def embedFig(target="IBD"):
#Embed
f = plt.figure(figsize=(15, 5))
X_train, X_val, X_test, y_train, y_val, y_test = hf.getMlInput(
otu_train,
otu_test,
map_train,
map_test,
target=target,
embed=True,
qual_vecs=qual_vecs)
X_train = pd.concat([X_train, X_val], axis=0)
y_train = y_train + y_val
plt.subplot(1, 2, 1)
m, auc_embed, auc_train_embed, fpr_embed, tpr_embed, prec_embed, f1_embed, f2_embed, _ = hf.predictIBD(
X_train,
y_train,
X_test,
y_test,
graph_title="Embedding weighted by averaging taxa " +
str(X_train.shape[1]) + " features",
max_depth=5,
n_estimators=95,
weight=20,
plot=True,
plot_pr=True)
f.savefig(os.path.join(fig_dir, "curves_AGP_test_embed.pdf"))
def pcaFig(target="IBD"):
f = plt.figure(figsize=(15, 5))
X_train, X_val, X_test, y_train, y_val, y_test = hf.getMlInput(
otu_train,
otu_test,
map_train,
map_test,
target=target,
pca_reduced=True,
numComponents=100)
X_train = pd.concat([X_train, X_val], axis=0)
y_train = y_train + y_val
plt.subplot(1, 2, 1)
m, auc_pca, auc_train_pca, fpr_pca, tpr_pca, prec_pca, f1_pca, f2_pca, _ = hf.predictIBD(
X_train,
y_train,
X_test,
y_test,
graph_title="PCA dimensionality reduced " + str(X_train.shape[1]) +
" features",
max_depth=5,
n_estimators=50,
weight=20,
plot=True,
plot_pr=True)
f.savefig(os.path.join(fig_dir + "curves_AGP_test_pca.pdf"))
def asinFig(target="IBD"):
#Normalize with asinh
f = plt.figure(figsize=(15, 5))
X_train, X_val, X_test, y_train, y_val, y_test = hf.getMlInput(
otu_train,
otu_test,
map_train,
map_test,
target=target,
asinNormalized=True)
X_train = pd.concat([X_train, X_val], axis=0)
y_train = y_train + y_val
plt.subplot(1, 2, 1)
m, auc_asin, auc_train_asin, fpr_asin, tpr_asin, prec_asin, f1_asin, f2_asin, _ = hf.predictIBD(
X_train,
y_train,
X_test,
y_test,
graph_title="Normalized asinh Taxa Abundances " +
str(X_train.shape[1]) + " features",
max_depth=5,
n_estimators=170,
weight=20,
plot=True,
plot_pr=True)
f.savefig(os.path.join(fig_dir, "asin_otu.pdf"))
# Classifying embedded data i.e. 113 features using Naive Bayes X_embed_train, X_embed_test, y_embed_train, y_embed_test = train_test_split(X_embed, y_embed, test_size = 0.2, random_state = 10) # Input data has negative values, MultinomialNB and ComplementNG cannot be used. clf = GaussianNB() model = clf.fit(X_embed_train, y_embed_train) predicted_y = model.predict(X_embed_test) f = plt.figure(figsize=(15,5)) roc_auc, fpr, tpr, average_precision, f1, f2 = hf.computeMLstats(model, X_embed_test, y_embed_test, plot=True, plot_pr=True, graph_title = "Naive Bayes Classifier on embedded data", flipped = False) f.savefig(os.path.join(fig_dir, "naive_bayes_classifier_embed.pdf")) # Classifying OTU data i.e. 26k+ features using Naive Bayes X_train, X_val, X_test, y_train, y_val, y_test = hf.getMlInput(otu_train, otu_test, map_train, map_test, target = "IBD", asinNormalized=True) X_train = pd.concat([X_train, X_val], axis = 0) y_train = y_train + y_val # Input data has negative values, MultinomialNB and ComplementNG cannot be used. clf = GaussianNB() model = clf.fit(X_train, y_train) predicted_y = model.predict(X_test) f = plt.figure(figsize=(15,5)) roc_auc, fpr, tpr, average_precision, f1, f2 = hf.computeMLstats(model, X_test, y_test, plot=True, plot_pr=True, graph_title = "Naive Bayes Classifier on OTU table", flipped = False) f.savefig(os.path.join(fig_dir, "naive_bayes_classifier_otu.pdf"))
weight=20,
plot=True,
plot_pr=True)
f.savefig(os.path.join(fig_dir + "curves_AGP_test_pca.pdf"))
pcaFig()
importlib.reload(hf)
target = "IBD"
# hf.getMlInput generates training, validation, and testing data
# if embed is True, the data is normalized using asinh i.e. hyperbolic inverse sin function
X_train, X_val, X_test, y_train, y_val, y_test = hf.getMlInput(
otu_train,
otu_test,
map_train,
map_test,
target=target,
embed=True,
qual_vecs=qual_vecs)
X = pd.concat([X_train, X_val, X_test], axis=0)
y = y_train + y_val + y_test
# These values are never used again (?)
auc_crossVal, auc_prec_crossVal, f1_crossVal, feat_imp_embed = hf.crossValPrediction(
X, y, max_depth=2, n_estimators=50, weight=20)
# weights/importance of 113 features (100 properties + 13 demographic features)
feat_imp_df = hf.getFeatImpDf(feat_imp_embed)
# mapping of property to function from kegg database
pathway_table = pd.read_csv(data_dir + "/property_pathway_dict.txt",