def test_PhaGat2_classification():
import phagraphnn.utilities as ut
from phagraphnn.PhaGraph import PhaGraph,PhaNode
data = ut.readChemblXls("./tests/data/CHE_3.xls")
graph_list = []
for i in range(0,len(data)):
graph = PhaGraph()
mol = ut.CDPLmolFromSmiles(data[i][1],True)
graph(ut.CDPLphaGenerator(None,mol,"lig_only"))
graph.setProperty("ic50",(0,1))
graph_list.append(graph)
from phagraphnn.DataPreparer import DataPreparer
loader = DataPreparer(graph_list,3,property_string="ic50",mpn="gru",is_path=False)
from phagraphnn.PhaGatModel2 import PhaGatModel2 as gat
import tensorflow as tf
seq = tf.keras.Sequential([
tf.keras.layers.Dense(16, activation='relu', input_shape=(128,),name="first_layer"),
tf.keras.layers.Dense(8, activation='relu',name="second_layer"),
tf.keras.layers.Dense(2,activation= None)],name="output_NN")
gat = gat(hidden_dim=32,output_nn=seq,regression = False)
lr = 0.001
gat.compile(loss=tf.keras.losses.BinaryCrossentropy(),
optimizer=tf.keras.optimizers.Adam(lr))
rec = tf.keras.metrics.MeanAbsoluteError()
for batch in loader:
inputs,af,other = batch
gat(inputs)
for epoch in range(0,10):
pred,loss = gat.train(inputs=inputs,outputs=af,learning_rate=lr)
16, activation='relu', input_shape=(32, ), name="first_layer"),
tf.keras.layers.Dense(8, activation='relu', name="second_layer"),
tf.keras.layers.Dense(1, activation=None)
],
name="output_NN")
seq_gru3 = tf.keras.Sequential([
tf.keras.layers.Dense(
16, activation='relu', input_shape=(32, ), name="first_layer"),
tf.keras.layers.Dense(8, activation='relu', name="second_layer"),
tf.keras.layers.Dense(1, activation=None)
],
name="output_NN")
# here the model is defined
gat = gat(hidden_dim=32, emb_dim=32, output_nn=seq_gat)
gat2 = gat2(hidden_dim=32, emb_dim=32, output_nn=seq_gat2, num_heads=3)
gru = gru(32, 1, seq_gru)
gru2 = gru2(32, 1, seq_gru2)
gru3 = gru3(32, 1, seq_gru3)
lr = 0.001
gat.compile(loss=tf.keras.losses.mse,
optimizer=tf.keras.optimizers.RMSprop(lr))
gat2.compile(loss=tf.keras.losses.mse,
optimizer=tf.keras.optimizers.RMSprop(lr))
gru.compile(loss=tf.keras.losses.mse,
optimizer=tf.keras.optimizers.RMSprop(lr))
gru2.compile(loss=tf.keras.losses.mse,
optimizer=tf.keras.optimizers.RMSprop(lr))
gru3.compile(loss=tf.keras.losses.mse,
optimizer=tf.keras.optimizers.RMSprop(lr))