ML services validation update

Author: perdy

ml_services_validation_19-05/slides/Slides.pdf

@@ -20,6 +20,26 @@ \section{Machine Learning}
     \end{block}
 \end{frame}
 
+\begin{frame}{AI in perspective}
+    \only<1>{
+        \begin{figure}
+            \includegraphics[width=\textwidth]{img/tree-ai.png}
+        \end{figure}
+    }
+
+    \only<2>{
+        \begin{figure}
+            \includegraphics[width=\textwidth]{img/venn-ai.jpeg}
+        \end{figure}
+    }
+
+    \only<3>{
+        \begin{figure}
+            \includegraphics[width=\textwidth]{img/venn-ai-simplified.jpeg}
+        \end{figure}
+    }
+\end{frame}
+
 \begin{frame}{What is Machine Learning?}
     The science of getting a computer to act without programming. There are three types of machine learning algorithms:
 
@@ -42,34 +62,14 @@ \section{Machine Learning}
     \end{block}
 \end{frame}
 
-\begin{frame}{AI in perspective}
-    \only<1>{
-        \begin{figure}
-            \includegraphics[width=\textwidth]{img/tree-ai.png}
-        \end{figure}
-    }
-
-    \only<2>{
-        \begin{figure}
-            \includegraphics[width=\textwidth]{img/venn-ai.jpeg}
-        \end{figure}
-    }
-
-    \only<3>{
-        \begin{figure}
-            \includegraphics[width=\textwidth]{img/venn-ai-simplified.jpeg}
-        \end{figure}
-    }
-\end{frame}
-
 \begin{frame}{Supervised Learning}
     \begin{figure}
         \includegraphics[width=\textwidth]{img/training-process.png}
     \end{figure}
 \end{frame}
 
 \begin{frame}{The Model}
-    We are going to build a model that performs a sentiment analysis over a text and concludes it it is positive or negative.
+    We are going to build a model that performs a sentiment analysis over a text and concludes if it is positive or negative.
 
     \textbf{Input:} A text (a list of integers representing each word).
 

ml_services_validation_19-05/slides/machine_learning.tex

@@ -3,8 +3,12 @@ \section{Building a Machine Learning Service}
 \begin{frame}{How to expose the ML model?}
     Machine Learning models can be used either as an internal piece of a service or as a service itself.
 
+    \pause
+
     If it is used as an \textbf{internal piece} you won't notice it, such as scoring or recommendation systems within bigger products like Spotify or Netflix.
 
+    \pause
+
     But you can also find them as a \textbf{service that exposes an API} to directly interact with the model. There are many examples of that in AWS, Google Cloud, Azure...
 \end{frame}
 
@@ -21,17 +25,17 @@ \section{Building a Machine Learning Service}
             \begin{description}
                 \item[Verb] \texttt{\textbf{GET}}
                 \item[URL] \texttt{https://service.url/analyze/}
-                \item[Params] \texttt{text=The girl is having fun while playing with her toys}
+                \item[Params] \texttt{text=The\%20girl\%20is\%20having\%20fun\%20while\%20playing}
             \end{description}
         }
     \end{exampleblock}
 
     \begin{exampleblock}{Response}
         \begin{minted}[fontsize=\footnotesize]{js}
 {
-  "score": 0.5628073215484619,
-  "text": "The girl is having fun while playing with her toys",
-  "sentiment": "Positive"
+  "text": "The girl is having fun while playing",
+  "sentiment": "Positive",
+  "score": 0.6321590542793274
 }
         \end{minted}
     \end{exampleblock}
@@ -40,14 +44,16 @@ \section{Building a Machine Learning Service}
 \begin{frame}{Building a REST API with Flama}
     To build a REST API we need to define:
 
-    \begin{enumerate}
+    \begin{enumerate}[<+->]
         \item A \textbf{component} that loads our ML model.
         \item The \textbf{data schema} for our response.
         \item The \textbf{view} function that will be called through requests to \texttt{/analyze/} endpoint.
         \item The whole API \textbf{application}.
     \end{enumerate}
 
-    Everything put together is less than 100 lines of python code.
+    \only<4>{
+        Everything put together is less than 100 lines of python code.
+    }
 \end{frame}
 
 \begin{frame}[fragile]{ML Component}

ml_services_validation_19-05/slides/service.tex

@@ -47,10 +47,6 @@ class SentimentAnalysis(Schema):
     sentiment = fields.String(title="sentiment", description="Sentiment class (Positive or Negative)")
 
 
-def home():
-    return "Welcome to Sentiment Analysis API"
-
-
 def analyze(text: str, model: SentimentAnalysisModel) -> SentimentAnalysis:
     """
     tags:
@@ -66,21 +62,17 @@ def analyze(text: str, model: SentimentAnalysisModel) -> SentimentAnalysis:
     text = unquote(text)
     score, sentiment = model.predict(text)
 
-    return {
-        "text": text,
-        "score": score,
-        "sentiment": sentiment,
-    }
+    return {"text": text, "score": score, "sentiment": sentiment}
 
 
 app = Flama(
     components=[SentimentAnalysisModelComponent(model_path="model.h5", words_path="words.json")],
     title="Sentiment Analysis",  # API title
     version="0.1",  # API version
     description="A sentiment analysis API for movies reviews",  # API description
+    redoc="/redoc/",
 )
 
-app.add_route("/", home, methods=["GET"])
 app.add_route("/analyze/", analyze, methods=["GET"])
 
 

ml_services_validation_19-05/src/api/api.py

@@ -5,3 +5,4 @@ h5py
 tensorflow
 flama[full]
 uvicorn
+marshmallow==3.0.0rc5