def _findConstant(savePath):
pathForDeterminingConstant = os.path.join(savePath,
"toDefineModelConstant")
# Let us determine the constant that goes with the chemical model, by simply defining a small network.
smallMasks = [np.array([[1, -1]])]
complexity = "simple"
useProtectionOnActivator = False
useEndoOnOutputs = True
useEndoOnInputs = False
generateTemplateNeuralNetwork(
pathForDeterminingConstant,
smallMasks,
complexity=complexity,
useProtectionOnActivator=useProtectionOnActivator,
useEndoOnOutputs=useEndoOnOutputs,
useEndoOnInputs=useEndoOnInputs)
parsedEquation, constants, nameDic = read_file(
pathForDeterminingConstant + "/equations.txt",
pathForDeterminingConstant + "/constants.txt")
KarrayA, stochio, maskA, maskComplementary = sparseParser(
parsedEquation, constants)
_, T0, C0, _ = setToUnits(constants, KarrayA, stochio)
constantList = [
0.9999999999999998, 0.1764705882352941, 1.0, 0.9999999999999998,
0.1764705882352941, 1.0, 0.9999999999999998, 0.1764705882352941, 1.0,
0.018823529411764708
]
constantList += [constantList[-1]]
enzymeInit = 5 * 10**(-7) / C0
activInit = 10**(-4) / C0
inhibInit = 10**(-4) / C0
return constantList, enzymeInit, activInit, inhibInit, C0
def load(directory_name="weightDir",
directory_for_network="",
useEndo=False,
complexity="simple",
useProtectionOnActivator=False,
useEndoOnOutputs=True,
useEndoOnInputs=False):
"""
Generate the biochemical equations from the weights of a deep neural network.
The weights should be stored in the format csv, with the following name:
weight_LayerNb
The weight dir should be a directory containing only weights files.
Bias should not be given in this file as we ignore it in the current algorithm.
"""
files = os.listdir(directory_name)
files = np.sort(files)
# Load the masks:
# We use 0.2 to discriminate the clipping values, as was done during training and validation
masks = []
for file in files:
if (file.split("_")[0] == "weight"):
df = pandas.read_csv(os.path.join(directory_name, file))
W = np.transpose(df.values[:, 1:])
WZeros = np.zeros((W.shape))
WPos = np.zeros((W.shape)) + 1
WNeg = np.zeros((W.shape)) + -1
masks += [
np.where(W < -0.2, WNeg, np.where(W > 0.2, WPos, WZeros))
]
# We generate the chemical network
# For constants we can either let the default value or give pre-defined values, for now use default value
if useEndo:
generateTemplateNeuralNetwork(
directory_for_network,
masks,
complexity=complexity,
useProtectionOnActivator=useProtectionOnActivator,
useEndoOnOutputs=useEndoOnOutputs,
useEndoOnInputs=useEndoOnInputs)
else:
generateTemplateNeuralNetwork(
directory_for_network,
masks,
complexity=complexity,
endoConstants=None,
useProtectionOnActivator=useProtectionOnActivator,
useEndoOnInputs=useEndoOnInputs,
useEndoOnOutputs=useEndoOnOutputs)
print("Generated network at " + str(directory_for_network))
return directory_for_network, masks
useDerivativeLeak = True
leak = 10**(-10)
layerInit = 10**(-13) #initial concentation value for species in layers
initValue = 10**(-13) #initial concentration value for all species.
enzymeInit = 5 * 10**(-7)
endoInit = 10**(-5) #only used if useEndo == True
activInit = 10**(-4)
inhibInit = 10**(-4)
if useEndo:
generateTemplateNeuralNetwork(
name,
masks,
complexity=complexity,
useProtectionOnActivator=useProtectionOnActivator,
useEndoOnOutputs=useEndoOnOutputs,
useEndoOnInputs=useEndoOnInputs)
else:
generateTemplateNeuralNetwork(
name,
masks,
complexity=complexity,
endoConstants=None,
useProtectionOnActivator=useProtectionOnActivator,
useEndoOnInputs=useEndoOnInputs,
useEndoOnOutputs=useEndoOnOutputs)
X1, X2, x_test, otherActivInitialC, otherInhibInitialC = _generate_initialConcentration_firstlayer(
activatorsOnObserved,