Complex model weights with Tensorflow 2.0 and Keras

complex_weights_with_tensorflow_2.0.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install tf-nightly-2-0-preview==2.0.0.dev20190812\n",
    "!pip install tf-nightly-gpu-2-0-preview==2.0.0.dev20190812"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow.keras.layers import LSTM, Dense, Input\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "BATCH_SIZE = 5\n",
    "TIME_STEP_SIZE = 50\n",
    "FEATURE_SIZE = 1024\n",
    "BATCH_SHAPE = (BATCH_SIZE, TIME_STEP_SIZE, FEATURE_SIZE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = tf.keras.Sequential([\n",
    "    Input(batch_shape=(BATCH_SIZE, TIME_STEP_SIZE, FEATURE_SIZE),\n",
    "          dtype=tf.complex64),\n",
    "    LSTM(units=FEATURE_SIZE, dtype=tf.complex64, return_sequences=True),\n",
    "    Dense(units=FEATURE_SIZE, dtype=tf.complex64, activation='relu'),\n",
    "    LSTM(units=FEATURE_SIZE, dtype=tf.complex64, return_sequences=True),\n",
    "    Dense(units=FEATURE_SIZE, dtype=tf.complex64, activation='relu'),\n",
    "])\n",
    "model.compile(optimizer='adam', loss='MAPE')\n",
    "model.summary(line_length=75)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def batch_generator():\n",
    "    xBatchRow = tf.complex(\n",
    "        real=np.random.rand(BATCH_SIZE, TIME_STEP_SIZE, FEATURE_SIZE),\n",
    "        imag=np.random.rand(BATCH_SIZE, TIME_STEP_SIZE, FEATURE_SIZE)\n",
    "    )\n",
    "    yBatchRow = tf.complex(\n",
    "        real=np.random.rand(BATCH_SIZE, TIME_STEP_SIZE, FEATURE_SIZE),\n",
    "        imag=np.random.rand(BATCH_SIZE, TIME_STEP_SIZE, FEATURE_SIZE)\n",
    "    )\n",
    "    yield xBatchRow, yBatchRow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.fit_generator(generator=batch_generator())"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}