SMatejcek

smatejcek.py

@@ -0,0 +1,48 @@
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+from sklearn.linear_model import LogisticRegression
+from sklearn.metrics import confusion_matrix
+from sklearn.model_selection import train_test_split
+from pandas.plotting import scatter_matrix
+from itertools import combinations
+from sklearn.model_selection import cross_val_score
+
+df = pd.read_csv("grad.csv")
+
+print(df.describe().T)
+
+# No time to self-teach myself crosstbabing
+
+# This was no help at all
+scatter_matrix(df, figsize = (12, 12))
+plt.show()
+
+df.plot(kind = "bar")
+
+dependent = "admit"
+
+columns = [c for c in list(df) if c != dependent]
+results = []
+for i in range(1, len(columns) + 1):
+    combos = list(combinations(columns, i))
+    for c in combos:
+        y = df[dependent]
+        X = pd.DataFrame(df, columns = c) #Alternative assignment method
+        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.33, random_state = 42)
+        for j in range(2):
+            model = LogisticRegression(fit_intercept = False).fit(X_train, y_train)
+            score = cross_val_score(model, X, y, cv = 10, scoring = "accuracy").mean()
+            results.append([score, "Intercept: {}".format(j == 1), c, "Coef: {}".format(model.coef_), "Int: {}".format(model.intercept_)])
+
+print(max(results))
+
+# Best results for no Intercept == False
+print(max([r for r in results if r[1] == "Intercept: False"]))
+
+# Accuracy is .7023
+
+# No preference over sensitivy/recall and
+# precision.  Higher sensitivity, we're rejecting
+# qualified appicants at a higher rate but,
+# accepting fewer unqualified applicants.