{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"
\n\n\n
\n\n
\n

\n
\"Google

Isolated Sign Language Recognition

HOW TO SUBMIT!

\n \n
\"asl
\n
CREATED BY: DARIEN SCHETTLER

\n\n
\n\n

\n\n\n\n
\n\n---\n\n
\n\n
\n 🛑   WARNING:

THIS IS A WORK IN PROGRESS
\n
\n\n\n
\n 👏   IF YOU FORK THIS OR FIND THIS HELPFUL   👏

PLEASE UPVOTE!

This was a lot of work for me and while it may seem silly, it makes me feel appreciated when others like my work. 😅\n
","metadata":{}},{"cell_type":"markdown","source":"
\n\nIn this notebook we will make a simple – untrained – tf.keras model that we will convert into tflite to show how to make a submission.\n\n
\n\nThe following is some details about Tensorflow Lite and the conversion process:\n\n
\n\n---\n\n
TensorFlow Lite\n\n
\n\nTextbook Definition (Key Points) [REF]\n\n\n\n\n\n
\n\nELI5 Competition Definition\n\nTensorFlow Lite is a specialized version of TensorFlow, a powerful machine learning framework that has been optimized for running on mobile and embedded devices. It allows us to run our models on a variety of devices, including smartphones and tablets, and provides fast, low-latency performance.\n\n
Explain With Pictures\n\nTF Model Conversion Workflow\n\n\n\n
\n\n---","metadata":{}},{"cell_type":"code","source":"###############################################\n######## Step 0 – Imports #######\n###############################################\nimport tensorflow as tf\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\nimport json\nimport os\n\n\n###############################################\n######## Step 1 – Setup #######\n###############################################\ndef read_json_file(file_path):\n with open(file_path, 'r') as file:\n return json.load(file)\n \nDATA_DIR = \"/kaggle/input/asl-signs\"\ntrain_df = pd.read_csv(os.path.join(DATA_DIR, \"train.csv\"))\ns2p_map = {k.lower():v for k,v in read_json_file(os.path.join(DATA_DIR, \"sign_to_prediction_index_map.json\")).items()}\np2s_map = {v:k for k,v in read_json_file(os.path.join(DATA_DIR, \"sign_to_prediction_index_map.json\")).items()}\nencoder = lambda x: s2p_map.get(x.lower())\ndecoder = lambda x: p2s_map.get(x)\n\ntrain_df[\"path\"] = DATA_DIR+\"/\"+train_df[\"path\"]\ndisplay(train_df)\n\nROWS_PER_FRAME = 543 # number of landmarks per frame\nINPUT_SHAPE = (543,3) # ...reshape(n_frames, ROWS_PER_FRAME, len(data_columns))\nN_LABELS = 250 # train_df.sign.nunique()\n\n\n###############################################\n### Step 2 – Host Provided Load Fn ###\n###############################################\n\ndef load_relevant_data_subset(pq_path):\n data_columns = ['x', 'y', 'z']\n data = pd.read_parquet(pq_path, columns=data_columns)\n n_frames = int(len(data) / ROWS_PER_FRAME)\n data = data.values.reshape(n_frames, ROWS_PER_FRAME, len(data_columns))\n return data.astype(np.float32)\n\n\n###############################################\n### Step 3 – Model Creation Fn ###\n###############################################\n\ndef get_tf_keras_model(optimizer=\"adam\", loss=\"sparse_categorical_crossentropy\"):\n \"\"\" Basic model creation fn with NaN handling \"\"\"\n \n # (batch_size, n_landmarks, n_coords)\n _inputs = tf.keras.layers.Input(shape=INPUT_SHAPE, name=\"inputs\") \n \n # Replace all nan with 0\n x = tf.keras.layers.Lambda(\n lambda _x: tf.where(tf.math.is_nan(_x), tf.zeros_like(_x), _x), name=\"nan_to_zero\",\n )(_inputs)\n \n # Flatten all data points and reduce 'batch_dim' (which is actually frame_dim)\n # to 1 via averaging --> yields (1, n_landmarks*n_coords)\n x = tf.keras.layers.Lambda(\n lambda _x: tf.reshape(tf.reduce_mean(_x, axis=0), \n (1, INPUT_SHAPE[0]*INPUT_SHAPE[1])), \n name=\"flatten_all\",\n )(x)\n \n # (1, n_labels)\n _outputs = tf.keras.layers.Dense(N_LABELS, activation=\"softmax\", name=\"outputs\")(x)\n\n # Make model with optimizer and loss (probably not necessary)\n _model = tf.keras.Model(inputs=_inputs, outputs=_outputs)\n _model.compile(optimizer, loss)\n return _model\n\n\n##################################################\n### Step 4 – Create & Save Model [BRANCHING] ###\n##################################################\n\n# BRANCH A --> tf.lite.TFLiteConverter.from_saved_model() \nmodel = get_tf_keras_model()\nprint(model.summary())\nmodel.save(\"./model\")\n\n# BRANCH B --> tf.lite.TFLiteConverter.from_keras_model() \n### NOTHING REQUIRED AT THIS STEP ###\n\n\n####################################################\n### Step 5 – Convert Model to TFLite [BRANCHING] ###\n####################################################\n\n# BRANCH A --> tf.lite.TFLiteConverter.from_saved_model() \nsaved_model_converter = tf.lite.TFLiteConverter.from_saved_model(\"./model\")\ntflite_model = saved_model_converter.convert()\n\n### BRANCH B --> tf.lite.TFLiteConverter.from_keras_model() \n# keras_converter = tf.lite.TFLiteConverter.from_keras_model(model)\n# tflite_model = keras_model_converter.convert()\n\n\n####################################################\n### Step 6 – Save Newly Created TFLite Model ###\n####################################################\n\n# Save the model.\nwith open('model.tflite', 'wb') as f:\n f.write(tflite_model)\n \n####################################################\n### Step 7 – ZIP Saved File For Submission ###\n####################################################\n!zip submission.zip ./model.tflite\n\n# Check our pwd\n!ls ./ ","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2023-02-27T17:26:14.153791Z","iopub.execute_input":"2023-02-27T17:26:14.154341Z","iopub.status.idle":"2023-02-27T17:26:20.832177Z","shell.execute_reply.started":"2023-02-27T17:26:14.154292Z","shell.execute_reply":"2023-02-27T17:26:20.830120Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
\n\nTest Our Model","metadata":{}},{"cell_type":"code","source":"!pip install tflite-runtime\nimport tflite_runtime.interpreter as tflite\n\ninterpreter = tflite.Interpreter(\"./model.tflite\")\nfound_signatures = list(interpreter.get_signature_list().keys())\n# if REQUIRED_SIGNATURE not in found_signatures:\n# raise KernelEvalException('Required input signature not found.')\nprediction_fn = interpreter.get_signature_runner(\"serving_default\")\n\noutput = prediction_fn(inputs=load_relevant_data_subset(train_df.path[0]))\nsign = np.argmax(output[\"outputs\"])\n\nprint(\"PRED : \", decoder(sign))\nprint(\"GT : \", train_df.sign[0])","metadata":{"execution":{"iopub.status.busy":"2023-02-27T17:26:53.857454Z","iopub.execute_input":"2023-02-27T17:26:53.859662Z","iopub.status.idle":"2023-02-27T17:26:53.877951Z","shell.execute_reply.started":"2023-02-27T17:26:53.859336Z","shell.execute_reply":"2023-02-27T17:26:53.876585Z"},"trusted":true},"execution_count":null,"outputs":[]}]}