Add dropout slider

chatbot_constructor.py

@@ -29,7 +29,7 @@ def todset(text: str):
 def hash_str(data: str):
     return hashlib.md5(data.encode('utf-8')).hexdigest()
 
-def train(message: str = "", epochs: int = 16, learning_rate: float = 0.001, emb_size: int = 128, input_len: int = 16, kernels_count: int = 8, kernel_size: int = 8, data: str = ""):
+def train(message: str = "", dropout: float = 0.5, learning_rate: float = 0.001, epochs: int = 16, emb_size: int = 128, input_len: int = 16, kernels_count: int = 8, kernel_size: int = 8, data: str = ""):
     data_hash = None
     if "→" not in data or "\n" not in data:
             if data in os.listdir("cache"):
@@ -54,7 +54,8 @@ def train(message: str = "", epochs: int = 16, learning_rate: float = 0.001, emb
     else:
         input_layer = Input(shape=(inp_len,))
         emb_layer = Embedding(input_dim=vocab_size, output_dim=emb_size, input_length=inp_len)(input_layer)
-        attn_layer = MultiHeadAttention(num_heads=4, key_dim=128)(emb_layer, emb_layer, emb_layer)
+        dropout1_layer = Dropout(0.25)(emb_layer)
+        attn_layer = MultiHeadAttention(num_heads=4, key_dim=128)(dropout1_layer, dropout1_layer, dropout1_layer)
         noise_layer = GaussianNoise(0.1)(attn_layer)
         conv1_layer = Conv1D(kernels_count, kernel_size, padding='same', activation='relu', strides=1, input_shape=(64, 128))(noise_layer)
         conv2_layer = Conv1D(16, 4, padding='same', activation='relu', strides=1)(conv1_layer)
@@ -65,12 +66,15 @@ def train(message: str = "", epochs: int = 16, learning_rate: float = 0.001, emb
         conv2_flatten_layer = Flatten()(conv2_layer)
         conv3_flatten_layer = Flatten()(conv3_layer)
         concat1_layer = Concatenate()([flatten_layer, attn_flatten_layer, conv1_flatten_layer, conv2_flatten_layer, conv3_flatten_layer])
-        dense1_layer = Dense(512, activation="linear")(concat1_layer)
+        dropout2_layer = Dropout(0.5)(concat1_layer)
+        dense1_layer = Dense(512, activation="linear")(dropout2_layer)
         prelu1_layer = PReLU()(dense1_layer)
-        dropout_layer = Dropout(0.3)(prelu1_layer)
-        dense2_layer = Dense(256, activation="tanh")(dropout_layer)
-        dense3_layer = Dense(256, activation="relu")(dense2_layer)
-        dense4_layer = Dense(100, activation="tanh")(dense3_layer)
+        dropout3_layer = Dropout(0.5)(prelu1_layer)
+        dense2_layer = Dense(256, activation="tanh")(dropout3_layer)
+        dropout4_layer = Dropout(0.5)(dense2_layer)
+        dense3_layer = Dense(256, activation="relu")(dropout4_layer)
+        dropout5_layer = Dropout(0.5)(dense3_layer)
+        dense4_layer = Dense(100, activation="tanh")(dropout5_layer)
         concat2_layer = Concatenate()([dense4_layer, prelu1_layer, attn_flatten_layer, conv1_flatten_layer])
         dense4_layer = Dense(resps_len, activation="softmax")(concat2_layer)
         model = Model(inputs=input_layer, outputs=dense4_layer)
@@ -97,8 +101,9 @@ def train(message: str = "", epochs: int = 16, learning_rate: float = 0.001, emb
 
 if __name__ == "__main__":
     iface = gr.Interface(fn=train, inputs=["text",
-                                           gr.inputs.Slider(1, 64, default=32, step=1, label="Epochs"),
+                                           gr.inputs.Slider(0, 0.5, default=0.5, step=0.00000001, label="Dropout"),
                                            gr.inputs.Slider(0.00000001, 0.1, default=0.001, step=0.00000001, label="Learning rate"),
+                                           gr.inputs.Slider(1, 64, default=32, step=1, label="Epochs"),
                                            gr.inputs.Slider(1, 256, default=100, step=1, label="Embedding size"),
                                            gr.inputs.Slider(1, 128, default=16, step=1, label="Input Length"),
                                            gr.inputs.Slider(1, 128, default=64, step=1, label="Convolution kernel count"),