added similarity code

demo/digit_classifier.py

@@ -4,6 +4,7 @@ import tensorflow as tf
 import gradio
 import gradio as gr
 from urllib.request import urlretrieve
+import os
 
 urlretrieve("https://gr-models.s3-us-west-2.amazonaws.com/mnist-model.h5", "mnist-model.h5")
 model = tf.keras.models.load_model("mnist-model.h5")
@@ -14,12 +15,15 @@ def recognize_digit(image):
     prediction = model.predict(image).tolist()[0]
     return {str(i): prediction[i] for i in range(10)}
 
+im = gradio.inputs.Image(shape=(28, 28), image_mode='L', invert_colors=True)
 
 io = gr.Interface(
     recognize_digit, 
-    "sketchpad", 
+    im, 
     gradio.outputs.Label(num_top_classes=3),
-    live=True,
+    examples=[['digits/' + x] for x in os.listdir('digits/')],
+    # live=True,
+    interpretation="default",
     capture_session=True,
 )
 

demo/image_classifier.py

@@ -1,7 +1,7 @@
 # Demo: (Image) -> (Label)
 
 import gradio as gr
-import tensorflow as tf
+# import tensorflow as tf
 
 import numpy as np
 from PIL import Image
@@ -13,12 +13,13 @@ import json
 with open("files/imagenet_labels.json") as labels_file:
     labels = json.load(labels_file)
 
-mobile_net = tf.keras.applications.MobileNetV2()
+# mobile_net = tf.keras.applications.MobileNetV2()
 
 
 def image_classifier(im):
+    return 0
     arr = np.expand_dims(im, axis=0)
-    arr = tf.keras.applications.mobilenet.preprocess_input(arr)
+    # arr = tf.keras.applications.mobilenet.preprocess_input(arr)
     prediction = mobile_net.predict(arr).flatten()
     return {labels[i]: float(prediction[i]) for i in range(1000)}
 
@@ -34,4 +35,4 @@ io = gr.Interface(image_classifier, image, label,
         ["images/lion.jpg"]
     ])
 
-io.launch()
+io.launch()

gradio/inputs.py

@@ -640,6 +640,8 @@ class Image(InputComponent):
             output_scores = (output_scores - min_val) / (max_val - min_val)
         return output_scores.tolist()
 
+    def embed(self, x):
+        return x.flatten()
 
 class Audio(InputComponent):
     """

gradio/interface.py

@@ -15,6 +15,7 @@ import inspect
 import sys
 import weakref
 import analytics
+import numpy as np
 import os
 import copy
 
@@ -42,6 +43,7 @@ class Interface:
                  description=None, thumbnail=None, server_port=None, 
                  server_name=networking.LOCALHOST_NAME,
                  allow_screenshot=True, allow_flagging=True,
+                 embedding_fn="default",
                  flagging_dir="flagged", analytics_enabled=True):
 
         """
@@ -241,6 +243,16 @@ class Interface:
             predictions[i]) for i, output_interface in enumerate(self.output_interfaces)]
         return processed_output, durations
     
+    def embedding_fn(self, raw_input):
+        if self.interpretation == "default":
+            processed_input = [input_interface.preprocess(raw_input[i])
+                            for i, input_interface in enumerate(self.input_interfaces)]
+            embedding = np.concatenate([input_interface.embed(processed_input[i])
+                            for i, input_interface in enumerate(self.input_interfaces)])
+        else:
+            raise NotImplementedError("Only default embedding is currently supported")
+        return embedding
+
     def interpret(self, raw_input):
         """
         Runs the interpretation command for the machine learning model. Handles both the "default" out-of-the-box

gradio/networking.py

@@ -11,16 +11,17 @@ from flask_cors import CORS
 import threading
 import pkg_resources
 from distutils import dir_util
-import gradio as gr
 import time
 import json
-from gradio.tunneling import create_tunnel
 import urllib.request
 from shutil import copyfile
 import requests
 import sys
 import csv
 import logging
+import gradio as gr
+from gradio.similarity import calculate_similarity
+from gradio.tunneling import create_tunnel
 
 INITIAL_PORT_VALUE = int(os.getenv(
     'GRADIO_SERVER_PORT', "7860"))  # The http server will try to open on port 7860. If not available, 7861, 7862, etc.
@@ -120,6 +121,19 @@ def predict():
     return jsonify(output)
 
 
+@app.route("/api/score_similarity/", methods=["POST"])
+def score_similarity():
+    raw_input = request.json["data"]
+    input_embedding = app.interface.embedding_fn(raw_input)
+    scores = list()
+    for example in app.interface.examples:
+        preprocessed_example = [iface.preprocess_example(example)
+            for iface, example in zip(app.interface.input_interfaces, example)]
+        example_embedding = app.interface.embedding_fn(preprocessed_example)
+        scores.append(calculate_similarity(input_embedding, example_embedding))
+    return jsonify({"data": scores})
+
+
 @app.route("/api/predict_examples/", methods=["POST"])
 def predict_examples():
     example_ids = request.json["data"]

gradio/similarity.py

@@ -0,0 +1,8 @@
+import numpy as np
+
+def calculate_similarity(embedding1, embedding2):
+    """
+    Scores the similarity between two embeddings by taking the L2 distance
+    """
+    return np.linalg.norm(np.array(embedding1) - np.array(embedding2))
+    

gradio/static/js/all_io.js

@@ -36,8 +36,33 @@ var io_master_template = {
       this.target.find(".loading_in_progress").hide();
       this.target.find(".loading_failed").show();
     });
+  },
+  score_similarity: function() {
+    this.target.find(".loading").removeClass("invisible");
+    this.target.find(".loading_in_progress").show();
+    this.target.find(".loading_failed").hide();
+    this.target.find(".output_interfaces").css("opacity", 0.5);
+
+    this.fn(this.last_input, "score_similarity").then((output) => {
+      console.log(output.data)
+      this.target.find(".loading").addClass("invisible");
+      this.target.find(".output_interfaces").css("opacity", 1);
+      let html = "<th>DIFFS</th>"
+      this.target.find(".examples > table > thead > tr").append(html); 
+      for (let i = 0; i < output["data"].length; i++) {
+        let html = "<td>" + output["data"][i] + "</td>"
+        this.target.find(".examples_body tr[row='" + i + "']").append(html);
+      }
+    })
+
+
   },
   submit_examples: function() {
+    this.target.find(".loading").removeClass("invisible");
+    this.target.find(".loading_in_progress").show();
+    this.target.find(".loading_failed").hide();
+    this.target.find(".output_interfaces").css("opacity", 0.5);
+
     let example_ids = [];
     if (this.loaded_examples == null) {
       this.loaded_examples = {};
@@ -48,6 +73,9 @@ var io_master_template = {
       }
     }
     this.fn(example_ids, "predict_examples").then((output) => {
+      this.target.find(".loading").addClass("invisible");
+      this.target.find(".output_interfaces").css("opacity", 1);
+
       output = output["data"];
       if (!this.has_loaded_examples) {
         this.has_loaded_examples = true;
@@ -147,3 +175,5 @@ var io_master_template = {
     }
   }
 };
+
+

gradio/static/js/gradio.js

@@ -320,7 +320,8 @@ function gradio(config, fn, target, example_file_path) {
   target.find(".interpret").click(function() {
     target.find(".interpretation_explained").removeClass("invisible");
     if (io_master.last_output) {
-      io_master.interpret();
+      io_master.score_similarity();
+      // io_master.interpret(); // TODO(UNDO)
     }
   });
   target.find(".run_examples").click(function() {