Commit 894ff63d authored by jagar2's avatar jagar2
Browse files

added my codes back

parent 53bcde5e
import os
import numpy as np
from scipy import (signal, io)
import h5py
import tensorflow as tf
from keras.models import Sequential
from keras.layers import (Dense, Conv1D, GRU, LSTM, Recurrent, Bidirectional, TimeDistributed,
Dropout, Flatten, RepeatVector, Reshape, MaxPooling1D, UpSampling1D)
def load_data(data_path, resample_size=256):
f = h5py.File(data_path, 'r')
loop_data = f['filt_AI_mat'][()]
X = np.rollaxis(loop_data.reshape(loop_data.shape[0], -1), 1)
X_resample = np.zeros((X.shape[0], 40 * resample_size))
for i in range(X.shape[0]):
X_resample[i] = signal.resample(X[i], 40 * resample_size)
X_resample = X_resample.reshape((-1, resample_size))
X_resample -= np.mean(X_resample) # TODO remove mean of each singal?
X_resample /= np.std(X_resample)
X_resample = np.atleast_3d(X_resample)
return X_resample
def get_run_id(sample, layer_type, size, num_layers, embedding, lr, drop_frac, batch_size, kernel_size, **kwargs):
run = (f"{sample}_{layer_type}{size:03d}_x{num_layers}_emb{embedding:03d}_{lr:1.0e}"
f"_drop{int(100 * drop_frac)}_batch{batch_size}").replace('e-', 'm')
if layer_type == 'conv':
run += f'_k{kernel_size}'
return run
def rnn_auto(layer, size, num_layers, embedding, n_step, drop_frac=0., bidirectional=True,
if bidirectional:
wrapper = Bidirectional
wrapper = lambda x: x
model = Sequential()
model.add(wrapper(layer(size, return_sequences=(num_layers > 1)),
input_shape=(n_step, 1)))
for i in range(1, num_layers):
model.add(wrapper(layer(size, return_sequences=(i < num_layers - 1), dropout=drop_frac)))
model.add(Dense(embedding, activation='linear', name='encoding'))
for i in range(num_layers):
model.add(wrapper(layer(size, return_sequences=True, dropout=drop_frac)))
model.add(TimeDistributed(Dense(1, activation='linear')))
return model
def conv_auto(size, num_layers, embedding, n_step, kernel_size, drop_frac=0., **kwargs):
"""TODO: batch norm?"""
model = Sequential()
model.add(Conv1D(size, kernel_size, padding='same', activation='relu',
input_shape=(n_step, 1)))
for i in range(1, num_layers):
model.add(Conv1D(size, kernel_size, padding='same', activation='relu'))
model.add(Dense(embedding, activation='linear', name='encoding'))
# model.add(Conv1D(embedding, kernel_size=1, activation='linear', name='encoding'))
for i in range(1, num_layers):
model.add(Conv1D(size, kernel_size, padding='same', activation='relu'))
model.add(Conv1D(size, kernel_size, padding='same', activation='relu'))
return model
def main(arg_dict=None):
from keras.optimizers import Adam
import shutil
from keras.callbacks import TensorBoard, ModelCheckpoint
from keras_tqdm import TQDMCallback
from argparse import ArgumentParser, Namespace
parser = ArgumentParser()
parser.add_argument("--data_path", type=str, default='data/cleaned_data.mat')
parser.add_argument("--size", type=int)
parser.add_argument("--num_layers", type=int)
parser.add_argument('--embedding', type=int)
parser.add_argument("--drop_frac", type=float, default=0.)
parser.add_argument("--batch_size", type=int, default=1024)
parser.add_argument("--n_cycles", type=int, default=256)
parser.add_argument("--epochs", type=int, default=10)
parser.add_argument("--lr", type=float)
parser.add_argument("--layer_type", type=str)
parser.add_argument("--N_train", type=int)
parser.add_argument("--kernel_size", type=int, default=5)
# parser.add_argument("--patience", type=int, default=20)
parser.add_argument('--bidirectional', dest='bidirectional', action='store_true')
parser.add_argument('--overwrite', dest='overwrite', action='store_true')
parser.add_argument('--log_dir', type=str, default='log')
parser.add_argument('--cache', type=bool, default=False)
parser.add_argument('--sample', type=str)
parser.set_defaults(bidirectional=True, overwrite=False)
args = parser.parse_args(None if arg_dict is None else []) # don't read argv if arg_dict present
if arg_dict: # merge additional arguments w/ defaults
args = Namespace(**{**args.__dict__, **arg_dict})
# TODO fix the problems with the NANs
if args.sample in ['PZT_2080_BEPS', 'PZT_Mixed_BEPS']:
data = io.matlab.loadmat('Data.mat')
if args.sample in ['PZT_2080_BEPS']:
X = data['Loopdata_caca'].reshape(-1, 64)
# Sets the nan values = 0 TODO improve with in interpolator
X[np.where(np.isnan(X))] = 0
# subtracts the mean and takes the standard deviation of the data
X -= np.mean(X)
X /= np.std(X)
X = np.atleast_3d(X)
elif args.sample in ['PZT_Mixed_BEPS']:
X = data['Loopdata_mixed'].reshape(-1, 96)
# Sets the nan values = 0 TODO improve with in interpolator
X[np.where(np.isnan(X))] = 0
# subtracts the mean and takes the standard deviation of the data
X -= np.mean(X)
X /= np.std(X)
X = np.atleast_3d(X)
elif args.sample in ['PZT_Mixed_GEPS']:
if not args.cache:
X = load_data(args.data_path, args.n_cycles)
X = np.load('data/X_256.npy')
if args.n_cycles != 256:
raise ValueError("256 hard coded")
raise ValueError('Name of file does not exist')
if args.N_train:
train = np.arange(args.N_train)
train = np.arange(len(X))
run = get_run_id(**args.__dict__)
log_dir = os.path.join(args.log_dir, run)
print("Logging to {}".format(os.path.abspath(log_dir)))
weights_path = os.path.join(log_dir, 'weights.h5')
if os.path.exists(weights_path):
if args.overwrite:
print(f"Overwriting {log_dir}")
shutil.rmtree(log_dir, ignore_errors=True)
raise ValueError("Model file already exists")
layer = {'lstm': LSTM, 'gru': GRU, 'conv': Conv1D}[args.layer_type]
if issubclass(layer, Recurrent):
model = rnn_auto(layer, args.size, args.num_layers, args.embedding, n_step=X.shape[1],
elif issubclass(layer, Conv1D):
#def conv_auto(size, num_layers, embedding, n_step, kernel_size, drop_frac=0., **kwargs):
model = conv_auto(args.size, args.num_layers, args.embedding, n_step=X.shape[1],
kernel_size=args.kernel_size, drop_frac=args.drop_frac)
raise NotImplementedError()
model.compile(Adam(, loss='mse')
history =[train], X[train], epochs=args.epochs, batch_size=args.batch_size,
TensorBoard(log_dir=log_dir, write_graph=False),
return X, model, history
if __name__ == '__main__':
X, model, history = main()
Pycroscopy's processing module
.. autosummary::
:toctree: _autosummary
from . import plotting_format
from .plotting_format import *
from . import cKPFM
from .cKPFM import *
from . import Autoencoder
from .Autoencoder import *
def fit_cKPFM_linear(bias_vec_r,resp_mat_vsr):
bias_vec_r : numpy array
bias vector for the read voltage
resp_mat_vsr : numpy array
array containing the DC response curves
fit_slope : numpy array
slope fit results
fit_yintercept : numpy array
y_intercept results
jCPD_mat : numpy array
junction contact potential
fit_slope = np.zeros(resp_mat_vsr.shape)
fit_yintercept = np.zeros(resp_mat_vsr.shape)
jCPD_mat = np.zeros(resp_mat_vsr.shape)
for row in range(resp_mat_vsr.shape[0]):
for col in range(resp_mat_vsr.shape[1]):
for vw in range(resp_mat_vsr.shape[2]):
idx = np.isfinite(resp_mat_vsr[col, row, vw, :])
fit_coeff = np.polyfit(
resp_mat_vsr[col, row, vw, idx], 1)
fit_slope[col, row, vw] = fit_coeff[0]
fit_yintercept[col, row, vw] = fit_coeff[1]
jCPD_mat[col, row, vw] = -fit_coeff[1] / fit_coeff[0]
return(fit_slope, fit_yintercept, jCPD_mat)
axes.linewidth: 1.5
xtick.major.size: 6
xtick.minor.size: 2
xtick.major.width: 1.5
xtick.minor.width: 1.5
ytick.major.size: 6
ytick.minor.size: 2
ytick.major.width: 1.5
ytick.minor.width: 1.5 DejaVu Sans
axes.labelweight: bold
axes.labelpad: 1
axes.labelsize: 12
xtick.major.pad: 1
ytick.major.pad: 1
xtick.labelsize: 12
ytick.labelsize: 12 True
ytick.right: True
xtick.direction: in
ytick.direction: in
image.interpolation: nearest
\ No newline at end of file
This diff is collapsed.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment