Upload oxide_vacancy

.gitattributes

@@ -47,3 +47,4 @@ model_steel_yield/model.dill filter=lfs diff=lfs merge=lfs -text
 model_semiconductor_lvls/model.dill filter=lfs diff=lfs merge=lfs -text
 model_perovskite_Opband/model.dill filter=lfs diff=lfs merge=lfs -text
 model_perovskite_ASR/model.dill filter=lfs diff=lfs merge=lfs -text
+model_oxide_vacancy/model.dill filter=lfs diff=lfs merge=lfs -text

model_oxide_vacancy/.gitattributes

@@ -0,0 +1,2 @@
+# Auto detect text files and perform LF normalization
+* text=auto

model_oxide_vacancy/README.md

@@ -0,0 +1,2 @@
+# model_oxide_vacancy
+Random forest model to predict the formation energy of O vacancies in oxides

model_oxide_vacancy/RandomForestRegressor.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e98a5f48973e2e29a68f488422ea7cec8c20560ebbce7ac20fe36d095108f594
+size 59635443

model_oxide_vacancy/StandardScaler.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:345d92d75ec0456b756b4754fced973ecd78f2e762db2aa92bdd5412cdf73b33
+size 2381

model_oxide_vacancy/model.dill

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d45c7fe032a383352e6f428df7843eed7208e5371f7c6f01c8e52affc490411e
+size 93007744

model_oxide_vacancy/predict_oxide_vacancy.py

@@ -0,0 +1,83 @@
+import os
+import pandas as pd
+import numpy as np
+import joblib
+import dill
+from mastml.feature_generators import ElementalFeatureGenerator, OneHotGroupGenerator
+
+def get_preds_ebars_domains(df_test):
+    d = 'model_oxide_vacancy'
+    scaler = joblib.load(os.path.join(d, 'StandardScaler.pkl'))
+    model = joblib.load(os.path.join(d, 'RandomForestRegressor.pkl'))
+    df_features = pd.read_csv(os.path.join(d, 'X_train.csv'))
+    recal_params = pd.read_csv(os.path.join(d, 'recal_dict.csv'))
+
+    features = df_features.columns.tolist()
+    df_test = df_test[features]
+
+    X = scaler.transform(df_test)
+
+    # Make predictions
+    preds = model.predict(X)
+
+    # Get ebars and recalibrate them
+    errs_list = list()
+    a = recal_params['a'][0]
+    b = recal_params['b'][0]
+    c = recal_params['c'][0]
+    for i, x in X.iterrows():
+        preds_list = list()
+        for pred in model.model.estimators_:
+            preds_list.append(pred.predict(np.array(x).reshape(1, -1))[0])
+        errs_list.append(np.std(preds_list))
+    ebars = a * np.array(errs_list)**2 + b * np.array(errs_list) + c
+
+    # Get domains
+    with open(os.path.join(d, 'model.dill'), 'rb') as f:
+        model_domain = dill.load(f)
+
+    domains = model_domain.predict(X)
+
+    return preds, ebars, domains
+
+def process_data(comp_list):
+    X = pd.DataFrame(np.empty((len(comp_list),)))
+    y = pd.DataFrame(np.empty((len(comp_list),)))
+
+    df_test = pd.DataFrame({'Material composition': comp_list})
+
+    # Try this both ways depending on mastml version used.
+    try:
+        X, y = ElementalFeatureGenerator(composition_df=df_test['Material composition'],
+                                    feature_types=['composition_avg', 'arithmetic_avg', 'max', 'min','difference'],
+                                    remove_constant_columns=False).evaluate(X=X, y=y, savepath=os.getcwd(), make_new_dir=False)
+    except:
+        X, y = ElementalFeatureGenerator(featurize_df=df_test['Material composition'],
+                                         feature_types=['composition_avg', 'arithmetic_avg', 'max', 'min',
+                                                        'difference'], remove_constant_columns=False).evaluate(X=X, y=y, savepath=os.getcwd(), make_new_dir=False)
+
+    df_test = pd.concat([df_test, X], axis=1)
+
+    return df_test
+
+def make_predictions(comp_list):
+
+    # Process data
+    df_test = process_data(comp_list)
+
+    # Get the ML predicted values
+    preds, ebars, domains = get_preds_ebars_domains(df_test)
+
+    pred_dict = {'Compositions': comp_list,
+                 'Predicted O vacancy formation energy (eV)': preds,
+                 'Ebar O vacancy formation energy (eV)': ebars}
+
+    for d in domains.columns.tolist():
+        pred_dict[d] = domains[d]
+
+    del pred_dict['y_pred']
+    #del pred_dict['d_pred']
+    del pred_dict['y_stdu_pred']
+    del pred_dict['y_stdc_pred']
+
+    return pd.DataFrame(pred_dict)

model_oxide_vacancy/recal_dict.csv

@@ -0,0 +1,2 @@
+a,b,c
+0.07968009200757614,0.6292581403715423,0.0007229036713515384

model_oxide_vacancy/requirements.txt

@@ -0,0 +1,5 @@
+scikit-learn
+numpy 
+pandas
+mastml
+pymatgen