def main(*args):
file_name = os.path.basename(os.path.splitext(__file__)[0])
time = datetime.datetime.now().isoformat()
description = f'redwine'
# description = 'TEST'
logger = Logger(log_dir=os.path.join(file_name, description, time))
logger.log_file(__file__)
print('log_dir', logger.log_dir)
# Red wine data
data = pd.read_csv('../data/winequality-red.csv')
X = np.asarray(data.drop(['quality'],
axis=1)) # Drop the labels for the X-data
# Adjust to quality scale between 0 and 6
y = np.asarray(data['quality'] - min(data['quality']), int)
# --- Model and loss ---
model = Model()
optimizer = tf.optimizers.Adam(0.001)
loss = tf.keras.losses.SparseCategoricalCrossentropy()
model.compile(optimizer=optimizer, loss=loss, metrics=['accuracy'])
# model.build((1, 2))
# print(model.summary())
print(*model.trainable_variables, sep='\n')
keep_best_callback = KeepBestCallback()
history: tf.keras.callbacks.History = model.fit(
X,
y,
validation_split=0.2,
epochs=1000,
callbacks=[keep_best_callback],
verbose=2)
print('Best', 'val_accuracy',
history.history['val_accuracy'][keep_best_callback.best_epoch])
print('Best', 'val_loss',
history.history['val_loss'][keep_best_callback.best_epoch])
def main(*args):
file_name = os.path.basename(os.path.splitext(__file__)[0])
time = datetime.datetime.now().isoformat()
data_idx = int(args[0])
# data_idx = 6
description = f'no_qc_16neurons_relu_data{data_idx}'
# description = 'TEST'
logger = Logger(log_dir=os.path.join(file_name, description, time))
logger.log_file(__file__)
print('log_dir', logger.log_dir)
with open(__file__) as this_file:
this_file = this_file.read()
# This data goes into C1 and T1
plain_x, labels = my_datasets.article_2003_09887_data(data_idx)
colors = ['blue' if label == 1 else 'red' for label in labels]
plain_x = normalize_data(plain_x)
plain_x = tf.convert_to_tensor(plain_x, dtype=float_type)
labels = tf.convert_to_tensor(labels, dtype=float_type)
# --- Feature extraction ---
features = tf.convert_to_tensor([
plain_x[..., 0],
plain_x[..., 1],
])
features = tf.transpose(features)
# --- Model and loss ---
model = U3_U()
optimizer = tf.optimizers.Adam(0.01)
# loss = P00MaximisationLoss()
model.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['accuracy'])
# model.build((1, 2))
# print(model.summary())
print(*model.trainable_variables, sep='\n')
keep_best_callback = KeepBestCallback()
history = model.fit(features, labels,
validation_split=0.2,
epochs=100,
callbacks=[keep_best_callback],
use_multiprocessing=True)
# --- Test result ---
model_fit = model
model_fit(features[:2,...]) # Phony init of model
model_fit.set_weights(keep_best_callback.best_weights)
for i in range(len(model_fit.weights)):
model_fit.weights[i].assign(keep_best_callback.best_weights[i])
# out = model_fit(features)
# x = plain_x.numpy()
# c_out = out.numpy().flatten()
# # labels = labels.numpy()
# fig = plt.figure()
# plt.title(f'{description}')
# plt.scatter(x[:, 0], x[:, 1], c=c_out, cmap=plt.get_cmap('bwr'))
# plt.show()
X = plain_x.numpy()
lab = labels.numpy()
int_labels = np.array(lab>.5, int)
fig = plt.figure(figsize=(5, 5))
plot_decision_regions(X=X, y=int_labels, clf=model_fit, legend=2)
train_labels = model_fit.predict(X, use_multiprocessing=True)
plt.title(description)
plt.show()
logger.log_variables(fig, 'fig',
X, 'X',
int_labels, 'y',
train_labels, 'train_labels',
keep_best_callback.best_weights, 'weights',
history.history, 'history')
logger.log_figure(fig, 'fig.pdf')
import tensorflow as tf
import os
import tensorflow as tf
import numpy as np
import datetime
from logtools import Logger
π = np.pi
file_name = os.path.basename(os.path.splitext(__file__)[0])
time = datetime.datetime.now().isoformat()
description = 'MNIST_intermetdiate_layer'
# description = 'TEST'
logger = Logger(log_dir=os.path.join(file_name, description, time))
logger.log_file(__file__)
print('log_dir', logger.log_dir)
# sHHHH = tensor(4*[H]) @ s0000
class NN(tf.keras.Sequential):
def __init__(self):
input_dense = tf.keras.layers.Dense(10, name='input_linear_combi')
# output_dense = tf.keras.layers.Dense(10, name='mnist')
super(NN, self).__init__(layers=[
tf.keras.layers.Flatten(),
input_dense,
# QC(), # This works as an activation function
# output_dense
])
def main(data_idx):
# Data from -1 to 1
data_generator = my_datasets.complex_1D_functions
data_index = data_idx
x, y = data_generator(data_index)
file_name = os.path.basename(os.path.splitext(__file__)[0])
time = datetime.datetime.now().isoformat()
description = f'{data_generator.__name__}({data_index})_smart_iSWAP_rxzx_6'
# description = 'TEST_U_bias'
logger = Logger(log_dir=os.path.join(file_name, description, time))
logger.log_file(__file__)
print('log_dir', logger.log_dir)
# --- Feature extraction ---
def calc_features(x):
features = tf.convert_to_tensor([ # Features as descried in the article
np.arccos(x**2),
np.arcsin(x),
])
return tf.transpose(features) # (batch, feature)
features = calc_features(x)
# --- Model and loss ---
model = Model()
optimizer = tf.optimizers.Adam(0.001)
model.compile(optimizer=optimizer, loss='mean_squared_error')
print(*model.trainable_variables, sep='\n')
keep_best_callback = KeepBestCallback()
early_stopping_callback = tf.keras.callbacks.EarlyStopping(min_delta=0.00001, patience=100)
history = model.fit(features, y,
validation_split=0.2,
epochs=10000,
callbacks=[keep_best_callback],
use_multiprocessing=True,
verbose=2)
# --- Test result ---
model_fit = model
model_fit(features[:10,...]) # Phony init of model
model_fit.set_weights(keep_best_callback.best_weights)
for i in range(len(model_fit.weights)):
model_fit.weights[i].assign(keep_best_callback.best_weights[i])
fig = plt.figure()
x_fit = np.linspace(-1, 1, 1000)
features_fit = calc_features(x_fit)
y_fit = model_fit.predict(features_fit)
plt.plot(x, y, '.b', label='Data')
plt.plot(x_fit, y_fit, '-r', label='Fit')
plt.title(description)
plt.show()
# X = f.numpy()
# lab = labels.numpy()
# int_labels = np.array(lab>.5, int)
# fig = plt.figure(figsize=(5, 5))
# plot_decision_regions(X=X, y=int_labels, clf=model_fit, legend=2)
# train_labels = model_fit.predict(X, use_multiprocessing=True)
# plt.title(description)
# plt.show()
logger.log_variables(fig, 'fig',
x, 'X',
y, 'y',
keep_best_callback.best_weights, 'weights',
history.history, 'history')
logger.log_text(f'best_val_loss = {keep_best_callback.best_val_loss}', 'best_val_loss.txt')
logger.log_text(f'{keep_best_callback.best_weights}', 'weights.txt')
logger.log_figure(fig, 'fig.pdf')