def init_models(num_models, input_num):
global global_genes, global_num_nodes, current_pop
global_num_nodes = input_num + 1
for i in range(num_models):
model = Model()
# model.add(Dense(output_dim=3, input_dim=input_num))
# model.add(Activation("sigmoid"))
# model.add(Dense(output_dim=1))
# model.add(Activation("sigmoid"))
for j in range(input_num):
if len(global_genes) < input_num:
global_genes.append(Gene(j, input_num, j))
model.add(global_genes[j])
model.set_num_nodes(input_num+1)
# sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True) # stochastic gradient descent, (lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
# model.compile(loss="mse", optimizer=sgd, metrics=["accuracy"])
current_pop.append(model)
fitness.append(-100)
if load_current_pool:
current_pop = load_saved()
for i in range(num_models):
print(current_pop[i].get_weights())
class StructureBuilder:
"""
Deals with contructing the Structure object. The StructureBuilder class is used
by the PDBParser classes to translate a file to a Structure object.
"""
def __init__(self):
self.line_counter=0
self.header={}
def _is_completely_disordered(self, residue):
"Return 1 if all atoms in the residue have a non blank altloc."
atom_list=residue.get_unpacked_list()
for atom in atom_list:
altloc=atom.get_altloc()
if altloc==" ":
return 0
return 1
# Public methods called by the Parser classes
def set_header(self, header):
self.header=header
def set_line_counter(self, line_counter):
"""
The line counter keeps track of the line in the PDB file that
is being parsed.
Arguments:
o line_counter - int
"""
self.line_counter=line_counter
def init_structure(self, structure_id):
"""Initiate a new Structure object with given id.
Arguments:
o id - string
"""
self.structure=Structure(structure_id)
def init_model(self, model_id):
"""Initiate a new Model object with given id.
Arguments:
o id - int
"""
self.model=Model(model_id)
self.structure.add(self.model)
def init_chain(self, chain_id):
"""Initiate a new Chain object with given id.
Arguments:
o chain_id - string
"""
if self.model.has_id(chain_id):
self.chain=self.model[chain_id]
if __debug__:
warnings.warn("WARNING: Chain %s is discontinuous at line %i."
% (chain_id, self.line_counter),
PDBConstructionWarning)
else:
self.chain=Chain(chain_id)
self.model.add(self.chain)
def init_seg(self, segid):
"""Flag a change in segid.
Arguments:
o segid - string
"""
self.segid=segid
def init_residue(self, resname, field, resseq, icode):
"""
Initiate a new Residue object.
Arguments:
o resname - string, e.g. "ASN"
o field - hetero flag, "W" for waters, "H" for
hetero residues, otherwise blank.
o resseq - int, sequence identifier
o icode - string, insertion code
"""
if field!=" ":
if field=="H":
# The hetero field consists of H_ + the residue name (e.g. H_FUC)
field="H_"+resname
res_id=(field, resseq, icode)
if field==" ":
if self.chain.has_id(res_id):
# There already is a residue with the id (field, resseq, icode).
# This only makes sense in the case of a point mutation.
if __debug__:
warnings.warn("WARNING: Residue ('%s', %i, '%s') "
"redefined at line %i."
% (field, resseq, icode, self.line_counter),
PDBConstructionWarning)
duplicate_residue=self.chain[res_id]
if duplicate_residue.is_disordered()==2:
# The residue in the chain is a DisorderedResidue object.
# So just add the last Residue object.
if duplicate_residue.disordered_has_id(resname):
# The residue was already made
self.residue=duplicate_residue
duplicate_residue.disordered_select(resname)
else:
# Make a new residue and add it to the already
# present DisorderedResidue
new_residue=Residue(res_id, resname, self.segid)
duplicate_residue.disordered_add(new_residue)
self.residue=duplicate_residue
return
else:
# Make a new DisorderedResidue object and put all
# the Residue objects with the id (field, resseq, icode) in it.
# These residues each should have non-blank altlocs for all their atoms.
# If not, the PDB file probably contains an error.
if not self._is_completely_disordered(duplicate_residue):
# if this exception is ignored, a residue will be missing
self.residue=None
raise PDBConstructionException(\
"Blank altlocs in duplicate residue %s ('%s', %i, '%s')" \
% (resname, field, resseq, icode))
self.chain.detach_child(res_id)
new_residue=Residue(res_id, resname, self.segid)
disordered_residue=DisorderedResidue(res_id)
self.chain.add(disordered_residue)
disordered_residue.disordered_add(duplicate_residue)
disordered_residue.disordered_add(new_residue)
self.residue=disordered_residue
return
residue=Residue(res_id, resname, self.segid)
self.chain.add(residue)
self.residue=residue
def init_atom(self, name, coord, b_factor, occupancy, altloc, fullname,
serial_number=None, element=None):
"""
Initiate a new Atom object.
Arguments:
o name - string, atom name, e.g. CA, spaces should be stripped
o coord - Numeric array (Float0, size 3), atomic coordinates
o b_factor - float, B factor
o occupancy - float
o altloc - string, alternative location specifier
o fullname - string, atom name including spaces, e.g. " CA "
o element - string, upper case, e.g. "HG" for mercury
"""
residue=self.residue
# if residue is None, an exception was generated during
# the construction of the residue
if residue is None:
return
# First check if this atom is already present in the residue.
# If it is, it might be due to the fact that the two atoms have atom
# names that differ only in spaces (e.g. "CA.." and ".CA.",
# where the dots are spaces). If that is so, use all spaces
# in the atom name of the current atom.
if residue.has_id(name):
duplicate_atom=residue[name]
# atom name with spaces of duplicate atom
duplicate_fullname=duplicate_atom.get_fullname()
if duplicate_fullname!=fullname:
# name of current atom now includes spaces
name=fullname
if __debug__:
warnings.warn("WARNING: atom names %s and %s differ "
"only in spaces at line %i."
% (duplicate_fullname, fullname,
self.line_counter),
PDBConstructionWarning)
atom=self.atom=Atom(name, coord, b_factor, occupancy, altloc,
fullname, serial_number, element)
if altloc!=" ":
# The atom is disordered
if residue.has_id(name):
# Residue already contains this atom
duplicate_atom=residue[name]
if duplicate_atom.is_disordered()==2:
duplicate_atom.disordered_add(atom)
else:
# This is an error in the PDB file:
# a disordered atom is found with a blank altloc
# Detach the duplicate atom, and put it in a
# DisorderedAtom object together with the current
# atom.
residue.detach_child(name)
disordered_atom=DisorderedAtom(name)
residue.add(disordered_atom)
disordered_atom.disordered_add(atom)
disordered_atom.disordered_add(duplicate_atom)
residue.flag_disordered()
if __debug__:
warnings.warn("WARNING: disordered atom found "
"with blank altloc before line %i.\n"
% self.line_counter,
PDBConstructionWarning)
else:
# The residue does not contain this disordered atom
# so we create a new one.
disordered_atom=DisorderedAtom(name)
residue.add(disordered_atom)
# Add the real atom to the disordered atom, and the
# disordered atom to the residue
disordered_atom.disordered_add(atom)
residue.flag_disordered()
else:
# The atom is not disordered
residue.add(atom)
def set_anisou(self, anisou_array):
"Set anisotropic B factor of current Atom."
self.atom.set_anisou(anisou_array)
def set_siguij(self, siguij_array):
"Set standard deviation of anisotropic B factor of current Atom."
self.atom.set_siguij(siguij_array)
def set_sigatm(self, sigatm_array):
"Set standard deviation of atom position of current Atom."
self.atom.set_sigatm(sigatm_array)
def get_structure(self):
"Return the structure."
# first sort everything
# self.structure.sort()
# Add the header dict
self.structure.header=self.header
return self.structure
def set_symmetry(self, spacegroup, cell):
pass
class StructureBuilder:
"""
Deals with contructing the Structure object. The StructureBuilder class is used
by the PDBParser classes to translate a file to a Structure object.
"""
def __init__(self):
self.line_counter = 0
self.header = {}
def _is_completely_disordered(self, residue):
"Return 1 if all atoms in the residue have a non blanc altloc."
atom_list = residue.get_unpacked_list()
for atom in atom_list:
altloc = atom.get_altloc()
if altloc == " ":
return 0
return 1
# Public methods called by the Parser classes
def set_header(self, header):
self.header = header
def set_line_counter(self, line_counter):
"""
The line counter keeps track of the line in the PDB file that
is being parsed.
Arguments:
o line_counter - int
"""
self.line_counter = line_counter
def init_structure(self, structure_id):
"""Initiate a new Structure object with given id.
Arguments:
o id - string
"""
self.structure = Structure(structure_id)
def init_model(self, model_id):
"""Initiate a new Model object with given id.
Arguments:
o id - int
"""
self.model = Model(model_id)
self.structure.add(self.model)
def init_chain(self, chain_id):
"""Initiate a new Chain object with given id.
Arguments:
o chain_id - string
"""
if self.model.has_id(chain_id):
self.chain = self.model[chain_id]
if __debug__:
sys.stderr.write(
"WARNING: Chain %s is discontinuous at line %i.\n" %
(chain_id, self.line_counter))
else:
self.chain = Chain(chain_id)
self.model.add(self.chain)
def init_seg(self, segid):
"""Flag a change in segid.
Arguments:
o segid - string
"""
self.segid = segid
def init_residue(self, resname, field, resseq, icode):
"""
Initiate a new Residue object.
Arguments:
o resname - string, e.g. "ASN"
o field - hetero flag, "W" for waters, "H" for
hetero residues, otherwise blanc.
o resseq - int, sequence identifier
o icode - string, insertion code
"""
if field != " ":
if field == "H":
# The hetero field consists of H_ + the residue name (e.g. H_FUC)
field = "H_" + resname
res_id = (field, resseq, icode)
if field == " ":
if self.chain.has_id(res_id):
# There already is a residue with the id (field, resseq, icode).
# This only makes sense in the case of a point mutation.
if __debug__:
sys.stderr.write(
"WARNING: Residue ('%s', %i, '%s') redefined at line %i.\n"
% (field, resseq, icode, self.line_counter))
duplicate_residue = self.chain[res_id]
if duplicate_residue.is_disordered() == 2:
# The residue in the chain is a DisorderedResidue object.
# So just add the last Residue object.
if duplicate_residue.disordered_has_id(resname):
# The residue was already made
self.residue = duplicate_residue
duplicate_residue.disordered_select(resname)
else:
# Make a new residue and add it to the already
# present DisorderedResidue
new_residue = Residue(res_id, resname, self.segid)
duplicate_residue.disordered_add(new_residue)
self.residue = duplicate_residue
return
else:
# Make a new DisorderedResidue object and put all
# the Residue objects with the id (field, resseq, icode) in it.
# These residues each should have non-blanc altlocs for all their atoms.
# If not, the PDB file probably contains an error.
if not self._is_completely_disordered(duplicate_residue):
# if this exception is ignored, a residue will be missing
self.residue = None
raise PDBConstructionException(\
"Blank altlocs in duplicate residue %s ('%s', %i, '%s')" \
% (resname, field, resseq, icode))
self.chain.detach_child(res_id)
new_residue = Residue(res_id, resname, self.segid)
disordered_residue = DisorderedResidue(res_id)
self.chain.add(disordered_residue)
disordered_residue.disordered_add(duplicate_residue)
disordered_residue.disordered_add(new_residue)
self.residue = disordered_residue
return
residue = Residue(res_id, resname, self.segid)
self.chain.add(residue)
self.residue = residue
def init_atom(self,
name,
coord,
b_factor,
occupancy,
altloc,
fullname,
serial_number=None):
"""
Initiate a new Atom object.
Arguments:
o name - string, atom name, e.g. CA, spaces should be stripped
o coord - Numeric array (Float0, size 3), atomic coordinates
o b_factor - float, B factor
o occupancy - float
o altloc - string, alternative location specifier
o fullname - string, atom name including spaces, e.g. " CA "
"""
residue = self.residue
# if residue is None, an exception was generated during
# the construction of the residue
if residue is None:
return
# First check if this atom is already present in the residue.
# If it is, it might be due to the fact that the two atoms have atom
# names that differ only in spaces (e.g. "CA.." and ".CA.",
# where the dots are spaces). If that is so, use all spaces
# in the atom name of the current atom.
if residue.has_id(name):
duplicate_atom = residue[name]
# atom name with spaces of duplicate atom
duplicate_fullname = duplicate_atom.get_fullname()
if duplicate_fullname != fullname:
# name of current atom now includes spaces
name = fullname
if __debug__:
sys.stderr.write(
"WARNING: atom names %s and %s differ only in spaces at line %i.\n"
% (duplicate_fullname, fullname, self.line_counter))
atom = self.atom = Atom(name, coord, b_factor, occupancy, altloc,
fullname, serial_number)
if altloc != " ":
# The atom is disordered
if residue.has_id(name):
# Residue already contains this atom
duplicate_atom = residue[name]
if duplicate_atom.is_disordered() == 2:
duplicate_atom.disordered_add(atom)
else:
# This is an error in the PDB file:
# a disordered atom is found with a blanc altloc
# Detach the duplicate atom, and put it in a
# DisorderedAtom object together with the current
# atom.
residue.detach_child(name)
disordered_atom = DisorderedAtom(name)
residue.add(disordered_atom)
disordered_atom.disordered_add(atom)
disordered_atom.disordered_add(duplicate_atom)
residue.flag_disordered()
if __debug__:
sys.stderr.write(
"WARNING: disordered atom found with blanc altloc before line %i.\n"
% self.line_counter)
else:
# The residue does not contain this disordered atom
# so we create a new one.
disordered_atom = DisorderedAtom(name)
residue.add(disordered_atom)
# Add the real atom to the disordered atom, and the
# disordered atom to the residue
disordered_atom.disordered_add(atom)
residue.flag_disordered()
else:
# The atom is not disordered
residue.add(atom)
def set_anisou(self, anisou_array):
"Set anisotropic B factor of current Atom."
self.atom.set_anisou(anisou_array)
def set_siguij(self, siguij_array):
"Set standard deviation of anisotropic B factor of current Atom."
self.atom.set_siguij(siguij_array)
def set_sigatm(self, sigatm_array):
"Set standard deviation of atom position of current Atom."
self.atom.set_sigatm(sigatm_array)
def get_structure(self):
"Return the structure."
# first sort everything
# self.structure.sort()
# Add the header dict
self.structure.header = self.header
return self.structure
def set_symmetry(self, spacegroup, cell):
pass
from Layers import Dense_layer, Conv_layer, Pooling_layer, Dropout_layer
from Activation_Functions import ReLU, Softmax
from Loss_functions import Loss, Softmax, CategoricalCrossentropy, Act_Softmax_Loss_CCentropy
from Optimizers import SGD, Adam
from Model import Model, Accuracy, Accuracy_Categorical
# Exempel model
hugh = Model()
hugh.add(
Dense_layer(2,
512,
weight_regularizer_l2=0.0004,
bias_regularizer_l2=0.0004))
hugh.add(ReLU())
hugh.add(Dropout_layer(0.2))
hugh.add(Dense_layer(512, 3))
hugh.add(Softmax())
hugh.setters(loss=CategoricalCrossentropy(),
optimizer=Adam(learning_rate=0.05, decay=0.00005),
accuracy=Accuracy_Categorical())
hugh.finalize()
hugh.train(X_train,
y_train,
validation_data=(X_val, y_val),
epochs=1000,
print_every=100)
from Model import Model
from layers.Layers import Dense, Conv, Flatten
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("../data", one_hot=True)
x_train = np.reshape(mnist.train.images[:800], (800, 28, 28, 1))
y_train = mnist.train.labels[:800]
x_test = np.reshape(mnist.test.images[:1000], (1000, 28, 28, 1))
y_test = mnist.test.labels[:1000]
# print(x_train.shape)
model = Model()
model.add(Conv(16, (7, 7), (28, 28, 1), strides=(2, 2), padding="VALID", activation='tanh'))
model.add(Conv(32, (5, 5), (11, 11, 16), strides=(2, 2), padding="VALID", activation='tanh'))
# model.add(Conv(64, (3, 3), (4, 4, 32), strides=(1, 1), padding="VALID", activation='tanh'))
model.add(Flatten())
model.add(Dense((512, 64), activation='tanh', name='dense'))
model.add(Dense((64, 10), activation='softmax', name='dense2'))
# model.add(Dense((64, 1), activation='none', name='dense2'))
model.compile('cross_entropy') # cross_entropy
# model.compile('mse') # cross_entropy
model.fit_eval(x_train, y_train, 0.005, 20, x_test, y_test)
res = model.predict(x_test)
n = 0
for i, y in enumerate(y_test):
pred = np.argmax(res[i])
y_ = np.argmax(y)
# print("预测", pred, "真实", y_)
from Model import Model
model = Model("html")
body = Model("body")
body.add(Model("h1"))
body.add(Model("p"))
body.add(Model("div", "hello"))
model.add(body)
print(model.html())
input_shape = tuple(input_shape[:-1])
y = y.ravel()
if not num_classes:
num_classes = np.max(y) + 1
n = y.shape[0]
categorical = np.zeros((n, num_classes), dtype=dtype)
categorical[np.arange(n), y] = 1
output_shape = input_shape + (num_classes, )
categorical = np.reshape(categorical, output_shape)
return categorical
# Flatten the images to vectors
x_train = x_train.reshape(60000, 784)
x_test = x_test.reshape(10000, 784)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print(x_train.shape, 'train samples')
print(x_test.shape, 'test samples')
# convert class vectors to binary class matrices
y_train = to_categorical(y_train, num_classes)
y_test = to_categorical(y_test, num_classes)
model = Model()
model.add(Dense(10, input_shape=784))
model.train(x_train, y_train)
model.predict(x_test, y_test)
