mlFramework/KerasModel.py at master · mflechl/mlFramework · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
import ROOT
ROOT.PyConfig.IgnoreCommandLineOptions = True  # disable ROOT internal argument parser
import numpy as np
from pandas import DataFrame,concat
import json
np.random.seed(0)
import conf.keras_models as keras_models
from keras.callbacks import EarlyStopping, ModelCheckpoint
from keras.models import load_model as lm
from keras.utils.np_utils import to_categorical
from collections import deque
import time
import os

class KerasObject():

    def __init__(self, parameter_file = "", variables=[], target_names = {}, filename = "" ):

        self.variables = variables
        self.models = []

        try:
            if filename: self.load(filename)
            elif not parameter_file or not variables:
                raise Warning("Warning! Object not defined. Load from file or set 'params' and 'variables'")
            else:
                with open(parameter_file,"r") as FSO:
                    params = json.load(FSO)
                self.params = params["model"]
        except Warning as e:
            print e
            self.params = []

        if target_names: self.target_names = target_names


    def load(self, filename):
        with open(filename + ".dict", 'rb') as FSO:
            tmp_dict = json.load(FSO)

        print "Loading model from: " + filename
        self.__dict__.clear()
        self.__dict__.update(tmp_dict)

        self.models = []
        for model in tmp_dict["models"]:
            self.models.append( lm(model) )

    def save(self, filename):
        placeholders = []
        tmp_models = []
        for i,model in enumerate(self.models):
            modelname = filename + ".fold{0}".format(i)
            model.save( modelname )
            tmp_models.append(model)
            placeholders.append( modelname )
        self.models = placeholders

        with open(filename + ".dict", 'wb') as FSO:
            json.dump(self.__dict__, FSO)

        self.models = tmp_models


    def train(self, samples):

        if type(samples) is list:
            samples = deque(samples)

        for i in xrange( len(samples) ):
            test = samples[0]
            train = [ samples[1] ]

            for j in xrange(2, len(samples) ):
                train.append( samples[j] )

            train = concat(train , ignore_index=True).reset_index(drop=True)

            self.models.append( self.trainSingle( train, test ) )
            samples.rotate(-1)

        print "Finished training!"


    def trainSingle(self, train, test):


        # writing targets in keras readable shape
        best = str(int(time.time()))
        y_train = to_categorical( train["target"].values )
        y_test  = to_categorical( test["target"].values )

        N_classes = len(y_train[0])

        model_impl = getattr(keras_models, self.params["name"])
        model = model_impl(len(self.variables), N_classes)
        model.summary()
        history = model.fit(
            train[self.variables].values,
            y_train,
            sample_weight=train["train_weight"].values,
            validation_split = 0.25,
            # validation_data=(test[self.variables].values, y_test, test["train_weight"].values),
            batch_size=self.params["batch_size"],
            epochs=self.params["epochs"],
            shuffle=True,
            callbacks=[EarlyStopping(patience=self.params["early_stopping"]), ModelCheckpoint( best + ".model", save_best_only=True, verbose = 1) ])

        import matplotlib as mpl
        mpl.use('Agg')
        import matplotlib.pyplot as plt

        print "plotting training"
        epochs = xrange(1, len(history.history["loss"]) + 1)
        plt.plot(epochs, history.history["loss"], lw=3, label="Training loss")
        plt.plot(epochs, history.history["val_loss"], lw=3, label="Validation loss")
        plt.xlabel("Epoch")
        plt.ylabel("Loss")
        plt.legend()
        if not os.path.exists("plots"):
            os.mkdir("plots")
        plt.savefig("plots/fold_{0}_loss.png".format(best), bbox_inches="tight")


        print "Reloading best model"
        model = lm(best + ".model")
        os.remove( best + ".model" )

        return model

    def predict(self, samples, where=""):

        predictions = []
        if type(samples) is list:
            samples = deque(samples)

        for i in xrange( len(samples) ):
            predictions.append( self.testSingle( samples[0], i ) )
            samples.rotate(-1)

        samples[0].drop(samples[0].index, inplace = True)
        samples[1].drop(samples[1].index, inplace = True)

        return predictions


    def testSingle(self, test,fold ):

        prediction = DataFrame( self.models[fold].predict(test[self.variables].values) )

        return DataFrame(dtype = float, data = {"predicted_class":prediction.idxmax(axis=1).values,
                                 "predicted_prob": prediction.max(axis=1).values } )