rbm_mnist.ipynb

{
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "tags": [],
        "deepnote_to_be_reexecuted": false,
        "source_hash": "2b01372e",
        "execution_millis": 2722,
        "execution_start": 1619814113099,
        "cell_id": "00000-0edb4860-ebf0-44c4-b005-a1f8f250a006",
        "output_cleared": false,
        "deepnote_cell_type": "code"
      },
      "source": "import tensorflow as tf\nimport tensorflow_probability as tfp\nimport pandas as pd\nimport numpy as np\nimport keras\nimport matplotlib.pyplot as plt\nfrom matplotlib import pyplot\n\nimport json\n\nfrom rbm import RBM, train_rbm\nfrom rbm_utils import loss\nfrom dae import DAE\nfrom utils import dump_models, load_models, min_max_scale_data, display_output, plot_loss",
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "deepnote_variable_name": "training_data_size",
        "deepnote_variable_value": "60000",
        "cell_id": "00001-12f0b27c-a246-4b5b-a2d1-6182c70dae0a",
        "deepnote_to_be_reexecuted": false,
        "source_hash": "b623e53d",
        "execution_start": 1619814115823,
        "execution_millis": 5,
        "deepnote_cell_type": "input-text"
      },
      "source": "",
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "deepnote_variable_name": "epochs",
        "deepnote_variable_value": "2000",
        "cell_id": "00002-fe0149b1-2aa1-44b0-ae3e-525f7da57737",
        "deepnote_to_be_reexecuted": false,
        "source_hash": "b623e53d",
        "execution_start": 1619814115831,
        "execution_millis": 9,
        "deepnote_cell_type": "input-text"
      },
      "source": "",
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": "",
      "metadata": {
        "deepnote_variable_name": "batch_size",
        "deepnote_variable_value": "150",
        "cell_id": "00003-55f12e92-29d1-4c3d-a5ca-8664728fbcba",
        "deepnote_to_be_reexecuted": false,
        "source_hash": "b623e53d",
        "execution_start": 1619814115842,
        "execution_millis": 0,
        "deepnote_cell_type": "input-text"
      },
      "outputs": [],
      "execution_count": 4
    },
    {
      "cell_type": "code",
      "metadata": {
        "tags": [],
        "cell_id": "00001-83f7cb49-bfdf-4df3-bd5a-408b4018cf12",
        "deepnote_to_be_reexecuted": false,
        "source_hash": "b233a976",
        "execution_start": 1619814115844,
        "execution_millis": 642,
        "deepnote_cell_type": "code"
      },
      "source": "mnist = tf.keras.datasets.mnist.load_data(path=\"mnist.npz\") # returns (x_train, y_train), (x_test, y_test)\nx_train, y_train = mnist[0]\nx_tset, y_test = mnist[1]\n\nMNIST_NUM_PIXELS = 784 # 28x28\nx_train = x_train.reshape(-1,MNIST_NUM_PIXELS, 1)\nx_train_full = x_train\nx_train = x_train[:int(training_data_size), :]\nx_train = np.divide(x_train,255)\n\nx_train = tf.constant(x_train, dtype=tf.float32)\nx_train = tf.reshape(x_train, (-1,784))",
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "tags": [],
        "cell_id": "00010-cfe54d4e-9b10-4fe5-90a8-810fb34ff456",
        "deepnote_to_be_reexecuted": false,
        "source_hash": "124df82a",
        "execution_start": 1619814116493,
        "execution_millis": 3,
        "deepnote_cell_type": "code"
      },
      "source": "epochs = int(epochs)\nlr_b = 0.01\nlr_g = 0.001\nmodels_specs = [\n    {\n        \"visible_dim\": 784,\n        \"hidden_dim\": 1000, \n        \"num_epochs\": epochs, \n        \"learning_rate\": lr_b, \n        \"display_dim1\": 28, \n        \"display_dim2\": 28, \n        \"use_gaussian\": False\n    }, \n    {\n        \"visible_dim\": 1000,\n        \"hidden_dim\": 500, \n        \"num_epochs\": epochs, \n        \"learning_rate\": lr_b, \n        \"display_dim1\": 25, \n        \"display_dim2\": 40, \n        \"use_gaussian\": False\n    },\n    {\n        \"visible_dim\": 500,\n        \"hidden_dim\": 250, \n        \"num_epochs\": epochs, \n        \"learning_rate\": lr_b, \n        \"display_dim1\": 25, \n        \"display_dim2\": 20, \n        \"use_gaussian\": False\n    },\n    {\n        \"visible_dim\": 250,\n        \"hidden_dim\": 2, \n        \"num_epochs\": epochs * 3, \n        \"learning_rate\": lr_g, # use much lower LR for gaussian to avoid exploding gradient\n        \"display_dim1\": 25, \n        \"display_dim2\": 10, \n        \"use_gaussian\": True # use a Gaussian distribution for the last hidden layer to let it take advantage of continuous values\n    }\n]",
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": "# model1 = load_models('1_784_1000')[0]\n# model2 = load_models('2_1000_500')[0]\n# model3 = load_models('3_500_250')[0]\n# model4 = load_models('4_250_2')[0]\n# models=[model1,model2,model3,model4]\n# display_output(x_train, model1, 28,28)",
      "metadata": {
        "tags": [],
        "cell_id": "00008-8bc57fe9-8c29-4e28-91fe-194bd9106a78",
        "deepnote_to_be_reexecuted": false,
        "source_hash": "d6744954",
        "execution_start": 1619814116501,
        "execution_millis": 5,
        "deepnote_cell_type": "code"
      },
      "outputs": [],
      "execution_count": 7
    },
    {
      "cell_type": "code",
      "source": "# _, data1 = model1.sample_h(x_train)\n# _, data2 = model2.sample_h(data1)\n# _, data3 = model3.sample_h(data2)\ndatas=[x_train]#data1,data2,data3]",
      "metadata": {
        "tags": [],
        "cell_id": "00009-0332dac9-5942-49d6-aa18-bfd7901f1e6e",
        "deepnote_to_be_reexecuted": false,
        "source_hash": "82fb83d9",
        "execution_start": 1619814116509,
        "execution_millis": 2,
        "deepnote_cell_type": "code"
      },
      "outputs": [],
      "execution_count": 8
    },
    {
      "cell_type": "code",
      "metadata": {
        "tags": [],
        "cell_id": "00010-6a0441b2-34ec-49e1-816d-1a907c728ef4",
        "deepnote_to_be_reexecuted": false,
        "source_hash": "e2ff3284",
        "execution_start": 1619814116529,
        "execution_millis": 320,
        "output_cleared": false,
        "deepnote_cell_type": "code"
      },
      "source": "index = 0\ndata = datas[index]\nconfig = models_specs[index]\n\ndef train_layer(data, config):\n    # parse config\n    visible_dim = config[\"visible_dim\"]\n    hidden_dim = config[\"hidden_dim\"]\n    num_epochs = config[\"num_epochs\"]\n    lr = config[\"learning_rate\"]\n    d1 = config[\"display_dim1\"]\n    d2 = config[\"display_dim2\"]\n    use_gaussian = config[\"use_gaussian\"]\n    \n    # train RBM\n    print(f'Training RBM -> {visible_dim} to {hidden_dim}')\n    model, v, v_pred, losses = train_rbm(\n        data,\n        visible_dim, hidden_dim,\n        k=1, \n        num_epochs=num_epochs, batch_size=int(batch_size),\n        lr=lr,\n        use_gaussian=use_gaussian,\n        # pretrained_rbm=models[index]\n    )\n    \n    # display sample output and plot loss\n    display_output(data, model, d1, d2)\n    # dump_models(f'{index+1}_{visible_dim}_{hidden_dim}', [model])\n    return model\ntrain_layer(data, config)",
      "outputs": [
        {
          "name": "stdout",
          "text": "Training RBM -> 784 to 1000\n",
          "output_type": "stream"
        }
      ],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "tags": [],
        "cell_id": "00014-53ee2793-6d71-4490-835b-94b7cae9a337",
        "deepnote_to_be_reexecuted": true,
        "source_hash": "43a996b",
        "execution_start": 1619812397115,
        "execution_millis": 1623,
        "output_cleared": false,
        "deepnote_cell_type": "code"
      },
      "source": "ae = DAE(models)\ndisplay_output(x_train, ae, 28, 28)",
      "outputs": [
        {
          "data": {
            "text/plain": "<Figure size 1440x576 with 16 Axes>",
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light",
            "image/png": {
              "width": 1150,
              "height": 389
            }
          },
          "output_type": "display_data"
        }
      ],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "tags": [],
        "cell_id": "00012-e21c8075-2e9e-455a-ab0e-9656558befab",
        "deepnote_to_be_reexecuted": true,
        "source_hash": "e7a2ba0",
        "execution_start": 1619812398732,
        "execution_millis": 2,
        "deepnote_cell_type": "code"
      },
      "source": "def loss(prediction, label):\n    assert prediction.dtype == tf.float32, f'wrong dtype for prediction {prediction.dtype}'\n    assert label.dtype == tf.float32, f'wrong dtype for label {prediction.dtype}'\n\n    diff = tf.subtract(prediction, label)\n    sq = tf.square(diff)\n    return tf.cast(tf.reduce_mean(sq), tf.float32)\n",
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "tags": [],
        "cell_id": "00012-7324a161-49f4-43bf-80c6-3081e47d21b3",
        "deepnote_to_be_reexecuted": true,
        "source_hash": "49c07757",
        "execution_start": 1619812453885,
        "execution_millis": 10264,
        "output_cleared": false,
        "deepnote_cell_type": "code"
      },
      "source": "# fine-tune autoencoder\nopt = tf.optimizers.Adam(learning_rate=0.0001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, name='Adam')\nfor epoch in range(20):\n    losses = []\n    loss_ = lambda: loss(ae.predict(x_train), x_train)\n    opt.minimize(loss_, [*ae.encoders, *ae.decoders])\n    # for val in x_train:\n    #     # val = tf.constant(val.T, dtype=tf.float32)\n    #     val = tf.reshape(val, (1, -1))\n    #     loss_ = lambda: loss(ae.predict(val), val)\n    #     opt.minimize(loss_, [*ae.encoders, *ae.decoders])\n    #     losses.append(loss_())\n    if epoch%10 == 0: print(f'epoch {epoch}: {tf.reduce_sum(losses).numpy()}')\nprint(f'epoch {epoch}: {tf.reduce_sum(losses).numpy()}')\ndisplay_output(x_train, ae, 28, 28)",
      "outputs": [
        {
          "name": "stdout",
          "text": "epoch 0: 0.0\nepoch 10: 0.0\nepoch 19: 0.0\n",
          "output_type": "stream"
        },
        {
          "data": {
            "text/plain": "<Figure size 1440x576 with 16 Axes>",
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light",
            "image/png": {
              "width": 1150,
              "height": 389
            }
          },
          "output_type": "display_data"
        }
      ],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "source": "",
      "metadata": {
        "tags": [],
        "cell_id": "00014-b2790355-95b5-4310-9538-7c4905a769a9",
        "deepnote_to_be_reexecuted": true,
        "source_hash": "b623e53d",
        "deepnote_cell_type": "code"
      },
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "source": "<a style='text-decoration:none;line-height:16px;display:flex;color:#5B5B62;padding:10px;justify-content:end;' href='https://deepnote.com?utm_source=created-in-deepnote-cell&projectId=c8f7eba4-1ab3-4397-b00f-0691cd146402' target=\"_blank\">\n<img alt='Created in deepnote.com' style='display:inline;max-height:16px;margin:0px;margin-right:7.5px;' src='data:image/svg+xml;base64,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' > </img>\nCreated in <span style='font-weight:600;margin-left:4px;'>Deepnote</span></a>",
      "metadata": {
        "tags": [],
        "created_in_deepnote_cell": true,
        "deepnote_cell_type": "markdown"
      }
    }
  ],
  "nbformat": 4,
  "nbformat_minor": 2,
  "metadata": {
    "orig_nbformat": 2,
    "deepnote": {
      "is_reactive": false
    },
    "deepnote_notebook_id": "4f1c689b-f80e-4624-b352-e34be53b8bb2",
    "deepnote_execution_queue": [
      {
        "cellId": "00010-6a0441b2-34ec-49e1-816d-1a907c728ef4",
        "sessionId": "18ece3c8-6801-4ef8-af50-d8cbabb4321d",
        "msgId": "98e87cf1-11ed-4b69-9a62-3a8181fcb050"
      },
      {
        "cellId": "00014-53ee2793-6d71-4490-835b-94b7cae9a337",
        "sessionId": "18ece3c8-6801-4ef8-af50-d8cbabb4321d",
        "msgId": "0c45de90-5cc2-4bde-bf45-e30e0803a4c5"
      },
      {
        "cellId": "00012-e21c8075-2e9e-455a-ab0e-9656558befab",
        "sessionId": "18ece3c8-6801-4ef8-af50-d8cbabb4321d",
        "msgId": "7a4eb573-e6e7-4351-bbb8-cb3b14594d99"
      },
      {
        "cellId": "00012-7324a161-49f4-43bf-80c6-3081e47d21b3",
        "sessionId": "18ece3c8-6801-4ef8-af50-d8cbabb4321d",
        "msgId": "0c213a81-50aa-4fa0-af39-cddeee4469ed"
      },
      {
        "cellId": "00014-b2790355-95b5-4310-9538-7c4905a769a9",
        "sessionId": "18ece3c8-6801-4ef8-af50-d8cbabb4321d",
        "msgId": "8dc9f768-ce18-444f-9e4f-8d0218b82b8a"
      }
    ]
  }
}