Keras implementation

environment.yml

@@ -4,13 +4,15 @@ channels:
   - rdkit
   - conda-forge 
   - chembl
+  - anaconda
 dependencies:
   - pip
   - python=3.6
   - setuptools=>38.4.0
   - scikit-learn=>0.19.1
   - requests
   - rdkit
+  - keras
   - pyyaml
   - pytest
   - psycopg2

environment_gpu.yml

@@ -0,0 +1,28 @@
+name: flame_gpu
+channels:
+  - default
+  - rdkit
+  - conda-forge 
+  - chembl
+  - anaconda
+dependencies:
+  - pip
+  - python=3.6
+  - setuptools=>38.4.0
+  - scikit-learn=>0.19.1
+  - requests
+  - rdkit
+  - matplotlib
+  - keras-gpu
+  - pyyaml
+  - pytest
+  - psycopg2
+  - chembl_structure_pipeline
+  - pip:
+     - xgboost
+     - django==2.2.8
+     - djangorestframework
+     - django-cors-headers
+     - djangorestframework-yaml
+     - "https://github.com/phi-grib/standardiser/archive/master.zip"
+     - "https://github.com/josecarlosgomezt/nonconformist/archive/master.zip"

flame/apply.py

@@ -32,6 +32,7 @@
 from flame.stats.PLSDA import PLSDA
 from flame.stats.combo import median, mean, majority, matrix
 from flame.stats.XGboost import XGBOOST
+from flame.stats.Keras import Keras_nn
 from sklearn.metrics import mean_squared_error, matthews_corrcoef as mcc
 from sklearn.metrics import f1_score
 from sklearn.metrics import make_scorer
@@ -57,6 +58,7 @@ def __init__(self, parameters, conveyor):
                               ('GNB', GNB),
                               ('PLSR', PLSR),
                               ('PLSDA', PLSDA),
+                              ('Keras', Keras_nn),
                               ('median', median),
                               ('mean', mean),
                               ('majority', majority),

flame/children/my_keras.py

@@ -0,0 +1,44 @@
+#! -*- coding: utf-8 -*-
+
+# Description    Flame Apply internal class
+##
+# Authors:       Manuel Pastor (manuel.pastor@upf.edu)
+##
+# Copyright 2018 Manuel Pastor
+##
+# This file is part of Flame
+##
+# Flame is free software: you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation version 3.
+##
+# Flame is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+##
+# You should have received a copy of the GNU General Public License
+# along with Flame.  If not, see <http://www.gnu.org/licenses/>.
+
+import numpy as np
+from flame.stats.Keras import Keras_nn
+
+class my_keras(Keras_nn):
+    def __init__(self, X, Y, parameters, conveyor):
+        Keras_nn.__init__(self, X, Y, parameters, conveyor)
+
+# Function to create model, required for KerasClassifier
+    def create_model(self, dim=20):
+        # create model
+        model = Sequential()
+        model.add(Dense(10, input_dim=dim, activation='relu'))
+        model.add(Dense(20, activation='sigmoid'))
+        model.add(Dense(1, activation='sigmoid'))
+        # Compile model
+
+        if self.param.getVal('quantitative'):
+            loss = 'mean_squared_error'
+        else:
+            loss = 'binary_crossentropy'
+        model.compile(loss=loss, optimizer='adam', metrics=['accuracy'])
+        return model

flame/children/parameters.yaml

@@ -1067,6 +1067,59 @@ XGBOOST_optimize:
   comments: 
   group: modeling
 
+Keras_parameters:
+  advanced: advanced
+  object_type: dictionary
+  writable: false
+  options: null
+  value:
+    epochs:
+      object_type: list
+      writable: false
+      value: 100
+      options:
+        - 100
+        - 200
+        - 250
+      description:  Number of epochs
+    batch_size:
+      object_type: float
+      writable: true
+      value: 36
+      options:
+        - 1
+        - 10
+        - 30
+      description:  Batch size for model Keras batch learning
+  description: 
+  dependencies: 
+    model: SKLEARN
+  comments: 
+  group: modeling
+
+Keras_optimize:
+  advanced: advanced
+  object_type: dictionary
+  writable: false
+  options: null
+  value:
+    epoch:
+      object_type: list
+      writable: false
+      value: 
+        - 50
+        - 100
+        - 150
+      options:
+        - 50
+        - 100
+        - 150
+      description: 
+  description: Keras optimize parameters
+  dependencies: 
+    model: Keras
+  comments: 
+  group: modeling
 
 output_format:
   advanced: regular

flame/learn.py

@@ -24,6 +24,7 @@
 import os
 import pickle
 import numpy as np
+import time
 
 from sklearn.preprocessing import MinMaxScaler 
 from sklearn.preprocessing import StandardScaler 
@@ -36,6 +37,7 @@
 from flame.stats.PLSDA import PLSDA
 from flame.stats import feature_selection
 from flame.stats.XGboost import XGBOOST
+from flame.stats.Keras import Keras_nn
 from flame.stats.combo import median, mean, majority, matrix
 from flame.stats.imbalance import run_imbalance  
 
@@ -65,6 +67,7 @@ def __init__(self, parameters, conveyor):
                               ('GNB', GNB),
                               ('PLSR', PLSR),
                               ('PLSDA', PLSDA), 
+                              ('Keras', Keras_nn),
                               ('median', median),
                               ('mean', mean),
                               ('majority', majority),
@@ -397,6 +400,8 @@ def run(self):
         '''
         Builds the model using the appropriate toolkit (internal or custom).
         '''
+        # Count the time
+        start = time.time()
 
         toolkit = self.param.getVal('modelingToolkit')
 
@@ -411,5 +416,10 @@ def run(self):
             LOG.error("Modeling toolkit is not yet supported")
             self.conveyor.setError( 'modeling Toolkit ' + \
                 toolkit+' is not supported yet')
-
+
+        end = time.time()
+        hours, rem = divmod(end-start, 3600)
+        minutes, seconds = divmod(rem, 60)
+        LOG.info("{:0>2}:{:0>2}:{:05.2f}".format(int(hours),
+              int(minutes),seconds))
         return 

flame/stats/Keras.py

@@ -35,6 +35,7 @@
 from keras.models import Sequential
 from keras.layers import Dense
 from sklearn.base import clone
+import keras
 
 import numpy as np
 
@@ -140,7 +141,7 @@ def build(self):
                     **self.estimator_parameters, verbose=0)
                     results.append(('model', 'model type', 'Keras quantitative'))
                 else:
-
+                    print(self.estimator_parameters)
                     LOG.info("Building Qualitative Keras model")
                     self.estimator = KerasClassifier(build_fn=self.create_model, dim=self.X.shape[1],
                      **self.estimator_parameters, verbose=0)
@@ -209,13 +210,13 @@ def build(self):
             raise e
             return False, f'Exception building conformal Keras estimator with exception {e}'
 
-        return True, []
+        return True, results
 
 # Function to create model, required for KerasClassifier
-    def create_model(self, dim=20):
+    def create_model(self, dim=25):
         # create model
         model = Sequential()
-        model.add(Dense(10, input_dim=dim, activation='relu'))
+        model.add(Dense(30, input_dim=dim, activation='relu'))
         model.add(Dense(20, activation='sigmoid'))
         model.add(Dense(1, activation='sigmoid'))
         # Compile model
@@ -224,7 +225,15 @@ def create_model(self, dim=20):
             #loss = 'mean_squared_error'
         #else:
         loss = 'binary_crossentropy'
-        model.compile(loss=loss, optimizer='adam', metrics=['accuracy'])
+        optimizer = keras.optimizers.Adam(lr=0.1)
+        model.compile(loss=loss, optimizer=optimizer, metrics=['accuracy'])
+        # model.compile(
+        #             optimizer=keras.optimizers.Adam(
+        #                 hp.Choice('learning_rate',
+        #                 values=[1e-2, 1e-3, 1e-4])),
+        #             loss='sparse_categorical_crossentropy',
+        #             metrics=['accuracy'])
+
         return model
 
     # Overrides regular project to single class prediction

flame/stats/base_model.py

@@ -146,6 +146,15 @@ def __init__(self, X, Y, parameters, conveyor=None):
         self.Y = Y
         self.nobj, self.nvarx = np.shape(X)
 
+        # Check if model is a keras model and
+        # adjust n_jobs=1 if so
+        if self.param.getVal('model') == 'my_keras'\
+         or self.param.getVal('model') == 'Keras':
+            self.n_jobs = 1
+        else:
+            self.n_jobs = -1
+
+
         # Get cross-validator
         # Consider to include a Random Seed for cross-validator
         if self.param.getVal('ModelValidationCV'):
@@ -793,6 +802,7 @@ def quantitativeValidation(self):
             return False, f'Error computing goodness of fit with exception: {e}'
 
         # Compute Cross-validation quality metrics
+
         try:
             # Get predicted Y
             # y_pred = cross_val_predict(copy.copy(self.estimator), copy.copy(X), copy.copy(Y), cv=self.cv, n_jobs=1)
@@ -825,6 +835,8 @@ def quantitativeValidation(self):
         results ['Y_pred'] = y_pred
         return True, results
 
+
+
     def qualitativeValidation(self):
         ''' performs validation for qualitative models '''
 
@@ -864,7 +876,8 @@ def qualitativeValidation(self):
 
         # Get cross-validated Y 
         try:
-            y_pred = cross_val_predict(self.estimator, X, Y, cv=self.cv, n_jobs=-1)
+            y_pred = cross_val_predict(self.estimator, X, Y, cv=self.cv,
+                                       n_jobs=self.n_jobs)
         except Exception as e:
             return False, f'Cross-validation failed with exception: {e}'
 
@@ -1147,6 +1160,8 @@ def save_model(self):
             params = self.estimator_temp.get_params()
         else:
             params = self.estimator.get_params()
+        if self.param.getVal('model') == "Keras":
+            params = ["Keras sequential model"]
 
         self.conveyor.addVal(params, 'estimator_parameters',
                             'estimator parameters', 'method', 'single',
@@ -1180,4 +1195,4 @@ def load_model(self):
         if self.estimator is None:
             return False, 'No valid model estimator found. Try to rebuild the model'
 
-        return True, 'model loaded'
+        return True, 'model loaded'