【Python、Flask】自分の服を選んでくれるアプリを作ろうとした時の話（Webアプリ編）

1
@app.route("/recommend", methods=["GET", "POST"])
2
@login_required
3
def recommend():
4
    """Recommend user's best cloth today"""
5
    # User reached route via GET (as by clicking a link or via redirect)
6
    if not request.method == "POST":
7
        pass
8
    # User reached route via POST (as by submitting a form via POST)
9
    else:
10
        pass
11
    # DB connection and cursor generation
12
    connection = sqlite3.connect(dbpath)
13
    # connection.isolation_level = None # 自動コミットにする場合は下記を指定（コメントアウトを解除のこと）
14
    cursor = connection.cursor()
15

16
    # Prepare csv file to Machine Learning --------------------------------------
17
    # Get wardrobe-image url to display
18
    cursor.execute("SELECT * FROM history_wear WHERE userid = :userid AND NOT comfort_score = '';",
19
                    {"userid": session["user_id"]
20
                    })
21
    rows_hw = cursor.fetchall() # history_wear row
22

23
    # Generate list of "history_wear" (wardrobeid is converted to warmscore AND wear_date is removed)
24
    list_hw  = [] # List about a record of history_wear
25
    list_hws = [] # List about records of history_wear
26
    for num in range(len(rows_hw)):
27
        for i in ['wardrobeid_c', 'wardrobeid_o', 'wardrobeid_t', 'wardrobeid_i', 'wardrobeid_b', 'wardrobeid_s']:
28
            # Select DB same as wardrobe-id.
29
            cursor.execute("SELECT * FROM wardrobes WHERE id = :wardrobeid;",
30
                            {"wardrobeid": rows_hw[num][column_history_wear[i]]
31
                            })
32
            rows_wr = cursor.fetchall() # wardorbes row
33
            if rows_wr == []: # If rows_wr is empty,
34
                warmscore = 0
35
            else:
36
                list_wr = [] # wardorbes list あとで使うかも
37
                list_wr.append(rows_wr) # あとで使うかも
38
                warmscore = rows_wr[0][column_wardrobes['warmscore']] # Get warmscore
39
            list_hw.append(warmscore) # Store in a list
40
        for i in ['temperature_max', 'temperature_min', 'comfort_score']:
41
            list_hw.append(round(rows_hw[num][column_history_wear[i]], 0))
42
        list_hws.append(list_hw) # Store in a list
43
        list_hw = [] # Set default value
44

45
    # Part of Machine Learning -----------------------------------------
46
    # Convert list of "history_wear" to pd.DataFrame of "history_wear"
47
    feature_names = ["warmscore_c", "warmscore_o", "warmscore_t",
48
                     "warmscore_i", "warmscore_b", "warmscore_s",
49
                     "temperature_max", "temperature_min", "comfort_score"]
50
    data = pd.DataFrame(list_hws, columns=feature_names)
51
    model = model_learning(data)
52

53
    # Part of Machine Predicting -------------------------------------------
54
    # Select DB for wardrobes
55
    cursor.execute("SELECT * FROM wardrobes WHERE userid = :t_userid AND inuse = 1;",
56
                    {"t_userid": session["user_id"]
57
                    })
58
    rows_wr = cursor.fetchall()
59
    dict_category = category_dictionary() # Dictionary about wardrobe's category
60
    dict_wr_category = {"-Outer Category-": [], "-Tops Category-": [],
61
                        "-Inner Category-": [], "-Bottoms Category-": []}
62
    combi_content = {} # wardrobe-combination on category
63
    combi_score   = {} # score on wardrobe-combination on category
64
    num           = 0  # for index of "combi_content"
65
    iter_num_o    = 0  # for iterate like MECE for '-Outer Category-'
66
    iter_num_b    = 0  # for iterate like MECE for '-Bottoms Category-'
67
    __len_combi__ = 20000
68
    for i in range(__len_combi__):
69
        combi_content[i] = {}
70
    # Generate dictionary on category
71
    for i in rows_wr:
72
        for key, value in dict_category.items():
73
            if i[column_wardrobes['category']] in value: # If category is same,
74
                dict_wr_category[key].append(i[column_wardrobes['id']])
75
                break
76
            else:
77
                pass
78
    # Append NULL to dictionary on category
79
    for key in dict_wr_category.keys():
80
        dict_wr_category[key].append(None)
81
    # Generate wardrobe-combination on category
82
    len_o = len(dict_wr_category['-Outer Category-'])   # Length of dictionary
83
    len_b = len(dict_wr_category['-Bottoms Category-']) # Length of dictionary
84
    for c in range(0, len_o):
85
        for o in range(iter_num_o, len_o):
86
            bool_co = c == len_o - 1 and o == len_o - 1
87
            if bool_co or c != o : # Same wardrobes cannot be combi except both are None
88
                for t in dict_wr_category['-Tops Category-']:
89
                    for i in dict_wr_category['-Inner Category-']:
90
                        iter_num_b = 0 # Reset start of counting
91
                        for b in range(0, len_b):
92
                            for s in range(iter_num_b, len_b):
93
                                bool_bs = b == len_b - 1 and s == len_b - 1
94
                                if b != s or bool_bs: # Same wardrobes cannot be combi except both are None
95
                                    combi_content[num]['wardrobeid_c'] = dict_wr_category['-Outer Category-'][c]
96
                                    combi_content[num]['wardrobeid_o'] = dict_wr_category['-Outer Category-'][o]
97
                                    combi_content[num]['wardrobeid_t'] = t
98
                                    combi_content[num]['wardrobeid_i'] = i
99
                                    combi_content[num]['wardrobeid_b'] = dict_wr_category['-Bottoms Category-'][b]
100
                                    combi_content[num]['wardrobeid_s'] = dict_wr_category['-Bottoms Category-'][s]
101
                                    num += 1 # Count dictionary-length
102
                                    if num >= __len_combi__:
103
                                        combi_content[num] = {}
104
                                    else:
105
                                        pass
106
                            iter_num_b += 1 # Shift start of counting
107
        iter_num_o += 1 # Shift start of counting
108

109
    # Generate predicted values on wardrobe-combination
110
    __best_score__ = 5 # Best comfort score
111
    for i in range(num):
112
        combi_score[i] = [0,0,0,0,0,0] # Set default value
113
    for key, value in combi_content.items(): # Search in the dictionaries about wardrobe-combination
114
        for k, v in value.items():           # Search in the dictionary about wardrobe-combination
115
            for i in rows_wr:                # Search in the list about wardrobes-DB
116
                if v == i[column_wardrobes['id']]: # If two wardrobe-id are same
117
                    combi_score[key][column_history_wear[k] - column_history_wear['wardrobeid_c']] = i[column_wardrobes['warmscore']] # Store warmscore
118
                    break
119
    # Append temperature
120
    location = session["user_location"] # Get user's location
121
    cursor.execute("SELECT * FROM weather_today WHERE location = :t_location;",
122
                    {'t_location': location})
123
    rows_weather = cursor.fetchall()
124
    for key, value in combi_score.items():
125
        combi_score[key].extend([rows_weather[0][column_weather_today["temperature_max"]],
126
                                 rows_weather[0][column_weather_today["temperature_min"]]
127
                                ])
128

129
    # # If you wanna execute other environment(ex. jupyternotebook)
130
    # path_w = "combi_score.py"
131
    # generate_function_to_text("generate_dict_score", path_w, combi_score)
132

133
    # Generate pandas-dataframe from dictionary
134
    data_warmscore = pd.DataFrame.from_dict(combi_score, orient='index', columns=feature_names[0:7+1])
135
    list_predict_score = model.predict(data_warmscore) # Predict (Linear Multiple Regression)
136
    dict_predict_score = {}                            # predict-scores by model of machine learning
137
    for i in range(0, num): # Store absolute value of difference between "__best_score__" and "list_predict_score"
138
        dict_predict_score[i] = abs(list_predict_score[i] - __best_score__)
139
    dict_predict_best = {}  # Store best comfort-score
140
    num_candidate = 3       # Number of candidates of combination
141

142
    # Get "num_candidate" of the best "comfort-score"
143
    for k, v in  sorted(dict_predict_score.items(), key=lambda x: x[1]):
144
        dict_predict_best[k] = v
145
        num_candidate -= 1
146
        if num_candidate == 0:
147
            break
148

149
    # Get wardrobeid-combination
150
    # dict_predict_best = {3761: 0.0006331531891259345, 19205: 0.0006331531891259345, 3597: 0.0012729990001982827}
151
    combi_content_best = []
152
    num_candidate = 0
153
    for k in dict_predict_best.keys():
154
        combi_content_best.append(combi_content[k])
155
        num_candidate += 1
156

157
    # Part of "Today's Recommend" -----------------------------------------
158
    # Dicionary which will put image-url on wardrobeid
159
    dict_image_url = {"wardrobeid_c": 2, "wardrobeid_o": None, "wardrobeid_t": 7,
160
                      "wardrobeid_i": None, "wardrobeid_b": 3, "wardrobeid_s": None
161
                     }
162
    # dict_image_url = combi_content_best[0]
163
    # Dictionary which will put wardrobe-name on wardrobeid
164
    dict_wardrobename = {}
165
    # Display the wardrobes selected today
166
    for key, value in dict_image_url.items(): # if a wardrobe isn't selected,
167
        if value == None:
168
            dict_image_url[key] = nowardrobe_path
169
            dict_wardrobename[key] = "~NoWardrobe~"
170
        else: # if a wardrobe is selected,
171
            # Select DB for wardrobes to get wardrobename from wardrobeid
172
            cursor.execute("SELECT wardrobename FROM wardrobes WHERE id = :wardrobeid;",
173
                            {"wardrobeid": value
174
                            })
175
            # Get wardrobename
176
            rows_wn = cursor.fetchone()
177
            # Set image-url from wardrobename
178
            dict_image_url[key] = os.path.join("..", image_path, str(session["user_id"]), "{}.jpg".format(rows_wn[0]))
179
            dict_wardrobename[key] = rows_wn[0]
180

181
    # Part of "What's your outfit today?" --------------------------------------
182
    # Select DB for wardrobes
183
    cursor.execute("SELECT * FROM wardrobes as T1 LEFT JOIN (SELECT wardrobeid, date_to FROM history_own) T2 ON T1.id = T2.wardrobeid WHERE userid = :t_userid AND date_to = 'ongoing';",
184
                    {'t_userid': session["user_id"]
185
                    })
186
    tuple_row = cursor.fetchall()
187

188
    dict_category = category_dictionary() # Dictionary about wardrobe's category
189
    list_wardrobe_o = []                  # outer category
190
    list_wardrobe_t = []                  # tops category
191
    list_wardrobe_i = []                  # inner category
192
    list_wardrobe_b = []                  # bottoms category
193
    # Put wardrobe-name in list for "select" tag
194
    for i in tuple_row:
195
        if i[column_wardrobes["category"]] in dict_category["-Outer Category-"]:
196
            list_wardrobe_o.append(i[column_wardrobes["wardrobename"]])
197
        elif i[column_wardrobes["category"]] in dict_category["-Tops Category-"]:
198
            list_wardrobe_t.append(i[column_wardrobes["wardrobename"]])
199
        elif i[column_wardrobes["category"]] in dict_category["-Inner Category-"]:
200
            list_wardrobe_i.append(i[column_wardrobes["wardrobename"]])
201
        elif i[column_wardrobes["category"]] in dict_category["-Bottoms Category-"]:
202
            list_wardrobe_b.append(i[column_wardrobes["wardrobename"]])
203
    dict_list_wardrobe = {"c": list_wardrobe_o, "o": list_wardrobe_o,
204
                          "t": list_wardrobe_t, "i": list_wardrobe_i,
205
                          "b": list_wardrobe_b, "s": list_wardrobe_b
206
                         }
207

208
    # Remember which page user is in now
209
    session["nowpage"] = "recommend"
210

211
    return render_template("recommend.html", img_urls=dict_image_url, names=dict_wardrobename, wardrobes=dict_list_wardrobe) # おすすめの服を反映。

1
def model_learning(data):
2
    # # 線形単回帰 Linear Regression
3
    # for i in feature_names:
4
    #     train_X, test_X, train_y, test_y = train_test_split(
5
    #         data[i],data["comfort_score"],random_state=42) # トレーニングとテストデータを作成
6
    #     print(data[i][0])
7
    #     # Xのデータの形を修正
8
    #     train_X = train_X.values.reshape((-1, 1))
9
    #     test_X = test_X.values.reshape((-1, 1))
10
    #     model = LinearRegression() # Generate model
11
    #     model.fit(train_X, train_y) # Let model learn
12
    #     print(model.predict(test_X))
13
    #     print("線形単回帰：{}".format(i)) # データ列
14
    #     print("決定係数：{}\n".format(model.score(test_X, test_y))) # テストして出た決定係数
15

16
    # 線形重回帰 Linear Multiple Regression
17
    train_X, test_X, train_y, test_y = train_test_split(
18
        data.drop("comfort_score", axis=1), data["comfort_score"], random_state=42)
19
    model = LinearRegression()
20
    model.fit(train_X, train_y)
21

22
    # save model as .sav file
23
    pickle.dump(model, open('comfort_model.sav', 'wb'))
24

25
    print(test_X)
26
    # print(model.predict(test_X))
27
    print("線形重回帰：{}".format("comfort_score")) # データ列
28
    print("決定係数：{}\n".format(model.score(test_X, test_y))) # テストして出た決定係数
29

30
    # # ElasticNet回帰
31
    # train_X, test_X, train_y, test_y = train_test_split(data.drop("comfort_score", axis=1), data["comfort_score"], random_state=42)
32
    # model = ElasticNet(l1_ratio=0.0) # ラッソで鑑みる割合
33
    # model.fit(train_X, train_y) # Let model learn
34
    # print("ElasticNet回帰：{}".format(model.score(test_X, test_y)))
35

36
    return model