pres start

distances.py

@@ -0,0 +1,69 @@
+from sklearn.feature_extraction.text import CountVectorizer
+from sklearn.neighbors import NearestNeighbors
+import pandas as pd
+
+# toying around
+
+df = pd.read_csv('data/candy.csv')
+df = df[df['user'].isin(df['user'].sample(10))]
+df = df.pivot(index='item', columns='user', values='review').reset_index()
+df.head(1)
+
+df = df.melt(id_vars='item', var_name='user', value_name='review')
+df = df.dropna().reset_index(drop=True)
+df
+
+# data shaping
+
+df = pd.DataFrame([
+    [0, 1, 1, 0, 0, 0],
+    [0, 1, 1, 1, 0, 0],
+    [1, 0, 0, 1, 0, 0],
+    [0, 1, 1, 0, 0, 1],
+    [0, 0, 0, 1, 1, 1]
+])
+
+euclidean_distances(df)
+
+##
+
+
+df = pd.read_csv("data/candy.csv")
+
+df = df.groupby(["user"])["repo"].apply(lambda x: ",".join(x))
+df = pd.DataFrame(df)
+
+class NNRecommender:
+    def __init__(
+        self, n_neighbors=10, max_features=1000, tokenizer=lambda x: x.split(",")
+    ):
+        self.cv = CountVectorizer(tokenizer=tokenizer, max_features=max_features)
+        self.nn = NearestNeighbors(n_neighbors=n_neighbors)
+
+    def fit(self, X):
+        self.X = X
+        X = self.cv.fit_transform(X)
+        self.nn.fit(X)
+        return self
+
+    def predict(self, X):
+        Xp = []
+        for Xi in X:
+            Xt = self.cv.transform([Xi])
+            _, neighbors = self.nn.kneighbors(Xt)
+            repos = []
+            for n in neighbors[0]:
+                r = self.X.iloc[int(n)].split(",")
+                repos.extend(r)
+            repos = list(set(repos))
+            repos = [r for r in repos if r not in Xi.split(",")]
+            Xp.append(repos)
+        return Xp
+
+n_neighbors = 10
+max_features = 1000
+model = NNRecommender(n_neighbors, max_features)
+model.fit(df["repo"])
+
+with open("model/model.pkl", "wb") as f:
+    dill.dump(model, f)