def test_predict():
client = CitrinationClient(environ['CITRINATION_API_KEY'],
environ['CITRINATION_SITE'])
inputs = [
{
"CHEMICAL_FORMULA": "AlCu"
},
]
resp = client.predict("betterdensitydemo", inputs)
prediction = resp['candidates'][0]['Density']
assert abs(prediction[0] - 5.786) < 0.1
def main():
# Set up client
client = CitrinationClient(os.environ["CITRINATION_API_KEY"],
'https://citrination.com')
# Read in experimental data
filename = "C:/Users/mvane/Documents/Skunkworks/BMG/Data/BMG_full_dataset_with_energies.csv"
exp_data = pd.read_csv(filename)
formula = exp_data['formula'].as_matrix()
energy = exp_data['PROPERTY: Nearest DFT Formation Energy (eV)'].as_matrix(
)
form = "formula"
property = "Property Nearest DFT Formation Energy"
# Convert formulas & energies to dictionary. Make a list of dictionaries.
input = []
for i in range(0, len(formula)):
input.append({form: formula[i], property: energy[i]})
# Make predictions of Tg, Tx, and Tl. (These will also contain many of the Magpie descriptors used to train the model)
model_num = "4416"
predictions = client.predict(model_num, input)
# Write all these predictions to json files with date and time to differentiate predictions.
# Folder path:
folder_out = "C:/Users/mvane/Documents/GitHub/better-glasses/predictions_output.csv"
first = True
# Make CSV writer.
with open(folder_out, 'w', newline='') as csvfile:
writer = csv.writer(csvfile)
# Go through every prediction that we made from the chemical formula
for p in predictions:
keys = p.all_keys()
# Make a header row with all of the keys listed
if first:
row = []
row.append('Formula')
for k in keys:
row.append(k)
row.append(str(k) + ' Uncertainty')
writer.writerow(row)
first = False
row = []
# Write the formula in the first column of the CSV. Formula will also show up in a later column.
row.append(p.get_value('formula').value)
# Write the value and uncertainty of each key to the CSV
for k in keys:
val = p.get_value(k).value
loss = p.get_value(k).loss
row.append(str(val))
row.append(str(loss))
writer.writerow(row)
csvfile.close()
def predict_from_pifs(view_id,
pifs,
predict,
condition={},
exclude=[],
client=None):
'''
predict properties using inputs from pifs. returns dataframe with actual and predicted values
------------------
view_id: dataview id containing model
pifs: list of pifs to predict
predict: list of properties to predict
condition: dict of conditions and values
exclude: properties in pif to exclude from inputs (besides properties to predict)
client: CitrinationClient instance
'''
if client is None:
client = CitrinationClient(os.environ['CITRINATION_API_KEY'],
'https://citrination.com')
ids = []
inputs = []
predict = predict
actuals = []
for pif in pifs:
pids = {p.name: p.value for p in pif.ids}
ids.append(pids)
props = {p.name: p.scalars for p in pif.properties}
props['formula'] = pif.chemical_formula
inp = {
'Property {}'.format(k): v
for (k, v) in props.items() if k not in predict + exclude
}
inp.update(condition)
inputs.append(inp)
actuals.append({k: v for (k, v) in props.items() if k in predict})
modelout = client.predict(view_id, inputs)
predictions = []
for r in modelout:
pred = {
'pred_{}'.format(p): r.get_value('Property {}'.format(p)).value
for p in predict
}
predictions.append(pred)
dicts = []
for i, a, p in zip(ids, actuals, predictions):
td = {**i, **a, **p}
dicts.append(td)
result = pd.DataFrame(dicts)
return result
class CitrinationSaxsClassifer(object):
"""A set of classifiers to be used on SAXS spectra"""
def __init__(self, address, api_key_file):
with open(api_key_file, "r") as g:
api_key = g.readline()
a_key = api_key.strip()
self.client = CitrinationClient(site = address, api_key=a_key)
def citrination_classify(self,sample_params):
"""
Parameters
----------
sample_params : ordered dictionary
ordered dictionary of floats representing features of test sample
Returns
-------
flags : dict
dictionary of (boolean,float) tuples,
where the first item is the flag
and the second is the probability,
for each of the potential scattering populations
"""
inputs = self.append_str_property(sample_params)
flags = OrderedDict()
resp = self.client.predict("24", inputs) # "24" is ID of dataview on Citrination
flags['unidentified'] = resp['candidates'][0]['Property unidentified']
flags['guinier_porod'] = resp['candidates'][0]['Property guinier_porod']
flags['spherical_normal'] = resp['candidates'][0]['Property spherical_normal']
flags['diffraction_peaks'] = resp['candidates'][0]['Property diffraction_peaks']
return flags
# helper function
def append_str_property(self, sample_params):
inputs = {}
for k,v in sample_params.items():
k = "Property " + k
inputs[k] = v
return inputs
def citrination_predict(self, populations, sample_params, q_I):
"""Apply self.models and self.scalers to sample_params.
Parameters
----------
sample_params : ordered dictionary
ordered dictionary of floats representing features of test sample
Returns
-------
flags : dict
dictionary of (boolean,float) tuples,
where the first item is the flag
and the second is the probability,
for each of the potential scattering populations
"""
features = self.append_str_property(sample_params)
params = OrderedDict.fromkeys(saxs_math.all_parameter_keys)
if populations['unidentified'][0] == '1':
# TODO: we could still use a fit to 'predict' I0_floor...
return params # all params are "None"
if populations['spherical_normal'][0] == '1' and populations['diffraction_peaks'][0] == '0':
resp = self.client.predict("27", features) # "27" is ID of dataview on Citrination
params['r0_sphere'] = resp['candidates'][0]['Property r0_sphere']
additional_features = saxs_math.spherical_normal_profile(q_I)
additional_features = self.append_str_property(additional_features)
ss_features = dict(features)
ss_features.update(additional_features)
resp = self.client.predict("28", ss_features)
params['sigma_sphere'] = resp['candidates'][0]['Property sigma_sphere']
if populations['guinier_porod'][0] == '1':
additional_features = saxs_math.guinier_porod_profile(q_I)
additional_features = self.append_str_property(additional_features)
rg_features = dict(features)
rg_features.update(additional_features)
resp =self.client.predict("29", rg_features)
params['rg_gp'] = resp['candidates'][0]['Property rg_gp']
return params
def test_predict():
client = CitrinationClient(environ['CITRINATION_API_KEY'], environ['CITRINATION_SITE'])
inputs = [{"CHEMICAL_FORMULA": "AlCu"}, ]
resp = client.predict("betterdensitydemo", inputs)
prediction = resp['candidates'][0]['Density']
assert abs(prediction[0] - 5.786) < 0.1