassifier_1_.h5') # creates a HDF5 file 'CNNClassifier_1_.h5'
import pandas as pd
import numpy as np
import os
#Insert the path to the image files on your system here
path = r"C:\Users\YourUserName\Path"
train_path = os.path.join(path, "train_images.pickle")
test_path = os.path.join(path, "test_images.pickle")
nb_path = os.path.join(path, r"MLFinalProject-master\ML_Models_Saved\NBClassifier.pickle")
svm_path = os.path.join(path, r"MLFinalProject-master\ML_Models_Saved\SVMClassifier_1.pickle")
with open(nb_path, "rb") as model :
nb_classifier = pickle.load(model)
pipeline = DataPipeline("NB", train_file_path=train_path, test_file_path=test_path)
partitions = pipeline.get_partitions()
random_batch = np.random.randint(0, partitions-1)
X_, y_ = pipeline.get_test_data(random_batch)
sample_image = X_[0].reshape((750, 750))
sample_label = y_[0]
y_hat = nb_classifier.predict([X_[0]])
accuracy = nb_classifier.score(X_, y_)
results = pd.DataFrame({"Prediction": y_hat, "Actual": y_[0]})
print("Project Marrow Beta Demo: Multinomial Naive Bayes")
print("-------------------------------------------------")
glob.glob(os.path.join(path, r"MLFinalProject\ML_Models_Saved\*.h5"))
models = dict()
accuracies = dict()
for path in model_paths:
model_name = path.rsplit("\\", 1)[1].split(".")[0]
if "pickle" in path:
with open(path, "rb") as model_file:
model = pickle.load(model_file)
if "h5" in path:
model = load_model(path, compile=False)
model.compile(optimizer="sgd", loss="binary_crossentropy", metrics=["accuracy"])
models[model_name] = model
accuracies[model_name] = list()
pipeline_CNN = DataPipeline("CNN", batch_size=24, partitions=12, train_file_path=train_path, test_file_path=test_path)
pipeline_SGD = DataPipeline("SGD", batch_size=24, partitions=12, train_file_path=train_path, test_file_path=test_path)
pipeline_NB = DataPipeline("NB", batch_size=24, partitions=12, train_file_path=train_path, test_file_path=test_path)
partitions = pipeline_CNN.get_partitions()
for partition in range(partitions):
X_test_CNN, y_test_CNN = pipeline_CNN.get_test_data(partition)
X_test_NB, y_test_NB = pipeline_NB.get_test_data(partition)
X_test_SGD, y_test_SGD = pipeline_SGD.get_test_data(partition)
print("Partition: %d" %partition)
for model_name, model in models.items():
if "CNN" in model_name:
score = model.test_on_batch(X_test_CNN, y_test_CNN)[1]
accuracies[model_name].append(score)
elif "NB" in model_name:
def main():
# Verify Pipeline for Training Data Works #
data_pipeline = DataPipeline()
print("Training Methods Test")
print("---------------------")
train_check = []
train_iterations = data_pipeline.get_iterations()
# Get data in one epoch
for iteration in range(train_iterations):
features, labels = data_pipeline.get_training_batch(iteration)
if len(features) == len(labels):
train_check.append(True)
print("Batch Dimensions Verification: %d" % (sum(train_check)))
print("Training Iterations per Epoch: %d" % (train_iterations))
if sum(train_check) == train_iterations: print("Test 1 --- Passed")
try:
data_pipeline = DataPipeline(train_file_path="DNE")
except OSError:
print("Test 2 --- Passed\n")
# Verify Pipeline for Testing Data Works #
data_pipeline = DataPipeline()
print("Testing Methods Test")
print("--------------------")
test_partitions = data_pipeline.get_partitions()
test_check = []
# Get all testing data
for partition in range(test_partitions):
features, labels = data_pipeline.get_test_data(partition)
if len(features) == len(labels):
test_check.append(True)
print("Testing Partitions: %d" % (test_partitions))
print("Partition Dimensions Verification: %d" % (sum(test_check)))
if sum(test_check) == test_partitions:
print("Test 1 --- Passed")
try:
data_pipeline = DataPipeline(test_file_path="NULL")
for partition in range(test_partitions):
features, labels = data_pipeline.get_test_data(partition)
if len(features) == len(labels):
test_check.append(True)
except OSError:
print("Test 2 --- Passed\n")
# Verify Edge Cases Handled #
print("Exception & Warning Edge Cases")
print("------------------------------")
print("Batch Size Tests")
try:
data_pipeline = DataPipeline(batch_size=32 + 0.5)
except ValueError:
print("Test 1 --- Passed")
data_pipeline = DataPipeline(batch_size=0)
print("Test 2 --- Passed")
data_pipeline = DataPipeline(batch_size=10000)
print("Test 3 --- Passed\n")
print("Partition Count Tests")
try:
data_pipeline = DataPipeline(partitions=10 + 0.5)
except ValueError:
print("Test 4 --- Passed")
data_pipeline = DataPipeline(partitions=0)
print("Test 5 --- Passed")
data_pipeline = DataPipeline(partitions=10000)
print("Test 6 --- Passed\n")